At 20:28 or so, Bill of AMD misstates the WPRIME WR as 1.61, but it is actually 1.061. They will need higher clocks to beat the 16C WR and for now have only displaced the 7950X record for WPRIME, but not the 16C WR. The Cinebench numbers appear to have been accurately read and stated. Tons of CPU coverage on the way. You also now know why we were recently hanging out with Wendell to talk Intel failures! th-cam.com/video/oAE4NWoyMZk/w-d-xo.html Learn more about the Zen 5 CPU product details in our previous news video: th-cam.com/video/Y1yubL0h46U/w-d-xo.html Or the article for it here: gamersnexus.net/news/amd-announces-ryzen-9950x-9900x-9700x-9600x-zen-5-cpus-extends-am5-life-ai-cpus
I have a question: 4k gaming = CPU res to go by when DLLS is being used. If going from a res less than 4k to 4k is CPU going by 4k or the res less than? Don't want to spend more on CPU if upscaling is 4k to the CPU (GPU takes over part) thus, spend more for the Video Card.
5.8GHz = record 7950x 6.7GHz. 14% faster in this comparison. or +7%/year. what a "progress" with CPUs)) e.g. by Moore's law it should be twice as fast or +100% in 2 years.
We've been able to do this in windows for a while. Just had to know the right person to ask for a tool. And sometimes sign a nda or pinky promise to not leak it.
@@NegativeROG To be fair, in my day when I started overclocking with the P2s and P3s, you could get 50% daily use overclocks on chips ON AIR and sometimes x2 speed on water and more extreme cooling methods. I ran an Asus Dual P2 (and later P3) board for years with P2 300@450 stable/daily use and later a P3 450 at nearly 800. The generational jumps in frequency in those times were obscene compared to what we get now.
@@thetheoryguy5544That's marketing for you. Every single company does it. Is it good ? No, certainly not. But that's not the point of the video nor my initial comment.
@@GewelReal Because unfortunately for the average consumer, that's common practice in today's marketing. Like i said, every company does it, all of them. But at least, there's some reliable sources of information (like GN) that helps people to see through their BS.
@@simonseide9791 i guess there will be some good figures out 2-3 weeks after launch if most boards get a good bios. like going -0.2 or -0.3 in the mid range frequency if temp is at like 60°.
@@simonseide9791 You could certainly do a whole design of experiment but probably decent enough to just do their AI curve optimizer and then change a couple points from there. The automatic curve optimizer on my 5600 worked well enough but definitely didn't seem as stable on a couple points so I had to shift the whole thing up. Granted that's on an old X370 board so it's just nice to have all those features carry forward with new AGESA versions on older hardware.
So now we know what they were all doing in LA. Probably also why Adam and Willis were down there and did the full nerd podcast with Paul and Kyle last week as well.
@@rolandohiebert2144skill issue. If you can't figure out that your motherboard is cooking your cpu maybe you shouldn't get a unlocked model. Whether motherboard manufacturers are at fault or not isn't the issue here. Your lack of involvement for your own stuff is the issue.
@@antoniohagopian213Congratulations on your complete lack of a point. Raptor lake CPUs are degrading rapidly even with extremely conservative power limits, it's far from being user error.
Autodesk should prepare and make a 3D voltage/temperature/frequency map surface editor to easily edit the frequency surface (as it's now 3D, it's no longer a curve, it's a surface)
@@erkinalp Eh? What does Autodesk have to do with the whole thing? I used to work for them, never met people who were in any way connected with the computer biz that were less hardware oriented. Don't mean it as a bad thing, but just how they were 10+ yrs ago when I worked for them, the only hardware that was ever mentioned was the 3D mice. "mice", the 3dconnexion thingimabobs however you wanna call those things. So I doubt they would be interested in making something like that. But then again, any windows coder should be able to fairly quickly whip up an editor like that if they have devtools set up. I certainly could have when I was still a developer and had the tools for it.
@@nickmhc I don't see how a "neural engine" would be able to help with anything in this really, except maybe a strategy to find out optimal settings which you can do simpler, faster and easier without using a neural engine or gpu or anything like that, it's not complicated math. Easiest way would be to whip it up in excel or something, you don't have the granularity to make use of anything complicated really.
Man I'm excited for some new tech to drop, this is a great video. Love the relationship you have with AMD and other devs allowing for these interesting and informative videos.
Thanks for catching this. It looks like Bill misstated or misread the result he shared live. We have updated the pinned comment, added a pop-out corrections flag that will appear in-video in the top-right corner on most browsers, and updated the description.
3:17 this is exactly what was happening with my undervolted R9 5900X. It worked great at load but then at idle my PC would crash, and this wasn't mentioned in ANY of the undervolting videos I watched, and I watched atleast 20 different undervolting videos by major techtubers. It's the reason why I basically gave up on undervolting my CPU because it happened at almost every offset.
I think it's the same for everybody - I had a 5700X that would hold -30 all cores for Cinebench and OCCT, but fell over while watching TH-cam. You have to stability test in the low-load/high frequency zone and accommodate _that,_ even if you know there's more in the tank further down the curve. Well... you _used_ to have to do it that way😁
@@williamtopping The X3D always undervolt better because the boost frequency is capped - my 5800X3D does -30 for everything. But minimum LLC is always a good idea when undervolting - it's the undershoots that kill you down there.
