Update: Intel's PresentMon tool just got launched this morning. You can download it if you want to mess around with these tools: game.intel.com/story/intel-presentmon/ We're super excited about putting this to use -- it could be really helpful in identifying GPU or CPU bottlenecks and driver overhead across all vendors. Since this is vendor agnostic and because it's just an update to open source PresentMon, it's pretty easy for us to roll-out into testing.
This is going to be very interesting to see for you, but also for us. Plenty of times my PC has frozen during a battle in DoW Soulstorm Apoc (Very Modded) and see what’s the cause or how far one can push before it borks again.
This stuff is a little too geeky for me, but if it helps you inform us on which tech is best for the buck I'm all for it. It's definitely not something I'd bother with on my own.
Nvidia: "It just works!" Tom: "We developed a tool to help identify and explain in detail any CPU or GPU performance bottlenecks in realtime..." Nvidia: "SHUT HIM UP!!!!"
As a business, Nvidia doesn't need him. They make big money selling workstation and commercial products, gaming is fine as long as it works. Intel needs a guy like him to be the "face" of the product and get peoples goodwill to help their line be more successful. Intel is now where Nvidia was in the early 2000s for the GPU market, and it actually works well for them because in a social media focused world interviews like this get a lot of traction for the product they want to sell.
In 2016, Intel stopped putting effort in their cpus and instead focused on expanding their operations with data centers and whatnot because AMD was far behind. For the past 2 years, they’ve actually been putting ressources into consummer goods again and it’s really cool
Ty competition. in their case, intel really got the worst wake up call. AMD suddenly ryzing in the cpu market in all front (b2c, b2b), nvidia further shifting toward data centers and ai with their gpu based products and becoming even more dominant over intel, and some traditional customers walking away from intel (like apple with their m1, m2, etc.) either they adapt, either they become the next IBM, Kodak, or Nokia. at least IBM manage to get back together, but definitelly lost big market segments. it is good and cool to see intel acting to avoid this path, would be sad to lose them in the mid-long term
This is not a change in Intel's mentality this is just underdog strategy vs top dog strategy. Similar to how AMD did FSR as open source to try to drive it's adoption because they did not have the market share to make something proprietary (like nVidia's DLSS). Similarly Intel is quite the underdog in the GPU space so they need this sort of strategy that is much more consumer friendly. Even in the CPU space they've lost quite a lot of share to AMD in the last years , we now have a really nice competitive space between the two of them so Intel can't use top dog strategies there anymore either (like they did before Ryzen).
@@shanent5793 You're probably not computer smart enough to understand how a pipeline works if that's how you feel. The GPU can only go as fast as a slow CPU can. That's basically what this whole video states. Stop crying. What are you even getting at? That intel is artificially making their cpu slow down a gpu?
I find it crazy just how much can be done in fractions of a second. Yeah I know computers have been doing it for a long time - but it's no less amazing to me that all this is processed in less time than a human can even comprehend. I'm still impressed by my RTX 3060 and DDR4 system. I'm a cheap date.
And then there's the physical size -- we have all these huge, massive heatsinks for transistors (albeit lots of them) that are infinitesimally small in comparison.
Yeah! It was crazy some of the tricks we had to use just to get games somewhere between 25-30fps on the PS1. Also, minimising the video rendering for the many languages (by drawing the text over each video frame before rendering) in the EU occasional messed your head up trying to work out suitable ways of laying out the UI with such little screen and buffer real estate. Which the translations for dutch being a tug of war between the UI designer and the translator. One trying to shorten the phrase to the minimum and the other trying to make it still make sense. It seems a lot easier nowadays with bigger buffers and fast gpu and cpu timings. But optimising was always supported by lots of cafienne providered to late nights for coders
As a Linux gamer I think that an explanation of what WINE does at this level would be amazing. Understanding the translations to DirectX and how (or if) it impacts performance would be perfect.
I liked Tom's refusal to oversell it as being guaranteed to change the world, maybe Intel's PR team aren't clued in to this one or they've just relaxed a bit
idk if PR or marketing would like him to stop talking. but it wouldn't surprise me. their so backwards, while Tom is the OG mfr we want to be open and honest while pushing everything forward.
I hope they apply the same innovation and passion back to their CPU development as well. Seems like they've fallen asleep on the CPU side of things lately.
@@Hickeroarthey fell asleep 5 years ago and woke up a couple years ago, but with the complexity of CPUs, it isn’t an instant switch, it takes a few years for their increased focus and funding to be reflected in the products offered.
@@Somebody374-bv8cd Yeah I was more talking about Tom as an individual and not intel as a company. Specifically, how good he is at explaining highly technical concepts in a way that is both easy to understand and not boring. And you can tell he actually knows his stuff too because of how well he is able to respond to Steve’s questions on the fly.
@@PDXpunk I was just joking about the game that was killed being overwatch 2 because it's the most negatively reviewed game in steam's history, so "not much left to kill"
I'm actually really excited for battlemage. Intel has such an opportunity to get into the consumer market if nvidia continues its focus on mostly enterprise AI stuff.
@@GamersNexus It says a lot when Steve and Austin actually considered the Arc for both their budget builds on their latest video. I think they've gained enough goodwill by actually taking your advice and keeping their mouth shut as both AMD and Nvidia kept stumbling whenever they opened theirs.
@@GamersNexus At this point, I don’t know what would be more unexpected - Intel becoming a near-equal third GPU contender, or AMD/NVIDIA actually returning to around $200 for a “workable” low end GPU. Nah who am I kidding they’ll release a $3000 or $4000 next tier up before that
Agreed, that comparison with Counter-Strike 2 was fascinating. It isn't just "set everything to lowest for the smoothest, most consistent experience". In that case *increasing* settings and lowering framerate made the GPU and CPU much more in sync with their workload. So cool
As an UE game developer I love this kind of content. It's educational not only for developers and testers, but it also starts to bridge the gap with end users.
Intel keeping themselves talking about these interesting subjects is lovely to see.. hope their next gen graphics cards can really mature from the already good ARC cards
I know right, its so good to see them not dumping Arc and really pushing through the problems and issues. I really hope Battlemage is good both in terms of price and performance.
I just worry, you control the metrics or the testing tool and then you control the narrative. It is open source so hopefully to F@%#ery goes on. But just think if everyone is using this to test and suddenly only Intel products start looking good...
It feels odd that they're among very few major hardware manufacturers who realized that it's good marketing and PR to embrace this particular, uhm, "space" /gestures at where we are/ of enthusiasts in a more purely technical way, rather than just marketing cookie cutter cliche bells and whistles at the gamers and influencers.
@@Fezzy976The ARC A750 currently is better than an RX6600, RX7600 and RTX4060 for value. The ray tracing and 1440p perfomance is very good on that card.
It is so refreshing for Intel to have Tom be their GPU guy. Speaking about ARC makes me consider it even though they are still kinda rough around the edges.
I just hope that Battlemage improves on their attempts with Alchemist. The software seems to be improving majorly so fingers crossed that Intel can provide good mid-tier performance for good prices.
intel are nailing the software with huge resources being poured in, hopefully the hardware catches up. 1st gen was always going to be rough so it can't indicate the state of 2nd gen. AMD needs to put more resources into software. Nvidia can call me when they have open source worth a damn. Competition between intel and AMD would be nice to see when it comes to GPU. Nvidia not being dick bags would also be nice to see but lets have attainable goals shall we.
@@tappy8741Fun fact, a friend has an RX580 8GB and it's worse running Nintendo Switch games through Vulkan than my PC using the Intel UHD 770 graphics (also Vulkan API)... I used Presentmon and the Ryzen 7 1700 isn't the problem for my friend's PC, it's the RX580, *on a Nintendo SWITCH emulator* . Radeon drivers are such a massive meme.