Yeah, nothing super crazy but we will have to see how the launch units turn out. This video didn't give any inside into efficiency for example since you could only see total system power and ... well its XOC.
@@simonseide9791 The IPC improvement should be worth some efficiency, 5-10%, while the TSMC 5->4 node update is supposed to be worth 20%. So hopefully 25-30% overall.
@@concinnus That would be great since the power density was allready an issue on 5N. My 7800X3D for example is really efficient but it can still really get my fan running.
@@simonseide9791 The node upgrade also means a density increase, so heat density may be just as high. As long as AMD insists on cooler compatibility from AM4 and the attendant thicc IHS this will be an issue.
@@concinnus hmm the node update may or may not translate that directly to performance gains, more than likely it won't at least initially. The main issue is that the switching power levels do not scale the same way, and the same applies to a number of other relevant factors. If it wasn't like that, there would be no redesign needed, simply make the same exact chip scaled down in size for immediate performance increase from clock boost. We all know that isn't how it works. Would be cool if it was though. But not just that, IPC isn't even really so much where the game is, it's in things like cache optimization and even to a degree optimizing the execution path of specific instructions and using predictively fetched caches to eliminate wait states. So in other words optimizations not of instructions per clock but streamlining the execution of specific clusters or combinations of instructions that are used often like for example SIMD types triggering predictive fetch streams from cache and such. The low hanging fruit has been picked but the landscape isn't barren yet, by a long shot. It just gets more complicated and specific.
Their 1.393 wPrime is not world record x16, the WR by the intel is 1 second & 61ms, so 1.061 instead of 1.61. They misread, but were actually still off by quite some margin. I doublechecked hwbot just to be sure.
Now this is great tech content. Just HW Tech Enthusiasts doing what they love, overclocking and sharing knowledge. Thank you GN for the video, it's great!
@@Henrik_Holst AMD should make one wafer's worth of a new single-core chip. A vanity project, perhaps, but it would be cool to say "here's that 10 GHz chip y'all wanted. 😎"
@@benjaminoechsli1941 Impossible. Silicon is simply not suited for such high frequencies. That's what spurred the race to develop graphene processing in the first place (which is kind of irrelevant, now, as it appears quantum computing will leapfrog new electronic computing advancements)
@@benjaminoechsli1941 I don't think you have any idea of what you're talking about. Disable all but one of your cores. See how high you can clock it. If you're on AMD, you guaranteed didn't cross the 6GHz mark even on a single core.
The addition of the new curve shaper is really great! For my Linux server I use amdctl to 'shape' the voltage curves but it's only really limited due to the 3 available P-states
Super dope video, but if I could offer one piece of advice (IMO) for your audio mix, you have multiple audio sources, you can apply slight left right and centre levels to get more separation, it helps with being able to hear multiple voices at the same time without the mix getting too muddy.
At 4k gaming there is no difference between the 5800x3d and the 7800x3d (I know, I own both) I can't see this being ANY difference with the 9 series as that will also be GPU limited.
@@williamtoppingDepends what you are playing. Graphics stuff doesn't really need more right now, but there are plenty of strategy, simulation etc games where your CPU is the bottleneck on gameplay speed, not fps
These are some of the most fun videos, just because they're "just some nerds, hanging out and doing what they love" kind of things that takes all the industry BS out of the equation for the most part. Love the attitudes and just getting to see people have fun in spite of being in such a competitive or in some ways corporate space. It humanizes the huge companies.
It's even better for regular users as you can adjust things to more aggressively boost when the CPU is cool and back off as it gets warmer, resulting in lower temps but better performance for lighter threaded things like gaming.
that 3d curve is basically the same as ecu tuning rpm fuel and ignition timing an u sit an play with load sites an ur adjusting mixtures an using o2 senors an egt to make sure nothing is going to explode
Cool video (ha!), great way to slowly introduce us nerds to the next Ryzen chips without revealing too much. I wonder how XO is going to work when someone decided to ship an ARM desktop platform?
It's a pretty amazing journey. I've been around the oc community since the Celly 300 days, where we regularly ran 450mHz and some of us were close to 500mHz, *and* we had to fight the manufacturers to do this. And now, we have these crazy tools that are supported by the manufacturers to let us push the hardware with the craziest setups imaginable.
@@yes-ni1od I'm a professional writer and literally can mean both "exactly" and "figuratively." However, if your a stickler for tradition, then I guess I could revise to say, "When arctic cooling is literally from yes-n1od's mom."
Thanks for the video. Very interesting. The overlocking dealt more with tight voltage and harmonizing frequenting, instead of juicing more power and more cold. This is actually very good. This means there is more play in the frequency curves than chip manufacturers realize. They can make more efficient chips with current designs. Best comparison would be like tuning a CPU chip like a vehicle engine to get more operation efficiency.
I always wonder at these types of events, is there increased or active ventilation? 25L of LN2 expands to over 600 cubic feet, displacing and diluting oxygen in that enclosed space.
Hey, so a bit off topic - looking up any of your benchmark results on your website is a freakin nightmare. It's horribly organized and chaotic, and after half an hour trying to find something to help me pick out a CPU, I gave up and went on to a site I trust much less, but at least I can find some results. So you are by far the most trustworthy media out there, and your testing methodology far surpasses everyone else, but when time comes to pick out a configuration, I couldn't rely on you. :( Much love!
I'm likely buying this to replace my 14900k. I already had to replace the 13900k with the crashing issue (which intel did not want to RMA), would like to dump it before I start having issues with this one.
You had issues and they refused to help yet you bought intel again? 💀 I don't blame you though, it took many hours to get me convinced building Ryzen looking at AMD previous reputation.
@@facepalmqwerty I needed my machine, It's my work computer too. Building a new one right away wasn't in the cards. I bought the CPU before I even contacted intel and was testing after I was done with work on a spare drive.
Curve shaper looks almost identical to engine mapping. You take a few sensor inputs for both how the engine is actually running and what the actual input conditions are (fuel, air, timing, load etc), then use that to create a lookup table (which is often visualized with a map similar to the graphs shown early in the video) for how the engine should behave in any given scenario with a few additional safety tweaks based on input from other sensors (knock, EGT, oxygen etc). Genuinely if you know how to overclock and know how an engine works, you could probably tune one fairly competently without any real training or experience. It's almost scary how similar they are lol.
The design department hanging out with the XOC guys is a great idea. They get to witness their design being pushed to the absolute limit and taking notes on how they could improve upon it for the next generation. But did anyone else notice that the total system power, while braking world records was lower than the CPU power draw alone on the Intel CPUs. Absolutely incredible.
@@attepatte8485 Intel has been living off past success, offering little true innovation in recent years. Heck, they've been stuck on 8-cores since 2018. Nvidia is known to be just as arrogant and greedy as Intel but they bring true innovation. They're a multi trillion dollar company. The real deal. AMD must contend with both market leaders simultaneously. But one is much easier than the other.
When will we get a through explanation of each option and sub options of BIOS in BIOS help tooltips/help section, especially the OC settings? Will it ever happen before end of the mankind? 🤔
Intel also breaking the thermal records Their new Core Ultra processor has a TJMax limit of 110 C instead of 100, that means the Core Ultra processor won't throttle until reaching very close to 110 C. Some people are already skeptical about how it's going in the long term
@@sihamhamda47 Bruh that is wild. I bet many will have problems after about 2 years. Saying that because Nvidia 30 series GDDR6X cards had insane vram temps. 100c+. Not a single techtuber reported it. Maybe because vram junction temp sensor reporting was disabled back then or techtubers were lazy to check that. Either way, result was tons of cards with dead vram or artifacting. No other high end GPU 2nd hand value went down like 30 series did. I had two 3080s, had to use Copper mods to get the temps down to reasonable levels.
am5 until 2028, this curve shaper and ram software great, better efficiency and an actual new chip and not a repackage from last year with insane failure rates. Wtf intel and gg AMD again!
so glad I switched to amd, now I can actually get excited about future cpu releases cos I dont have to worry about changing my entire platform like with intel
@@LeftJoystickAnd Intel platform isn't even cheaper!!! If it was cheaper I'd be considering it. But I got a 5600 with B450 board and frikkin Bdie kit, all brand new, for LESS than 12400F with b660 and a cheap 3200 kit. (and not only is my 5600 faster at default settings, I got mine tuned to 4.9GHz and heavily tightened RAM). IMO only time Intel makes sense is either 2nd Haswell or older for pennies, because AMD's Zen 1 isn't that cheap yet, or (again 2nd hand) i3 12100(F) on H610 board for people that need highest possible single core perf without overspending
my first pc build was AMD 5900x 6900xt (custom water loop) and my next build was 13900kf (w/360mm aio) 4090 and I wish I would have stuck with AMD. The 5900x r23 score was 23k and will still do that same score today and it probably do the same score a few yrs from now and is game stable. my 13900kf did 41k r23 score (was game stable) but now wont do more then 37k and to get it game stable now I have to use default settings and that r23 score is 33-34k. I'm pretty sure the 13900kf is slowly dying.
Both have different architecture and different core layout. So there should not be similar issues with them. I would worry more about arrow lake having same issues as raptor lake.
I am eagerly awaiting on your follow up video to the Intel i9 stability scandal Steve. I really appreciate what you and Wendell have done by shining light on the situation and I really want you to keep it up and make Intel sweat.
2:45 what he says does also affect the older 5800X3D. With curve optimizer at -30, my CPU gets a boost from 4,25Ghz to 4,45Ghz while keeping the heat and powerdraw the same. On my 5950X -30 was not possible, since the idle voltage became too low and it crashed. With the curve shaper I hope that problem is gone. I also hope there is a (boost) cap we can enable, from my experience the last 10% clockspeed/performance raises the powerdraw/heat output by at least 30%. If I can tell the CPU to boost only to 5Ghz instead of 5,5Ghz, then it should run very efficient/cool. Right now I can only put in a manual clockspeed/voltage to achieve that, but that limits low idle powerdraw and higher single core boosts, since it is flat for all cores. For example: I wanted to lower the powerdraw of my 5950X, since it used 80W just for watching TH-cam. It boosted stock one core to the max 5Ghz @1.5V. Using CO or voltage-offset didn't help, since it lowered the idle voltage as well and crashed when lowering it too much. I only got -5W tops out of that. The only method that worked was to put in a manual clockspeed and voltage. The problem: It only works per CCD and not per core. So I had to set the clockspeed of all 16 core to 4,2Ghz@1.1V. Higher clockspeed was not really possible without raising the voltage and powerlimit, since it was constrained by the 142W stock PL. All that only to set a limit on the singlecore boost. In the end I lost 10% singlecore performance, but it lowered the powerdraw by 40% while surfing and 30% while gaming. And it lowered the max temp by 30°C!!! on a 280AIO.