@@tappy8741You would already wash your hands when it comes to Nvidia, as for AMD; they need to put some more focus on their software more, albeit FSR 3 is good for being open source, I think they need more to competition for themselves. As for Intel, it has been improving ever since, Tom really is nailing it at the sweet spot, i have high hopes for the new Intel GPUs.
I love what Intel has done with Arc. I have had the A770 since January, and the driver improvements have been dramatic, to say the least -- coming from a fairly hardcore CS:GO player. Good to see Intel doubling down on gfx development and I can't wait to see the next generation of cards!
CSGO runs worse than Portal 2 on my i7-12700K. No driver update can fix the optimization mess from that game, specially with the huge fps drop with smoke grenades (such a meme). Also, CSGO refuses to use my e-cores, it must be Denuvo or weird anti-cheat software there.
I've never rooted for a corporation as much as I'm for Intel in the GPU market right now. I might give them a chance once I feel the need to upgrade my current graphics card 2-3 years from now. Given, of course, their drivers continue on the right path, and their prices keep competitive.
It won't matter in the end even if Intel does claw market share away from Ngreedia and AMDump they will just start doing the same and minimizing uplift and maximizing profits. We will be here 10 years from now (if were lucky) and be watching the same videos about how Intel just released a gpu with 10% uplift for 50% more price than their last generation and maybe more because Intel has a Quality that people attach to the brand so they might charge even more. It might be good in the short term when they are trying to win over customers but once enough buyers are there and brand loyal they will fall in line and do the same as everyone else.
@@iconoclast485 That's a good thing and that's why we need competition. When Intel dGPUs become mainstream, either AMD or Nvidia will become more affordable to fight Intel.
@@iconoclast485 Nvidia and AMD aren't really minimizing uplift, they're just prioritizing passing on efficiency gains to deep-pocketed data centre customers who will pay through the nose for performance-per-watt increases that don't require a power/cooling reprovision for every rack they upgrade. So those customers get the best binned silicon and gamers get the worst bins with frequencies pushed well beyond the point of best efficiency - those changes in binning procedures are why you can find comparisons like the RTX A40 vs the RTX 3090 - they're both based on the exact same GA102 die, but the pro card has more shaders, more VRAM, higher clocks all round, and a LOWER power draw than the 3090 - all for three times the price.
Something that Steve didn't talk about, but I think will be a point in the next video, is visualizing the differences between frame generation technologies with this benchmarking software.
I didn't even think about that, since frame gen to me is like a joke. I couldn't tolerate lagging behind by 2 frames for the sake of artificial smoothness. but this tool might be able to give better insight on how to bring the best out of all these technologies. fantastic
@@MartynDerg The behaviour and latency of the human visual system is what makes frame gen a worthwhile pursuit. It's why you can easily read moving text on a 60hz CRT, but need to push OLED past 144hz for the same clarity. Maybe someday Neuralink will integrate with Intel's SDK and add HUMANbusy to the charts to quantity the delay between "image observed" and "object recognized".
Special recognition to the team behind the camera for the incredible work and Steve for building GN into the gold standard of technical analysis and journalism.
Hardly the gold standard of journalism, he didn't even reach out to one party before dragging em through the mud. That's a basic courtesy any journalist knows you don't skip.
As a gamedev, this is terrible, this will actually force me to optimize using both the full GPU and CPU pipelines in balanced ways, player's will actually start criticizing my poorly optimized games and calling me out for lazy design.
That's pretty cool. There's definitely a good market case for being able to break down wait into something like "31% raw compile, 27% texture load, 12% optimize, 30% shader generation". And be able to spot like shader-related compiles going off the chart - so you just turn them down/off to get a significant performance boost.
What I want is a way to "freeze frame" the problem areas before they scroll off the screen. I.e., set a threshold, say "10 ms wait" and capture those instances where it's exceeded, with a plus-minus time wrapper, for deeper analysis later.
It would be interesting if we could see the system calls that occur during large wait periods. My guess is it's almost always cache misses and reading from RAM, maybe multiple times in one frame. (exacerbated by bad drivers). Interesting to see if we get a lot more cache like AMD has been doing with x3d chips. Even ram is pretty slow compared to cache/registers.
Intel better never lose this guy. He always has positive energy and knows how to present it so people understand. I have an arc A770 and it was a monstrous pain in my rear months ago but really I can't say anything negative about it now.
I'm optimistic that in a few years Intel will be extremely competitive. Right now they have basically proofs of concept, but not a product many people are buying. Their commitment to fixing their drivers and improving their GPUs is great to see, and something the industry needs to keep GPUs from becoming stale and uncompetitive. The more competitors we have in the space, the better products, faster innovation, and cheaper prices we will see. AMD and Nvidia are starting to become complacent, and have essentially mutually agreed to increase prices and not compete to get them lower. Intel trying to enter the market will undoubtedly try to undercut AMD and Nvidia to gain market share, and I hope this gets the duopoly off their high horse. Future of GPUs is looking promising.
Hopefully something in Intel's to do list is to not change the CPU socket every 2 years or so. This is more or less my block from switching to an Intel system. That and I want to rebuild my PC when I can also afford a far better monitor than what I currently own.
Glad I was not the only one thinking this. Very few games today seem to do anything fancy with their automatic settings optimization. Forza games seem to be an exception.
Auto optimise can't actually be solved as different people have different priorities. Some people are fine with lower FPS but higher quality textures. Some others are the opposite in the same game. I suppose you could have several auto 'targets'
First off, i can't state enough, how good you guys are doing with all your interviews, and all your videos, you are thorough and objective. Now, i wanted to say that this new diagnostic method might actually be a nice way to detect some kind of bottlenecks in a PC, and may ALSO prove useful for actual game development. Although i think Profilers in game engines may serve the same purpose,I'm not sure. Outside of game dev, for monitroing my system, this is a nice tool in my arsenal!
I remember when frametime and 0.1% lows were entering the scene. It was VERY enlightening. Seems like this is just as good a metric to be inserting in the conversation.
This type of video / presentation demonstrates why Gamers Nexus is a superior channel... nothing flashy, just technical info delivered and broken down in a professional manner... and we don't get interrupted every 5 minutes to buy a screw driver or a grossly overpriced back pack.
The breakdown of GPUBUSY that was talked about, can be VERY interesting. A good analogy to understanding what information this could contain is actually a feature in Minecraft: if you press SHIFT+F3 you get a navigable pie-chart breaking down performance time in many more segments. Having this for the GPU would be HUGE!
As someone that monitors server telemetry for a living, and is constantly looking at hardware and component "busy" times, I concur with the usage explain at 10:52. I use similar charts to make similar assessment about resource draws for specific use-cases to determine the limiting factor. I'm actually surprised this is only starting to be used now. I wonder what other industry telemetry analysis techniques will eventually make their way over to consumer product testing.
Part of the difficulty is the sense of scale. Monitoring a GPU looks at tasks that happened hundreds or thousands of times per second (or even 100,000s of times/second). That has traditionally made telemetry difficult because of network and storage requirements.
Never change Steve. You have officially become the Tech Jesus. We need people like you to keep the industry accurate, professional and inline with us Nerds. Keep up the good work and thank you for staying awesome
I love when you get Tom on to discuss this stuff. It’s incredibly informative and really shows intel is still committed to improving. Them doing optimizations for i5s because that’s what users have is promising
@@GamersNexus - Kindly consult some 3D developers - if you wish to use this metric then a standard GPU benchmark suddenly becomes a minefield of variability. An identical scenario with modified code may see an engine improve its present latency but be slower on the CPU (and overall) because of it. The GPU is doing work and things essential for the next frame are required as soon as possible, even if that means waiting for it.