@@castome7pi Don't forget to close every doors, windows, ventilation, etc. We don't want some heat leaking that'll lead to thermal runaway. Enjoy the breeze of pure Nitrogen.
As someone who's also feeling the summer heat, I was glad to see the TDPs of these new chips are going to be lower. That's the real efficiency indicator.
only if it has 3dcache on all 16 cores. come on amd, i dont want a cpu with different types of cache for different cores. if amd does that, the most epic gaming cpu will have 16 cores and be the 9950x3d. if its still only max 8 cores with 3dcache, the worlds most epic gaming cpu will still be stuck on 8 cores in 2024/5 and be the 9800x3d.
@@slopedarmor having cache everywehere increases cost, and decrease performance, because the x3d ccd is capped in boost mode while the none x3d can go higher when cache is not as needed, allowing mixed workloads, which ALWAYS are there when gaming, because games dont use the 16 cores so game will ALWAYS only use one ccd x3d cache fully
Thank you AMD for letting GN in and allow your engineers to nerd out while being filmed. I know a lot of companies would rather have some PR speaker instead.
it would do less, the 3d cache is fragile in that it's much more thermally limited. the extra cache also doesn't do anything positive in these kinds of tests (productivity tests), it's just not what that x3d cpus are designed to do. if they do game performance benchmarks then the x3d would be relevant there but then they would also use the 7950x3d to compare it to, they wouldn't use the 7950x.
@RadialSeeker113 Yeah, I know in terms of raw performance compared to a 9950x, I'm kinda surprised what amd have been up to as of late. Their chips keeps getting better & and better As for Intel, idk about them, but they could make a cpu faster than this, maybe? Actually, no, it would thermal throttle at 110°C xD
If grandma had a wheel, she would probably be a bike, and AMD cpus are immediately cooked past 6.6k under 250w. 14900k was boosted to 9.1k stable in May, on the other hand. AMD is a year behind what Intel did a year ago. Don’t tell anyone!
At 20:28 or so, Bill of AMD misstates the WPRIME WR as 1.61, but it is actually 1.061. They will need higher clocks to beat the 16C WR and for now have only displaced the 7950X record for WPRIME, but not the 16C WR. The Cinebench numbers appear to have been accurately read and stated.
Tons of CPU coverage on the way. You also now know why we were recently hanging out with Wendell to talk Intel failures! th-cam.com/video/oAE4NWoyMZk/w-d-xo.html
Learn more about the Zen 5 CPU product details in our previous news video: th-cam.com/video/Y1yubL0h46U/w-d-xo.html
Or the article for it here: gamersnexus.net/news/amd-announces-ryzen-9950x-9900x-9700x-9600x-zen-5-cpus-extends-am5-life-ai-cpus
I have a question:
4k gaming = CPU res to go by when DLLS is being used. If going from a res less than 4k to 4k is CPU going by 4k or the res less than?
Don't want to spend more on CPU if upscaling is 4k to the CPU (GPU takes over part) thus, spend more for the Video Card.
Can't wait, this 900 series generation looks promising. Maybe a GN OC session with BeardedHardware?
5.8GHz = record 7950x 6.7GHz. 14% faster in this comparison. or +7%/year. what a "progress" with CPUs)) e.g. by Moore's law it should be twice as fast or +100% in 2 years.
5.8GHz = record 7950x 6.7GHz. 14% faster in this comparison. or +7%/year. what a "progress" with CPUs))
e.g. by Moore's law it should be twice as fast or +100% in 2 years, not laughable +14%
Some say the space cat mousepad adds another 0.1GHz, but science isn't ready to explain it yet.
Also plus 5 levels to sneak in skyrim
but only from 6.8 GHz
The explanation was there but Schrodinger's cat ate it.
The editing of tiny video fragments at th-cam.com/video/pYWtP4tZe30/w-d-xo.html is exceptional. Captures everything this is about in an instant.
This checks out.
The fact that you can now do this fully in windows is insane.
Back in my day a restart and bios change was required every time you do a run..
Back in my day, 1.023 MHz was standard clock speed, there was no boost, no PBO. Temps were never an issue, though.
And you forgot to put the jumper into its default position
We've been able to do this in windows for a while. Just had to know the right person to ask for a tool. And sometimes sign a nda or pinky promise to not leak it.
@@NegativeROGMhz? Was it the 70s?
@@NegativeROG To be fair, in my day when I started overclocking with the P2s and P3s, you could get 50% daily use overclocks on chips ON AIR and sometimes x2 speed on water and more extreme cooling methods.
I ran an Asus Dual P2 (and later P3) board for years with P2 300@450 stable/daily use and later a P3 450 at nearly 800.
The generational jumps in frequency in those times were obscene compared to what we get now.
AMD : Breaks records 🏆
Intel : Breaks itself 💀
AMD also blatantly lies to consumers with shady benchmarks and graphs.
@@thetheoryguy5544 Intel does the same so... Nvidia as well, they all do shady graphs etc that make themselves look better
@@thetheoryguy5544That's marketing for you. Every single company does it. Is it good ? No, certainly not. But that's not the point of the video nor my initial comment.
@@thetheoryguy5544idk why AMD's marketing team still hasn't been fired
@@GewelReal Because unfortunately for the average consumer, that's common practice in today's marketing. Like i said, every company does it, all of them. But at least, there's some reliable sources of information (like GN) that helps people to see through their BS.
AMD binned to overclock
Intel underclocked into the bin
lmao
This made me chuckle more than it should xD
These Intel roasts keep coming and more evolved each time
@@CsQ_RandomRepository There's no need for additional roasting, these CPUs are roasting themselves already!
I want this on a t shirt
Zen5 based threadrippers are gonna be insane
You buying one?
@@Fin1nishingMove sure, why not? 😎
@@Fin1nishingMove why?
@@Fin1nishingMove enjoy your intel cpu crashing at stock speed brother
@@kylothow Ok, when can I expect it to happen?
Now we can truely say that AMD's products are...over 9000!
Except they have Pride merch on their website…
Underrated comment
@@LuccianoNova ?
@@LuccianoNova It's funny how people dislike or even hate a brand because they promote lgbtqia+. Who cares what some people want to be?
@@cayo2588just a random looser
Curve shaper actually looks really promising for undervolting. I can't wait for that tech to be in a CPU on the market!
So I do do not need a new AM5 board? 😅
Probably it will come with a bios update so all am5 boards that support curve optimizer will get curve shaper.
Its actually kinda crazy though, seems like you could spend weeks to test all of that.
@@simonseide9791 i guess there will be some good figures out 2-3 weeks after launch if most boards get a good bios. like going -0.2 or -0.3 in the mid range frequency if temp is at like 60°.
@@simonseide9791 You could certainly do a whole design of experiment but probably decent enough to just do their AI curve optimizer and then change a couple points from there. The automatic curve optimizer on my 5600 worked well enough but definitely didn't seem as stable on a couple points so I had to shift the whole thing up. Granted that's on an old X370 board so it's just nice to have all those features carry forward with new AGESA versions on older hardware.
What is more stable a 9950x at 6.6ghz or a 14900k stock 🤔
oooof
There is probably some more left in the 9950x to achieve equal stability.
Bruh
@@GamersNexus Steve doing the 50% slowed roblox oof.wav
And which one consumes more power
Wendell spotting in the wild there...
So now we know what they were all doing in LA. Probably also why Adam and Willis were down there and did the full nerd podcast with Paul and Kyle last week as well.
I hear Sir David Attenborough narrating...
At 1:12 for anyone wondering.
Shilling for his masters.
This event must been where they were to have filmed the previous videos together
I'm so happy to see the 2 engineers with Steve again
Always nice when a company pays a business trip for you so that you can promote them in return.
@@Fin1nishingMove Steve and the team pay for all their own air fare / accommodations
While AMD rips records, Intel's doing a "Mr. Gelsinger, I don't feel so good."
To bad, Intel oh yeah has the WW at 9100mhz and amd hasn't clocked high since shitty fx bulldozer CPUs
@@τετέλεστα13th and 14th gen i9s are dying at an alarming rate even when run at lower power targets.
@@rolandohiebert2144skill issue. If you can't figure out that your motherboard is cooking your cpu maybe you shouldn't get a unlocked model. Whether motherboard manufacturers are at fault or not isn't the issue here. Your lack of involvement for your own stuff is the issue.
@@τετέλεστα Almost like the recipe for high clock speeds is a bad architecture.
@@antoniohagopian213Congratulations on your complete lack of a point. Raptor lake CPUs are degrading rapidly even with extremely conservative power limits, it's far from being user error.
That curve shaper BIOS menu is going to be the source of many quiet, frustrating, nights
Autodesk should prepare and make a 3D voltage/temperature/frequency map surface editor to easily edit the frequency surface (as it's now 3D, it's no longer a curve, it's a surface)
@@erkinalpgreat idea but one step at a time. Could even use a neural engine (GPU or integrated) to come up with an optimization
Motherboard makers better put up a true graph...hope AMD helps them get there!
@@erkinalp Eh? What does Autodesk have to do with the whole thing? I used to work for them, never met people who were in any way connected with the computer biz that were less hardware oriented. Don't mean it as a bad thing, but just how they were 10+ yrs ago when I worked for them, the only hardware that was ever mentioned was the 3D mice. "mice", the 3dconnexion thingimabobs however you wanna call those things. So I doubt they would be interested in making something like that.
But then again, any windows coder should be able to fairly quickly whip up an editor like that if they have devtools set up. I certainly could have when I was still a developer and had the tools for it.
@@nickmhc I don't see how a "neural engine" would be able to help with anything in this really, except maybe a strategy to find out optimal settings which you can do simpler, faster and easier without using a neural engine or gpu or anything like that, it's not complicated math. Easiest way would be to whip it up in excel or something, you don't have the granularity to make use of anything complicated really.
Man I'm excited for some new tech to drop, this is a great video. Love the relationship you have with AMD and other devs allowing for these interesting and informative videos.
20:22 That wprime record is not 1.61s, it's 1.061s. While the 1.393 is impressive, it's not close to the record there...
Yep, I was about to comment the same.
How did 9960X do that?? I guess it's the architectural differences
They said record for 16 cores
Can confirm, 1.393s is at 49-th place in 16 core table.