Yeah I'm curious what the best way to turn this into a number would be. It's tough because I think both the mean (absolute) delta and the variance matter - if there's a lot of variance then work presumably needs to be spread out better, whereas if there's a large difference between the mean frame time and the mean GPU busy then it's just the total *amount* of work that needs to be lowered.
I like the gold and blue colors of the graphs lines Steve 👌🏻 😊 thank you so much for supporting ❤ I know you are apolitical so you won't talk about it but thank you 👍🏻
Not enough games tell you what settings are impacting your frames so I would personally love a feature like that. Alex from Digital Foundry did a similar piece talking about Game settings and being more transparent about what impacts what.
LOD and rendering distance, that's your main CPU bottleneck. Just play Minecraft and use the chunk distance setting, or watch your average potato open world game on the Switch with that absolute garbage Tegra X1 SOC. I'm glad that i have a beefy CPU like the i7-12700K, a friend has a Ryzen 9 5900X (beefy 3080Ti as well), can push the rendering distance very, very far away in Dying Light 2 with ray tracing, basically no LOD, even at 4K with that system and DLSS (also CPU intensive).
This looks like a great tool to hold developers to task for bad ports too. I'd be curious to see how terribly optimised games like Jedi Survivor and Hogwarts Legacy look like with this metric.
This is going to be huge, specially for games with DirectStorage coming into the scene... as we would be able to have a view(even if rudimentary) of when the GPU is in heavy GPU decompresion and when it is not.... Something that Frametime is not sufficient for that.
I've always hated how reviewing and benchmarking hardware will use the highest end parts with the idea being to remove any possible bottlenecks to show the best possible numbers for hardware. It's not just unrealistic but it assumes a linear performance curve that I've never seen in real life. I've never known enough about the process to express why it doesn't ring true but GPU Busy looks like part of the metric I've been looking for. I understand review numbers are just supposed to be relative to other hardware being compared in that review but beyond it seeming like most people can't comprehend the idea that they can't apply those numbers to real systems, the scale of performance of the hardware is skewed by what we end up just calling "driver overhead." Matching an entry level CPU with a flagship GPU or vice versa for a review doesn't tell you how that lower end part will work in the future or how performance will scale with future upgrades because there's so many other factors that determine performance that change over time. The mismatch can cause weird stuff like lower 1% lows as the lower end hardware can do weird stuff when it can't keep up. Go get original Crysis or Witcher 2 and crank up all the settings at 4k and watch your modern system struggle while not maxing out your CPU or GPU and compare those numbers to review benchmarks of the day to see how that scaling falls apart over time with twice the power you could get when they were contemporary. To be clear I don't know a good solution to benchmarking hardware since the alternative seems to be, running 10 times as many tests and historically the way it is done now seems good enough most of the time. I guess I wish reviewers would understand people will take their numbers at face value and do more tests showing you don't need an i7 or Ryzen 9 to fully utilize your hardware and that the upgrades you will actually do in the lifetime of the hardware isn't worth the "futureproofing" premium most of the time. GN and HUB say this all the time but I think more examples would hammer it home even better. So I guess all of this is to say I'd love more reviews showing contemporary balanced builds and revisiting popular hardware from years ago running modern games is some of my favorite GN stuff I'd like to see more of. P.S. Everyone should cap their framerates more for smoother more stable gameplay while using less power and taxing their hardware less.
Dear TechJesus, This is truly impressive! We'll no longer hear claims like "These are the BEST competitive settings!" because everyone can tailor the settings perfectly to their system. :D (By "competitive," I mean in terms of high FPS, not clearer visibility.) Once the update with its more in-depth analysis is released, I'd love to see a follow-up video demonstrating how to adjust settings to maximize hardware performance. It would be an excellent way to illustrate how different settings impact various components of a system. Thanks for the fantastic content! in love Tech Disciple
I could see this being used as a starting point for an automated settings modifier. It would be for those who just want to optimize game settings to their system with little to no thought process. Run the theorhetical program that runs Presentmon in the background to analyze where the bottlenecks are, then automatically change game settings based on that data after a little bit of sample gameplay. Maybe you go to the graphics settings menu and hit an "Optimize" button and it changes everything for you. Every game would likely need to be programmed into this proposed magical app individually, as settings menus are all different, but that's why open source is so cool. On the more commercial end, companies like Nvidia could use a version of it in their "Gforce Experience" bloatware to make their auto-optimization feature not suck so much.
This suggests having a bit more respect for Intel. I hope they continue this direction and keep including the public in their process development. I also hope they keep building graphics cards and adapting their drivers. The industry needed this desperately. I never thought I would say this but good work Intel. Good video. Thanks for the download link. Could this finally be the end of the iron curtain/coldwar mentality from hardware developers? Dogs of war, I hope so.
I just want to acknowledge how lucky we are that outlet like GN exists in this space, this is not something to be taken for granted. In fact, I don't think I can name another outlet even outside of tech space that is this dedicated to truth, pure content without any bullshit or propaganda, kudos to this team, keep doing what you're doing guys!
2 MILLION SUBSCRIBERS!!!... Over 100K in just a couple days is insane, no points for guessing why of course but well deserved either way with or without the last couple of uploads!
Its just LTT is a general channel and GN is a very specific technical channel. This video would go over the heads of the general public. @@GamingOverGirls
@@GamingOverGirls LTT being entertainment in the tech space vs GN being much more firmly educational in the tech space. People enjoy(ed) watching Linus being a dufus, such is the nature of things...
I like the simplistic intro. to the graphics pipeline, but I think a more detailed description could really help and doesn't necessarily need to be complicated. Not sure how well versed GN is in deferred vs. forward rendering but there can be some interesting analysis angles e.g. deferred rendering benefits greatly from GPU memory bandwidth which is also why mobile/weak hardware games tend to be forward rendered.
Unsolicited graphics programmer commentary (thanks for the great vid explaining what we do to laypeople GN!) - I would LOVE to see review outlets not report average FPS as the primary metric but "average frametime" (measured in ms). Frametime (time between presents) is a linear metric that's easy to compare/contrast. - The "wait" between frames can sometimes be intentional because we want the game simulation to not get too far ahead of what is displayed, which would result in increased perceived latency. Waitable swapchains are one mechanism used to impose this delay mid-frame, and ensure that the compositor is ready by the time we render. Of course, other reasons given for imposed waits by the driver or compositor are also relevant and factored into the imposed wait. Part of what makes this tricky is that the art of predicting how long the future frame will take to complete is not an exact science. - Be advised that the overlay provided by the "user friendly" PresentMon *will affect timings*. Compositor-interaction will impose compositor overhead and also potentially disable hardware scanout flips. If possible, I think I would disable the overlay or get it to display on a separate display to get accurate timings.
Completely agree. Special K's framepacing widget shows the current presentation model and it's drawn into the game's SwapChain so it doesn't bring the compositor into the equation. For NVIDIA users, SK also has a visualization mode involving Reflex's latency markers that can drill-down deep (way deeper than this video) into the distribution between GPU render time, flip queue / presentation delays, and CPU-side work submission in D3D11/12/GL. I have the same data as this "GPU Busy" thing in Special K, but I don't have it in graph form at the moment and it's also represented as the inverse -- GPU Idle time, since ultimately Special K is a frame pacing machine and the actual metric that's important is when the GPU goes idle between frames.
These videos with Tom have been great both because we learn cool information about how graphics work, and we get to see someone genuinely enthusiastic about the topic who is actively working to make it better and share that progress with us. If everyone who worked in these spaces was as enthusiastic as Tom and Steve, I think we'd all benefit. Great video!
Is this footage from Tom's visit about a year ago, or did he swing by again more recently? Either way, great to see him. He's really friendly (at least seems that way on camera), and explains things really well so that a dummy like me can understand it. :-)
@@MidnightToker24I run the legendary Unreal Tournament 2004 (64 bit version, uses DX9) on an i7-12700K, 100% usage for one thread in one core, between 190 and 400fps. Legendary AI from that game as well, i'm glad that i have an i7-12700K for vintage PC games.
This is game changing for users tuning for performance on mid tier hardware. Awesome stuff intel! The dedication to improve the drivers especially for the "average" gamer is fantastic to see. I'm really excited to see the future cpu wait and gpu render breakdowns.
Very interesting topic, is the GPU/CPU load "ideal" a 1:1 (referencing the graph)? is there a metric number that is ideal, could a calculation be made to determine where your bottleneck exists?
close to 1:1 seems to be the goal to shoot for, if you're too far in either direction in the majority of games you play it means either the gpu or cpu in your computer isn't being used to its full potential
As far as I understood it, the graph CAN not expose a GPU bottleneck, because in that case frametime includes waiting for the GPU, so it'll show as a 1:1 graph. Would love if they were able to find a game (or furmark) to show this, and how close it would possibly get. I'd imagine if you don't get any significant deviations from a 1:1 graph, and you see that in most games, that could indicate you could be GPU bottlenecked. So a few CPU deviations might be good, because that means you CPU is faster than GPU sometimes/most of the time, but GPU isn't so slow that it is never faster than the CPU when it DOES have to compute more for a few frames
@@vocassen - you misunderstand. the Present wait from driver calls to flip is a fraction of the frame-rate assuming uncapped frames. So you can have two games that both run at 120fps but have different present times. The true measure of engine efficiency would be some metric of how 'much' is displayed on screen for a given present delay. Ie if you had two game engines rendering exactly the same geometry with exactly the same textures and shaders, you'd have different framerates - but that difference might have different ratios between CPU and GPU times on two different GPUs / OSes etc etc So GPUWait tells you something about driver efficiency and game engine efficiency - but it's still not enough information as sometimes, greater efficiency comes from spending more GPU time to reduce CPU time. A game might have greater GPU wait time than another - but perform better overall.
Having equal CPU and GPU usage is only ideal in terms of hardware utilization but that doesn't necessarily carry over to an ideal end user experience. Most games run GPU bound for good reason.
There's no "ideal" load but I'll always want it to be GPU bound as much as possible since it's the last in the chain (well I guess the monitor is but that's another topic). Why? You can easily fix a GPU bottleneck by turning on Reflex. If that's not available, you can simply reduce your graphics settings and/or resolution. You can even disable queueing entirely by turning on Ultra Low Latency/Anti-Lag but that isn't recommended as it causes your PC to be both CPU and GPU bound.
Fascinating content Steve. I love it when you have guests or go somewhere to do an interview as the content is always of a very high calibre. This is quickly becoming my favourite tech channel.
Update: Intel's PresentMon tool just got launched this morning. You can download it if you want to mess around with these tools: game.intel.com/story/intel-presentmon/
We're super excited about putting this to use -- it could be really helpful in identifying GPU or CPU bottlenecks and driver overhead across all vendors. Since this is vendor agnostic and because it's just an update to open source PresentMon, it's pretty easy for us to roll-out into testing.
@Steve, thank you from the bottom of my heart for exposing LTT
@@fakhirkhan9689Same! I'd love a final news blurb from GN after a few weeks pass and the dust settles.
This is going to be very interesting to see for you, but also for us. Plenty of times my PC has frozen during a battle in DoW Soulstorm Apoc (Very Modded) and see what’s the cause or how far one can push before it borks again.
audio level between sponsor and rest of video are off
This stuff is a little too geeky for me, but if it helps you inform us on which tech is best for the buck I'm all for it. It's definitely not something I'd bother with on my own.
Oh no, a new measurement tool for me to stare at and question where I've gone wrong in my build.
This gave me a chuckle, +1 internet points /g
@Tree-House69 why thank you.
There are no mistakes only happy little accidents...
No mistakes are made. If you enjoy your build. Play games, be happy.
This will have less to do with your hardware and more with how well a game is optimized to utilize GPU.
I can't believe NVIDIA let this guy go.
He is such a positive and knowledgeable guy.
He is just great PR for whoever he works for.
you just know someone is going to run this on the nvidia cards and find another reason to roast them for the vram/bus.
Nvidia: "It just works!"
Tom: "We developed a tool to help identify and explain in detail any CPU or GPU performance bottlenecks in realtime..."
Nvidia: "SHUT HIM UP!!!!"
As a business, Nvidia doesn't need him. They make big money selling workstation and commercial products, gaming is fine as long as it works. Intel needs a guy like him to be the "face" of the product and get peoples goodwill to help their line be more successful. Intel is now where Nvidia was in the early 2000s for the GPU market, and it actually works well for them because in a social media focused world interviews like this get a lot of traction for the product they want to sell.
I mean... How do you know he didn't just leave Nvidia to go to intel
Well he kind of annoyed the ceo in one of the press conference
I really like how Intel develops lately. Just imagine how weird it would have been if Intel had done this 3 years ago.
More like from the start? First they rip everyone off, then they lose 90% of their stock value, and now they want to gain favor with consumers. LOL
In 2016, Intel stopped putting effort in their cpus and instead focused on expanding their operations with data centers and whatnot because AMD was far behind. For the past 2 years, they’ve actually been putting ressources into consummer goods again and it’s really cool
Ty competition. in their case, intel really got the worst wake up call. AMD suddenly ryzing in the cpu market in all front (b2c, b2b), nvidia further shifting toward data centers and ai with their gpu based products and becoming even more dominant over intel, and some traditional customers walking away from intel (like apple with their m1, m2, etc.)
either they adapt, either they become the next IBM, Kodak, or Nokia. at least IBM manage to get back together, but definitelly lost big market segments.
it is good and cool to see intel acting to avoid this path, would be sad to lose them in the mid-long term
This is not a change in Intel's mentality this is just underdog strategy vs top dog strategy. Similar to how AMD did FSR as open source to try to drive it's adoption because they did not have the market share to make something proprietary (like nVidia's DLSS). Similarly Intel is quite the underdog in the GPU space so they need this sort of strategy that is much more consumer friendly. Even in the CPU space they've lost quite a lot of share to AMD in the last years , we now have a really nice competitive space between the two of them so Intel can't use top dog strategies there anymore either (like they did before Ryzen).
Just imagine how weird it would have been if they actually waited for development to launch the product, we would still be waiting
I like how Steve just listens to the experts, and then asks questions that push the conversation forward. "Thanks, Steve."
tech jesus delivered once again! =D
+1 thumb to "Thanks Steve"
@@shanent5793CPU bound feels nastyyyy
@@shanent5793 You're probably not computer smart enough to understand how a pipeline works if that's how you feel.
The GPU can only go as fast as a slow CPU can. That's basically what this whole video states. Stop crying.
What are you even getting at? That intel is artificially making their cpu slow down a gpu?
”back to you, Steve”
I find it crazy just how much can be done in fractions of a second. Yeah I know computers have been doing it for a long time - but it's no less amazing to me that all this is processed in less time than a human can even comprehend. I'm still impressed by my RTX 3060 and DDR4 system. I'm a cheap date.
Don't under estimate what the brain does in mere seconds.
And then there's the physical size -- we have all these huge, massive heatsinks for transistors (albeit lots of them) that are infinitesimally small in comparison.
Even programmers comment on the current scale as shocking - you're not alone.
Yeah! It was crazy some of the tricks we had to use just to get games somewhere between 25-30fps on the PS1. Also, minimising the video rendering for the many languages (by drawing the text over each video frame before rendering) in the EU occasional messed your head up trying to work out suitable ways of laying out the UI with such little screen and buffer real estate. Which the translations for dutch being a tug of war between the UI designer and the translator. One trying to shorten the phrase to the minimum and the other trying to make it still make sense. It seems a lot easier nowadays with bigger buffers and fast gpu and cpu timings. But optimising was always supported by lots of cafienne providered to late nights for coders
@@TheFluffyArmadillo Dutch need to cut down on the double vowels like; ee = é and ii = ì.