Thanks for catching this. It looks like Bill misstated or misread the result he shared live. We have updated the pinned comment, added a pop-out corrections flag that will appear in-video in the top-right corner on most browsers, and updated the description.
3:17 this is exactly what was happening with my undervolted R9 5900X. It worked great at load but then at idle my PC would crash, and this wasn't mentioned in ANY of the undervolting videos I watched, and I watched atleast 20 different undervolting videos by major techtubers. It's the reason why I basically gave up on undervolting my CPU because it happened at almost every offset.
I think it's the same for everybody - I had a 5700X that would hold -30 all cores for Cinebench and OCCT, but fell over while watching TH-cam.
You have to stability test in the low-load/high frequency zone and accommodate _that,_ even if you know there's more in the tank further down the curve.
Well... you _used_ to have to do it that way😁
Check your boards VDROOP.
It's a board issue more than a CPUI issue.
I have -30 offset on my 7800x3d with no issues.
Weird, I cap the 7950X at less than a volt and it holds pretty stable.
@@JMUDoc
Dude, you're so lucky. My 5600 won't boot below -20.
@@williamtopping The X3D always undervolt better because the boost frequency is capped - my 5800X3D does -30 for everything.
But minimum LLC is always a good idea when undervolting - it's the undershoots that kill you down there.
5.85GHz to beat 6.9GHz with 7950X, they do deliver the generational 15% "IPC" gain.
Yeah, nothing super crazy but we will have to see how the launch units turn out. This video didn't give any inside into efficiency for example since you could only see total system power and ... well its XOC.
@@simonseide9791 The IPC improvement should be worth some efficiency, 5-10%, while the TSMC 5->4 node update is supposed to be worth 20%. So hopefully 25-30% overall.
@@concinnus That would be great since the power density was allready an issue on 5N. My 7800X3D for example is really efficient but it can still really get my fan running.
@@simonseide9791 The node upgrade also means a density increase, so heat density may be just as high. As long as AMD insists on cooler compatibility from AM4 and the attendant thicc IHS this will be an issue.
@@concinnus hmm the node update may or may not translate that directly to performance gains, more than likely it won't at least initially. The main issue is that the switching power levels do not scale the same way, and the same applies to a number of other relevant factors. If it wasn't like that, there would be no redesign needed, simply make the same exact chip scaled down in size for immediate performance increase from clock boost. We all know that isn't how it works. Would be cool if it was though. But not just that, IPC isn't even really so much where the game is, it's in things like cache optimization and even to a degree optimizing the execution path of specific instructions and using predictively fetched caches to eliminate wait states. So in other words optimizations not of instructions per clock but streamlining the execution of specific clusters or combinations of instructions that are used often like for example SIMD types triggering predictive fetch streams from cache and such. The low hanging fruit has been picked but the landscape isn't barren yet, by a long shot. It just gets more complicated and specific.
Their 1.393 wPrime is not world record x16, the WR by the intel is 1 second & 61ms, so 1.061 instead of 1.61. They misread, but were actually still off by quite some margin. I doublechecked hwbot just to be sure.
Yeah I caught that too, they were pretty far off.
They have seprate records for makers becuase intel has always been faster at specfifc benchmarka like Wprime and superPI
so curve shaping fixes the 'crash at idle' situation when using aggressive voltage curves, huzzah!
Indeed. Excellent work, AMD. Now let's see it in the wild.
If it does that, it would be awesome! As that was always the only problem i had
it's actually "crash just above idle", idle is fine
@@erkinalp yeah. What CPU is truly idle these days, with all the opaque Windoze and crapware garbage going on in the background.
Now this is great tech content. Just HW Tech Enthusiasts doing what they love, overclocking and sharing knowledge. Thank you GN for the video, it's great!
Remember that one article 20+ years ago about how we are on track to reach 10ghz? Before thermals snuck up and said nope.
that was also when it was all single core
@@Henrik_Holst AMD should make one wafer's worth of a new single-core chip. A vanity project, perhaps, but it would be cool to say "here's that 10 GHz chip y'all wanted. 😎"
@@benjaminoechsli1941 Impossible. Silicon is simply not suited for such high frequencies. That's what spurred the race to develop graphene processing in the first place (which is kind of irrelevant, now, as it appears quantum computing will leapfrog new electronic computing advancements)
@@benjaminoechsli1941 I don't think you have any idea of what you're talking about. Disable all but one of your cores. See how high you can clock it. If you're on AMD, you guaranteed didn't cross the 6GHz mark even on a single core.
@@heatsink9198 i'm completely down for hardcore single-core overclocking
I love that I got the actual product launch date from this video first. Thanks GN!
The addition of the new curve shaper is really great! For my Linux server I use amdctl to 'shape' the voltage curves but it's only really limited due to the 3 available P-states
Super dope video, but if I could offer one piece of advice (IMO) for your audio mix, you have multiple audio sources, you can apply slight left right and centre levels to get more separation, it helps with being able to hear multiple voices at the same time without the mix getting too muddy.
Always good to see them again, Bill and Amit are awesome guests.
Definitely staying with my am4 5800x3d for a while but man the 9000 series looks awesome
At 4k gaming there is no difference between the 5800x3d and the 7800x3d (I know, I own both)
I can't see this being ANY difference with the 9 series as that will also be GPU limited.
@@williamtoppingDepends what you are playing. Graphics stuff doesn't really need more right now, but there are plenty of strategy, simulation etc games where your CPU is the bottleneck on gameplay speed, not fps
This arch is being back ported to AM 4 my friend. 5900x will be the new arch. AMD does take care of its customers
@@williamtoppingdepends on the game, even this channel shows some games gain a decent bit of performance based on cpu. Some do balance out ofc
Shoot while they on 9000 series I’m hoping the 5800X3D get a lil cheaper 😅 once the new cpus drops I need to retire my 3600
These are some of the most fun videos, just because they're "just some nerds, hanging out and doing what they love" kind of things that takes all the industry BS out of the equation for the most part. Love the attitudes and just getting to see people have fun in spite of being in such a competitive or in some ways corporate space. It humanizes the huge companies.
seems like a nice toy for overclockers, the curve shaper that is
It's even better for regular users as you can adjust things to more aggressively boost when the CPU is cool and back off as it gets warmer, resulting in lower temps but better performance for lighter threaded things like gaming.
It’ll make Undervolting better
@@PineyJustice we'll now need an equivalent of Blender/Maya to edit the frequency surface
The wPrime record was 1.061, but still, 1.393 is *super* impressive.
Awesome work GN team! Thanks again!
14:55 The dry ice and LN2 draws Steve closer when he senses it.
I hope some of these improvements come to their GPU offerings!
love these overclock vidz and with the amd crew, its just a great vibe when yall just entertained us with some good education.
It‘s always great to see Bill and Amit in your videos.
that 3d curve is basically the same as ecu tuning rpm fuel and ignition timing an u sit an play with load sites an ur adjusting mixtures an using o2 senors an egt to make sure nothing is going to explode
True, cool way to look at it.
Hopefully my CPU doesn't have knock
true
i was thinking now i'm going to be mapping pcs too lol
@@RazzbowAs long as you're not running too much boost (Intel) you should be fine.
Cool video (ha!), great way to slowly introduce us nerds to the next Ryzen chips without revealing too much.
I wonder how XO is going to work when someone decided to ship an ARM desktop platform?
It's a pretty amazing journey. I've been around the oc community since the Celly 300 days, where we regularly ran 450mHz and some of us were close to 500mHz, *and* we had to fight the manufacturers to do this. And now, we have these crazy tools that are supported by the manufacturers to let us push the hardware with the craziest setups imaginable.
I'm building a new PC this winter. I'm not an over-clocker, but this is all fascinating to show the potential for the 9000 series.
really cool stuff that they did this with gamers nexus and before the release of the 9950X
When arctic cooling is literally from the arctic.
Not sure if you know what literally means
@@yes-ni1od I'm a professional writer and literally can mean both "exactly" and "figuratively." However, if your a stickler for tradition, then I guess I could revise to say, "When arctic cooling is literally from yes-n1od's mom."
@@snowboard_coach HAHAHAH nice respose
@@snowboard_coach you're* sorry had to
@@salted2096 fair
I wish I had the time to play like this, awesome coverage as always!
i wish i had the money . I'm afraid tho when i'll have the money i won't have the time or interest anymore .
seeing the word 9950x in the title and thumbnail is like waking up turning over and finding a stranger in ur bed like HUH
Great vid, would love to see a "maximum efficiency" competition where they try to hit the advertised 5.7 Ghz with the least amount of voltage.
I hope the launch of these behemoths will be smooth.
Steve was so happy to be near an XOC rig again, fun times.
Curved display for new frequency curves. I get it!
Thanks for the video.
Very interesting. The overlocking dealt more with tight voltage and harmonizing frequenting, instead of juicing more power and more cold.
This is actually very good. This means there is more play in the frequency curves than chip manufacturers realize. They can make more efficient chips with current designs.
Best comparison would be like tuning a CPU chip like a vehicle engine to get more operation efficiency.
I wanted to write an "Hey Intel .... " comment too, but there are already a few xD
@18:45 "Wanna go to the edge, man?"
I always wonder at these types of events, is there increased or active ventilation? 25L of LN2 expands to over 600 cubic feet, displacing and diluting oxygen in that enclosed space.
mixing L and feet, my brain please help
@@s-x5373 Sorry, curse of being an American. ~17 cubic meters
Hey, so a bit off topic - looking up any of your benchmark results on your website is a freakin nightmare. It's horribly organized and chaotic, and after half an hour trying to find something to help me pick out a CPU, I gave up and went on to a site I trust much less, but at least I can find some results.
So you are by far the most trustworthy media out there, and your testing methodology far surpasses everyone else, but when time comes to pick out a configuration, I couldn't rely on you. :(
Much love!
skill issue
@@TheGuruStud Finding info on a website shouldn't require skill.
I'm likely buying this to replace my 14900k. I already had to replace the 13900k with the crashing issue (which intel did not want to RMA), would like to dump it before I start having issues with this one.
You had issues and they refused to help yet you bought intel again? 💀 I don't blame you though, it took many hours to get me convinced building Ryzen looking at AMD previous reputation.
What's your use case for that i9? If its gaming an 7800x3d would've been better. If its anything else then yeah I guess ryzen 9
@@facepalmqwerty I needed my machine, It's my work computer too. Building a new one right away wasn't in the cards. I bought the CPU before I even contacted intel and was testing after I was done with work on a spare drive.