As a Linux gamer I think that an explanation of what WINE does at this level would be amazing. Understanding the translations to DirectX and how (or if) it impacts performance would be perfect.
I liked Tom's refusal to oversell it as being guaranteed to change the world, maybe Intel's PR team aren't clued in to this one or they've just relaxed a bit
It's the trait of a good engineer to be reasonable about the expectations...
idk if PR or marketing would like him to stop talking. but it wouldn't surprise me. their so backwards, while Tom is the OG mfr we want to be open and honest while pushing everything forward.
Those slides are too fire to not have been touched by someone for marketing/PR. Core engineers are not that good at designing slides.
Realistically I don't see anything to sell to anyone that isn't a computer graphics need.
Under promise and over deliver. Generally when higher level executives follow this mantra, it's a sign of a good company.
Intel entering the GPU market has been pretty interesting and cool so far, let's hope they keep it up.
@@nathangoddard8115They've kept it up for more than 6 months, that should give you an idea :)
@@nathangoddard8115 A new challenger appears: Moore Threads
I hope they apply the same innovation and passion back to their CPU development as well. Seems like they've fallen asleep on the CPU side of things lately.
@@Hickeroarthey fell asleep 5 years ago and woke up a couple years ago, but with the complexity of CPUs, it isn’t an instant switch, it takes a few years for their increased focus and funding to be reflected in the products offered.
@@Hickeroarthe CPU side has been, and will continue, doing very cool things. Their process is just not as good as TSMC's, and is letting them down.
Intel entered the GPU market on the back foot, but they're definitely putting in some work. I'm hoping the Battlemage line really shakes things up.
I want battlemage for sure, but I need a computer with resizable bar / SAM
@@JackOfHearts42Anything from Zen 3 and foward is good enough.
Zen 3 is still very powerful, Alder Lake, Raptor Lake are niche stuff.
Tom Peterson absolutely killing the game as usual.
Given that it was overwatch 2 that was shown, there isn't a whole lot left to kill.
@@Somebody374-bv8cd Yeah I was more talking about Tom as an individual and not intel as a company. Specifically, how good he is at explaining highly technical concepts in a way that is both easy to understand and not boring. And you can tell he actually knows his stuff too because of how well he is able to respond to Steve’s questions on the fly.
@@PDXpunk I was just joking about the game that was killed being overwatch 2 because it's the most negatively reviewed game in steam's history, so "not much left to kill"
@@Somebody374-bv8cd At first, I didn't understand as well, but that was a good one
Yep, you know it's going to be a great video when Tom appears in the thumbnail
Tom is really putting in work for Intel's GPUs!
I'm actually really excited for battlemage. Intel has such an opportunity to get into the consumer market if nvidia continues its focus on mostly enterprise AI stuff.
There's a ton of room for Arc to fight it out in the entry-level and mid-range markets until NVIDIA and AMD decide to show up.
what i hope the most is for them to differentiate by not locking feature between product class. I want SR/IOV so freaking much...
@@GamersNexus It says a lot when Steve and Austin actually considered the Arc for both their budget builds on their latest video.
I think they've gained enough goodwill by actually taking your advice and keeping their mouth shut as both AMD and Nvidia kept stumbling whenever they opened theirs.
@@GamersNexus At this point, I don’t know what would be more unexpected - Intel becoming a near-equal third GPU contender, or AMD/NVIDIA actually returning to around $200 for a “workable” low end GPU. Nah who am I kidding they’ll release a $3000 or $4000 next tier up before that
intel facing so many problems but drivers is the biggest one
One small step for Intel, one GIANT LEAP for gamers.
one GIANT LEAP for gamers...wallet
The giant leap for gamers is start buying into Intel's GPU's. If gamers actually want competition.
I'm using Presentmon for my i7-12700K running console emulators.
Intel trying to get that GPU Bussy
Focusing on tighter more consistent frametimes as a goal is the best news tbh
Raising those 1% lows will make a more consistent and smooth experience!
I agree 👍
@@FlippyDipDipper Would be great, especially when running low FPS games on a device like a Steam Deck
Agreed, that comparison with Counter-Strike 2 was fascinating. It isn't just "set everything to lowest for the smoothest, most consistent experience". In that case *increasing* settings and lowering framerate made the GPU and CPU much more in sync with their workload. So cool
Exactly broski, the crusade against microstuttering
As an UE game developer I love this kind of content. It's educational not only for developers and testers, but it also starts to bridge the gap with end users.
Intel keeping themselves talking about these interesting subjects is lovely to see.. hope their next gen graphics cards can really mature from the already good ARC cards
I know right, its so good to see them not dumping Arc and really pushing through the problems and issues. I really hope Battlemage is good both in terms of price and performance.
I just worry, you control the metrics or the testing tool and then you control the narrative.
It is open source so hopefully to F@%#ery goes on. But just think if everyone is using this to test and suddenly only Intel products start looking good...
It looks promising. I never expected i might be going team blue for my next GPU.
It feels odd that they're among very few major hardware manufacturers who realized that it's good marketing and PR to embrace this particular, uhm, "space" /gestures at where we are/ of enthusiasts in a more purely technical way, rather than just marketing cookie cutter cliche bells and whistles at the gamers and influencers.
@@Fezzy976The ARC A750 currently is better than an RX6600, RX7600 and RTX4060 for value.
The ray tracing and 1440p perfomance is very good on that card.
It is so refreshing for Intel to have Tom be their GPU guy. Speaking about ARC makes me consider it even though they are still kinda rough around the edges.
I just hope that Battlemage improves on their attempts with Alchemist. The software seems to be improving majorly so fingers crossed that Intel can provide good mid-tier performance for good prices.
intel are nailing the software with huge resources being poured in, hopefully the hardware catches up. 1st gen was always going to be rough so it can't indicate the state of 2nd gen.
AMD needs to put more resources into software.
Nvidia can call me when they have open source worth a damn.
Competition between intel and AMD would be nice to see when it comes to GPU. Nvidia not being dick bags would also be nice to see but lets have attainable goals shall we.
@@tappy8741Fun fact, a friend has an RX580 8GB and it's worse running Nintendo Switch games through Vulkan than my PC using the Intel UHD 770 graphics (also Vulkan API)...
I used Presentmon and the Ryzen 7 1700 isn't the problem for my friend's PC, it's the RX580, *on a Nintendo SWITCH emulator* .
Radeon drivers are such a massive meme.
@@tappy8741You would already wash your hands when it comes to Nvidia, as for AMD; they need to put some more focus on their software more, albeit FSR 3 is good for being open source, I think they need more to competition for themselves. As for Intel, it has been improving ever since, Tom really is nailing it at the sweet spot, i have high hopes for the new Intel GPUs.
holding off till battlemage is released, I love Intel's approach to gpu market right now, prioritizing midrange market like what AMD did before
Intel simply can't sell a high-end product.
What midrange card has Intel released? Maybe we have two different definitions of the same word.
@@TripmasterG A750 fits the midrange category for me, it feels like rx580 all over again
@@Sam-KFor now...
@@TripmasterGarc 770/750 are mid range(1080/1440p gaming)
I love what Intel has done with Arc. I have had the A770 since January, and the driver improvements have been dramatic, to say the least -- coming from a fairly hardcore CS:GO player. Good to see Intel doubling down on gfx development and I can't wait to see the next generation of cards!
probably the way i'll go just because. What are they like emulation stuff?
CS:GO is all it can handle?
@@ChasseurTueur no, check some recent reviews. CS:GO support was pretty bad in the first batch of drivers, so I'm pointing out a specific improvement.
What rank?
CSGO runs worse than Portal 2 on my i7-12700K.
No driver update can fix the optimization mess from that game, specially with the huge fps drop with smoke grenades (such a meme).
Also, CSGO refuses to use my e-cores, it must be Denuvo or weird anti-cheat software there.
this sounds so useful, interested to see how it affects the benchmarks using it to come
Very interested to see! The engineers aren't even sure how it'll fully get used yet!
@@GamersNexus in any case thanks for making a video on this, I'm sure tools like this are going to help lots of consumers, and also great video!
Intel has been on the ball lately with this, genuinely surprising and refreshing.
They can't rely on the old tactics anymore. Great for the industry.
They have the easiest job, all the have to do is keep their mouths shut while Nvidia and AMD keep stumbling whenever they fail to keep theirs closed.
I've never rooted for a corporation as much as I'm for Intel in the GPU market right now. I might give them a chance once I feel the need to upgrade my current graphics card 2-3 years from now. Given, of course, their drivers continue on the right path, and their prices keep competitive.
It does feel different with an engineer at the head of the company over the years.
It won't matter in the end even if Intel does claw market share away from Ngreedia and AMDump they will just start doing the same and minimizing uplift and maximizing profits. We will be here 10 years from now (if were lucky) and be watching the same videos about how Intel just released a gpu with 10% uplift for 50% more price than their last generation and maybe more because Intel has a Quality that people attach to the brand so they might charge even more. It might be good in the short term when they are trying to win over customers but once enough buyers are there and brand loyal they will fall in line and do the same as everyone else.
@@iconoclast485 That's a good thing and that's why we need competition. When Intel dGPUs become mainstream, either AMD or Nvidia will become more affordable to fight Intel.
@@iconoclast485you really don’t understand economics.
@@iconoclast485 Nvidia and AMD aren't really minimizing uplift, they're just prioritizing passing on efficiency gains to deep-pocketed data centre customers who will pay through the nose for performance-per-watt increases that don't require a power/cooling reprovision for every rack they upgrade. So those customers get the best binned silicon and gamers get the worst bins with frequencies pushed well beyond the point of best efficiency - those changes in binning procedures are why you can find comparisons like the RTX A40 vs the RTX 3090 - they're both based on the exact same GA102 die, but the pro card has more shaders, more VRAM, higher clocks all round, and a LOWER power draw than the 3090 - all for three times the price.
Something that Steve didn't talk about, but I think will be a point in the next video, is visualizing the differences between frame generation technologies with this benchmarking software.
Yeah I wanna see where frame gen fits into all this
I didn't even think about that, since frame gen to me is like a joke. I couldn't tolerate lagging behind by 2 frames for the sake of artificial smoothness. but this tool might be able to give better insight on how to bring the best out of all these technologies. fantastic
@@MartynDerg The behaviour and latency of the human visual system is what makes frame gen a worthwhile pursuit. It's why you can easily read moving text on a 60hz CRT, but need to push OLED past 144hz for the same clarity. Maybe someday Neuralink will integrate with Intel's SDK and add HUMANbusy to the charts to quantity the delay between "image observed" and "object recognized".
Special recognition to the team behind the camera for the incredible work and Steve for building GN into the gold standard of technical analysis and journalism.
> gold standard of journalism
yeah no
Hardly the gold standard of journalism, he didn't even reach out to one party before dragging em through the mud. That's a basic courtesy any journalist knows you don't skip.
@@Kill4Gloryif he hadn't done that, they would have been able to hide their lie about contacting the company they stole from.
@@JoshsGameClubno need to reply this is basically a linus d sucker, new account and just bashing GN.
shut up linus@@Kill4Glory
As a gamedev, this is terrible, this will actually force me to optimize using both the full GPU and CPU pipelines in balanced ways, player's will actually start criticizing my poorly optimized games and calling me out for lazy design.
That's pretty cool. There's definitely a good market case for being able to break down wait into something like "31% raw compile, 27% texture load, 12% optimize, 30% shader generation". And be able to spot like shader-related compiles going off the chart - so you just turn them down/off to get a significant performance boost.
Leave it to Tom to come up with some crazy new monitoring tool. Dude keeps innovating no matter where he is. Awesome to see.
What I want is a way to "freeze frame" the problem areas before they scroll off the screen. I.e., set a threshold, say "10 ms wait" and capture those instances where it's exceeded, with a plus-minus time wrapper, for deeper analysis later.
It would be interesting if we could see the system calls that occur during large wait periods. My guess is it's almost always cache misses and reading from RAM, maybe multiple times in one frame. (exacerbated by bad drivers). Interesting to see if we get a lot more cache like AMD has been doing with x3d chips. Even ram is pretty slow compared to cache/registers.
Intel better never lose this guy. He always has positive energy and knows how to present it so people understand. I have an arc A770 and it was a monstrous pain in my rear months ago but really I can't say anything negative about it now.
I'm optimistic that in a few years Intel will be extremely competitive. Right now they have basically proofs of concept, but not a product many people are buying. Their commitment to fixing their drivers and improving their GPUs is great to see, and something the industry needs to keep GPUs from becoming stale and uncompetitive. The more competitors we have in the space, the better products, faster innovation, and cheaper prices we will see. AMD and Nvidia are starting to become complacent, and have essentially mutually agreed to increase prices and not compete to get them lower. Intel trying to enter the market will undoubtedly try to undercut AMD and Nvidia to gain market share, and I hope this gets the duopoly off their high horse. Future of GPUs is looking promising.
Hopefully something in Intel's to do list is to not change the CPU socket every 2 years or so. This is more or less my block from switching to an Intel system.
That and I want to rebuild my PC when I can also afford a far better monitor than what I currently own.
I bet within the next five years Intel is going to be toe to toe with Nvidia.
Steve, you get the best people on to help discuss technical topics with hardware and gaming. Bravo!
Gratz on 2mil subs.
I love when GN and TAP connect, always makes for an informative video.
This would be awesome if integrated into in-game benchmarks and then fed back into auto optimize graphic settings.
Glad I was not the only one thinking this. Very few games today seem to do anything fancy with their automatic settings optimization. Forza games seem to be an exception.
@@0xF00F00 it's mostly related to adjusting render resolution.
Some games kind of have that, it's usually called "Optimized Settings" which is obtained out of short background benchmarks.
Auto optimise can't actually be solved as different people have different priorities. Some people are fine with lower FPS but higher quality textures. Some others are the opposite in the same game. I suppose you could have several auto 'targets'
Mark my word. It will never happen.
First off, i can't state enough, how good you guys are doing with all your interviews, and all your videos, you are thorough and objective.
Now, i wanted to say that this new diagnostic method might actually be a nice way to detect some kind of bottlenecks in a PC, and may ALSO prove useful for actual game development.
Although i think Profilers in game engines may serve the same purpose,I'm not sure.
Outside of game dev, for monitroing my system, this is a nice tool in my arsenal!
I remember when frametime and 0.1% lows were entering the scene. It was VERY enlightening.
Seems like this is just as good a metric to be inserting in the conversation.
Mate this is way more than edutainment, this is literally a documentary / educational content. Great stuff!
This type of video / presentation demonstrates why Gamers Nexus is a superior channel... nothing flashy, just technical info delivered and broken down in a professional manner... and we don't get interrupted every 5 minutes to buy a screw driver or a grossly overpriced back pack.
The breakdown of GPUBUSY that was talked about, can be VERY interesting. A good analogy to understanding what information this could contain is actually a feature in Minecraft: if you press SHIFT+F3 you get a navigable pie-chart breaking down performance time in many more segments. Having this for the GPU would be HUGE!
21:50 Nice! I hope Frameview adds support.
As someone that monitors server telemetry for a living, and is constantly looking at hardware and component "busy" times, I concur with the usage explain at 10:52. I use similar charts to make similar assessment about resource draws for specific use-cases to determine the limiting factor. I'm actually surprised this is only starting to be used now. I wonder what other industry telemetry analysis techniques will eventually make their way over to consumer product testing.
Part of the difficulty is the sense of scale. Monitoring a GPU looks at tasks that happened hundreds or thousands of times per second (or even 100,000s of times/second). That has traditionally made telemetry difficult because of network and storage requirements.
Never change Steve. You have officially become the Tech Jesus. We need people like you to keep the industry accurate, professional and inline with us Nerds. Keep up the good work and thank you for staying awesome
I love when you get Tom on to discuss this stuff. It’s incredibly informative and really shows intel is still committed to improving. Them doing optimizations for i5s because that’s what users have is promising
This channel is about to explode in popularity. Great video. Keep it up!
Curious about using the delta (frame time - GPU busy) during CPU testing
Agreed! That's one of the metrics we want to mess around with!
fr
@@GamersNexus - Kindly consult some 3D developers - if you wish to use this metric then a standard GPU benchmark suddenly becomes a minefield of variability. An identical scenario with modified code may see an engine improve its present latency but be slower on the CPU (and overall) because of it. The GPU is doing work and things essential for the next frame are required as soon as possible, even if that means waiting for it.
Yeah I'm curious what the best way to turn this into a number would be. It's tough because I think both the mean (absolute) delta and the variance matter - if there's a lot of variance then work presumably needs to be spread out better, whereas if there's a large difference between the mean frame time and the mean GPU busy then it's just the total *amount* of work that needs to be lowered.
@@EmanuelHoogeveen they’d have to use measures of spread; mean +/- SD or with centile / box and whiskers. Which wouldn’t tell the complete story
I like the gold and blue colors of the graphs lines Steve 👌🏻 😊 thank you so much for supporting ❤ I know you are apolitical so you won't talk about it but thank you 👍🏻
Not enough games tell you what settings are impacting your frames so I would personally love a feature like that. Alex from Digital Foundry did a similar piece talking about Game settings and being more transparent about what impacts what.
LOD and rendering distance, that's your main CPU bottleneck. Just play Minecraft and use the chunk distance setting, or watch your average potato open world game on the Switch with that absolute garbage Tegra X1 SOC.
I'm glad that i have a beefy CPU like the i7-12700K, a friend has a Ryzen 9 5900X (beefy 3080Ti as well), can push the rendering distance very, very far away in Dying Light 2 with ray tracing, basically no LOD, even at 4K with that system and DLSS (also CPU intensive).
Wow! I can't believe you got Tom Petersen on the show!
What a fantastic, and literally, "distinguished" guest.
This looks like a great tool to hold developers to task for bad ports too. I'd be curious to see how terribly optimised games like Jedi Survivor and Hogwarts Legacy look like with this metric.
Is the "wait" the reason why when you have usage % up you never see 100% either on the cpu or the gpu?
(Usually 96-98% is considered to be maxed out)
96% you're still mostly GPU bound, i think 92%-94% is where you confidently get CPU bound.
@@MaxIronsThird I don't think you understood what I'm talking about in that comment.
@@FelixVyraok.
No, the wait had nothing to do with never seeing 100% usage.
@@MaxIronsThird Ok. So btw why is it happening?
This is going to be huge, specially for games with DirectStorage coming into the scene... as we would be able to have a view(even if rudimentary) of when the GPU is in heavy GPU decompresion and when it is not.... Something that Frametime is not sufficient for that.
I've always hated how reviewing and benchmarking hardware will use the highest end parts with the idea being to remove any possible bottlenecks to show the best possible numbers for hardware. It's not just unrealistic but it assumes a linear performance curve that I've never seen in real life. I've never known enough about the process to express why it doesn't ring true but GPU Busy looks like part of the metric I've been looking for. I understand review numbers are just supposed to be relative to other hardware being compared in that review but beyond it seeming like most people can't comprehend the idea that they can't apply those numbers to real systems, the scale of performance of the hardware is skewed by what we end up just calling "driver overhead."
Matching an entry level CPU with a flagship GPU or vice versa for a review doesn't tell you how that lower end part will work in the future or how performance will scale with future upgrades because there's so many other factors that determine performance that change over time. The mismatch can cause weird stuff like lower 1% lows as the lower end hardware can do weird stuff when it can't keep up. Go get original Crysis or Witcher 2 and crank up all the settings at 4k and watch your modern system struggle while not maxing out your CPU or GPU and compare those numbers to review benchmarks of the day to see how that scaling falls apart over time with twice the power you could get when they were contemporary.
To be clear I don't know a good solution to benchmarking hardware since the alternative seems to be, running 10 times as many tests and historically the way it is done now seems good enough most of the time. I guess I wish reviewers would understand people will take their numbers at face value and do more tests showing you don't need an i7 or Ryzen 9 to fully utilize your hardware and that the upgrades you will actually do in the lifetime of the hardware isn't worth the "futureproofing" premium most of the time. GN and HUB say this all the time but I think more examples would hammer it home even better.
So I guess all of this is to say I'd love more reviews showing contemporary balanced builds and revisiting popular hardware from years ago running modern games is some of my favorite GN stuff I'd like to see more of.
P.S. Everyone should cap their framerates more for smoother more stable gameplay while using less power and taxing their hardware less.
Hopefully by the time I will be done with my existing 3070, Intel will have something good to offer me
Good job on coming back to the content you're good at!
Great video! Keep up the great work GN!
The most educational tech channel on the Tube.
Love how after Steve has said "no more LTT videos" he actually follows through with his word and it's back to real business.
This is the best PC TH-cam Channel!
Some scaled ratio of frame rate to gpubusy would be a nice metric based on what Tom said the ideal case would be
These types of videos are truly great 👍
props to this dude. we need more of him! gj
Dear TechJesus,
This is truly impressive!
We'll no longer hear claims like "These are the BEST competitive settings!" because everyone can tailor the settings perfectly to their system. :D
(By "competitive," I mean in terms of high FPS, not clearer visibility.)
Once the update with its more in-depth analysis is released, I'd love to see a follow-up video demonstrating how to adjust settings to maximize hardware performance. It would be an excellent way to illustrate how different settings impact various components of a system.
Thanks for the fantastic content!
in love
Tech Disciple
I could see this being used as a starting point for an automated settings modifier.
It would be for those who just want to optimize game settings to their system with little to no thought process. Run the theorhetical program that runs Presentmon in the background to analyze where the bottlenecks are, then automatically change game settings based on that data after a little bit of sample gameplay. Maybe you go to the graphics settings menu and hit an "Optimize" button and it changes everything for you.
Every game would likely need to be programmed into this proposed magical app individually, as settings menus are all different, but that's why open source is so cool.
On the more commercial end, companies like Nvidia could use a version of it in their "Gforce Experience" bloatware to make their auto-optimization feature not suck so much.
This newest upload is an exceptionally insightful deep dive. Thank you GN! 🌺
This suggests having a bit more respect for Intel. I hope they continue this direction and keep including the public in their process development. I also hope they keep building graphics cards and adapting their drivers. The industry needed this desperately. I never thought I would say this but good work Intel. Good video. Thanks for the download link. Could this finally be the end of the iron curtain/coldwar mentality from hardware developers? Dogs of war, I hope so.
Really really love these kind of videos with Tom Petersen keep it up!!!
I love the "AMD RYZEN" on the shelf behind the Intel engineer who used to work for NVIDIA.
Congrats on 2 million subs, have been here since under 500k
I just want to acknowledge how lucky we are that outlet like GN exists in this space, this is not something to be taken for granted. In fact, I don't think I can name another outlet even outside of tech space that is this dedicated to truth, pure content without any bullshit or propaganda, kudos to this team, keep doing what you're doing guys!
I love watching Tom and Steve talking. 90% of this went over my head, but I still found it fascinating.
2 MILLION SUBSCRIBERS!!!... Over 100K in just a couple days is insane, no points for guessing why of course but well deserved either way with or without the last couple of uploads!
Everyone migrating from LTT to GN about to find out what proper tech due diligence looks like
it's baffling to me LTT has over 15M & GN only just hit 2M like what???
Its just LTT is a general channel and GN is a very specific technical channel. This video would go over the heads of the general public. @@GamingOverGirls
@@GamingOverGirls LTT being entertainment in the tech space vs GN being much more firmly educational in the tech space. People enjoy(ed) watching Linus being a dufus, such is the nature of things...
The difference between playing to win at all costs and playing to keep playing the game indefinitely.
Tom Petersen is a really good teacher.
Not gonna lie, I'm using this to help put me to sleep tonight but I'll re-watch it later I promise.
😆
Tom Petersen is AWESOME!
I'm SO glad that Intel lets him talk about this stuff.
I like the simplistic intro. to the graphics pipeline, but I think a more detailed description could really help and doesn't necessarily need to be complicated. Not sure how well versed GN is in deferred vs. forward rendering but there can be some interesting analysis angles e.g. deferred rendering benefits greatly from GPU memory bandwidth which is also why mobile/weak hardware games tend to be forward rendered.
As a subscriber since under 100k users it is so awesome to see Gamers Nexus break 2M. Great job Steve and team!
Great video! 8:15 It would be really cool to show off something like RenderDoc here. How meshes are getting drawn into buffers etc.
Great point! It's been a while since I used tools like that, but they're awesome to use. Thanks for the reminder.
Always a pleasure to see Tom in your videos! Tom definitely has that captivating teacher energy. 😄
Unsolicited graphics programmer commentary (thanks for the great vid explaining what we do to laypeople GN!)
- I would LOVE to see review outlets not report average FPS as the primary metric but "average frametime" (measured in ms). Frametime (time between presents) is a linear metric that's easy to compare/contrast.
- The "wait" between frames can sometimes be intentional because we want the game simulation to not get too far ahead of what is displayed, which would result in increased perceived latency. Waitable swapchains are one mechanism used to impose this delay mid-frame, and ensure that the compositor is ready by the time we render. Of course, other reasons given for imposed waits by the driver or compositor are also relevant and factored into the imposed wait. Part of what makes this tricky is that the art of predicting how long the future frame will take to complete is not an exact science.
- Be advised that the overlay provided by the "user friendly" PresentMon *will affect timings*. Compositor-interaction will impose compositor overhead and also potentially disable hardware scanout flips. If possible, I think I would disable the overlay or get it to display on a separate display to get accurate timings.
Completely agree. Special K's framepacing widget shows the current presentation model and it's drawn into the game's SwapChain so it doesn't bring the compositor into the equation.
For NVIDIA users, SK also has a visualization mode involving Reflex's latency markers that can drill-down deep (way deeper than this video) into the distribution between GPU render time, flip queue / presentation delays, and CPU-side work submission in D3D11/12/GL. I have the same data as this "GPU Busy" thing in Special K, but I don't have it in graph form at the moment and it's also represented as the inverse -- GPU Idle time, since ultimately Special K is a frame pacing machine and the actual metric that's important is when the GPU goes idle between frames.
These videos with Tom have been great both because we learn cool information about how graphics work, and we get to see someone genuinely enthusiastic about the topic who is actively working to make it better and share that progress with us. If everyone who worked in these spaces was as enthusiastic as Tom and Steve, I think we'd all benefit. Great video!
20 seconds in and you're already miles ahead of Linus Tech Flops
Is this footage from Tom's visit about a year ago, or did he swing by again more recently?
Either way, great to see him. He's really friendly (at least seems that way on camera), and explains things really well so that a dummy like me can understand it. :-)
CPU usage is so hard to measure these days. I love this.
Mine is pretty easy. Near 100% at all times during a game. It's an AMD 6800k.
@@MidnightToker24I run the legendary Unreal Tournament 2004 (64 bit version, uses DX9) on an i7-12700K, 100% usage for one thread in one core, between 190 and 400fps.
Legendary AI from that game as well, i'm glad that i have an i7-12700K for vintage PC games.
If this guys is on any other channel then i would have never clicked.. its GN who is presenting that means there is really something valuable..
That tool would be super useful in Linux, as some drivers there lije nVidia and Intel are sub par in optimizarion.
This is game changing for users tuning for performance on mid tier hardware. Awesome stuff intel! The dedication to improve the drivers especially for the "average" gamer is fantastic to see. I'm really excited to see the future cpu wait and gpu render breakdowns.
Tom's damn good at explaining this stuff to the layman.
I really love the CPU/GPU usage statistics for game tuning. You can squeeze out a ton of image quality or FPS by knowing which knobs to turn.
Im a simple guy. I see GN upload, i press like and then watch
I do that on every YT vid I have ever watched, that way I know if I've seen something but forgotten lol
Super Educational. Thank you Steve !
Funny thing is, SpecialK have had metrics like this for years in their tool... which is wild.
Great video! Tom Petersen always patient and pashionated explaning things, great stuff
Intel generating some seriously positive user sentiment for their GPU department.
I'm a software engineer so I love this kind of technical content!
This guy singlehandedly canceled Linus on a WHIM. What a gangster
So interesting! Thanks Tom and Steve for another great video!
Thanks Steve.
This looks super interesting. Great interview. Thanks Steve and Tom!
Very interesting topic, is the GPU/CPU load "ideal" a 1:1 (referencing the graph)? is there a metric number that is ideal, could a calculation be made to determine where your bottleneck exists?
close to 1:1 seems to be the goal to shoot for, if you're too far in either direction in the majority of games you play it means either the gpu or cpu in your computer isn't being used to its full potential
As far as I understood it, the graph CAN not expose a GPU bottleneck, because in that case frametime includes waiting for the GPU, so it'll show as a 1:1 graph. Would love if they were able to find a game (or furmark) to show this, and how close it would possibly get. I'd imagine if you don't get any significant deviations from a 1:1 graph, and you see that in most games, that could indicate you could be GPU bottlenecked. So a few CPU deviations might be good, because that means you CPU is faster than GPU sometimes/most of the time, but GPU isn't so slow that it is never faster than the CPU when it DOES have to compute more for a few frames
@@vocassen - you misunderstand. the Present wait from driver calls to flip is a fraction of the frame-rate assuming uncapped frames. So you can have two games that both run at 120fps but have different present times.
The true measure of engine efficiency would be some metric of how 'much' is displayed on screen for a given present delay.
Ie if you had two game engines rendering exactly the same geometry with exactly the same textures and shaders, you'd have different framerates - but that difference might have different ratios between CPU and GPU times on two different GPUs / OSes etc etc
So GPUWait tells you something about driver efficiency and game engine efficiency - but it's still not enough information as sometimes, greater efficiency comes from spending more GPU time to reduce CPU time.
A game might have greater GPU wait time than another - but perform better overall.
Having equal CPU and GPU usage is only ideal in terms of hardware utilization but that doesn't necessarily carry over to an ideal end user experience. Most games run GPU bound for good reason.
There's no "ideal" load but I'll always want it to be GPU bound as much as possible since it's the last in the chain (well I guess the monitor is but that's another topic). Why? You can easily fix a GPU bottleneck by turning on Reflex. If that's not available, you can simply reduce your graphics settings and/or resolution. You can even disable queueing entirely by turning on Ultra Low Latency/Anti-Lag but that isn't recommended as it causes your PC to be both CPU and GPU bound.
Fascinating content Steve. I love it when you have guests or go somewhere to do an interview as the content is always of a very high calibre. This is quickly becoming my favourite tech channel.