@@Derpynewb It's my everything, gaming and work PC. I will say Cities Skyline2 does max out the 14900k lol.
do it fast, when the word is all over you're only gonna be able to sell it really cheap.
0:26 a Wild Ed Corsa appears!
please put out a full review of the Kryo Sheets. or even a quick short review
Bro kryo sheet has been out for ages now.
@@JJFX- I know! And no one has done a review for it!! Which is extra weird bc GN usually reviews the things they advertise
These videos are so awesome, super excited to see the release and testing of this new amd chip
6.6GHz? In the backrooms? Before GTA 6?
It's so skibidi innit?
When 6.66 GHz?
@@daedalus6433 omg pls gib satans cpu! :D
@@daedalus6433 why not 666 ghz?
@@slopedarmor In the year 2666
Curve shaper looks almost identical to engine mapping. You take a few sensor inputs for both how the engine is actually running and what the actual input conditions are (fuel, air, timing, load etc), then use that to create a lookup table (which is often visualized with a map similar to the graphs shown early in the video) for how the engine should behave in any given scenario with a few additional safety tweaks based on input from other sensors (knock, EGT, oxygen etc).
Genuinely if you know how to overclock and know how an engine works, you could probably tune one fairly competently without any real training or experience. It's almost scary how similar they are lol.
love to see those engineers!
The design department hanging out with the XOC guys is a great idea. They get to witness their design being pushed to the absolute limit and taking notes on how they could improve upon it for the next generation. But did anyone else notice that the total system power, while braking world records was lower than the CPU power draw alone on the Intel CPUs. Absolutely incredible.
I wish the gpus were as exciting as the cpu department seems
@@attepatte8485 Intel has been living off past success, offering little true innovation in recent years. Heck, they've been stuck on 8-cores since 2018.
Nvidia is known to be just as arrogant and greedy as Intel but they bring true innovation. They're a multi trillion dollar company. The real deal.
AMD must contend with both market leaders simultaneously. But one is much easier than the other.
@@garyb7193 Yeah haha that is true. AMD is competing with two big corporations pushing them to innovate
@@attepatte8485 they need to bring back the ATI boys
the 99 cent fan on top cracks me up after all that effort
When will we get a through explanation of each option and sub options of BIOS in BIOS help tooltips/help section, especially the OC settings?
Will it ever happen before end of the mankind? 🤔
I can't wait to have to curve shape stress test per core!!
AMD: Breaking performance Records
Intel: Breaking Power Consumption records and Breaking itself
Intel also breaking the thermal records
Their new Core Ultra processor has a TJMax limit of 110 C instead of 100, that means the Core Ultra processor won't throttle until reaching very close to 110 C. Some people are already skeptical about how it's going in the long term
@@sihamhamda47 Bruh that is wild. I bet many will have problems after about 2 years.
Saying that because Nvidia 30 series GDDR6X cards had insane vram temps. 100c+. Not a single techtuber reported it. Maybe because vram junction temp sensor reporting was disabled back then or techtubers were lazy to check that. Either way, result was tons of cards with dead vram or artifacting. No other high end GPU 2nd hand value went down like 30 series did. I had two 3080s, had to use Copper mods to get the temps down to reasonable levels.
1:13 @Erdi Özüağ yine olay yerindesin 😄 başkaları totolarını kaldırmadan içerik üretir , sen yine okyanus ötesindesin. 👏
Damm, they even got cool ryzen hats and shirts.
Memory profile tuning on the fly is a dream come true
am5 until 2028, this curve shaper and ram software great, better efficiency and an actual new chip and not a repackage from last year with insane failure rates. Wtf intel and gg AMD again!
if you hear bill and amet is involved,, it's going to be legendary 🤙
Meanwhile at Intel: ⌛💀
so glad i didn't buy an intel when they're just failing left and right
@@arghpee Yeah. I do not understand how anyone who does any reading before buying their PC could buy an Intel CPU in 2024. AMD has been killing it.
so glad I switched to amd, now I can actually get excited about future cpu releases cos I dont have to worry about changing my entire platform like with intel
@@LeftJoystickAnd Intel platform isn't even cheaper!!! If it was cheaper I'd be considering it. But I got a 5600 with B450 board and frikkin Bdie kit, all brand new, for LESS than 12400F with b660 and a cheap 3200 kit.
(and not only is my 5600 faster at default settings, I got mine tuned to 4.9GHz and heavily tightened RAM).
IMO only time Intel makes sense is either 2nd Haswell or older for pennies, because AMD's Zen 1 isn't that cheap yet, or (again 2nd hand) i3 12100(F) on H610 board for people that need highest possible single core perf without overspending
@@LeftJoystick Easy. An Intel cpu is not shitting itself after I open a 24 MB excel document, as my 7800x3D does, what I will sell soon.
Love it that they are so proud and how easy they say " we are breaking records "
my first pc build was AMD 5900x 6900xt (custom water loop) and my next build was 13900kf (w/360mm aio) 4090 and I wish I would have stuck with AMD. The 5900x r23 score was 23k and will still do that same score today and it probably do the same score a few yrs from now and is game stable. my 13900kf did 41k r23 score (was game stable) but now wont do more then 37k and to get it game stable now I have to use default settings and that r23 score is 33-34k. I'm pretty sure the 13900kf is slowly dying.
Nice work guys, and with no pink nails. ))))
8:20 they should really just make a better version of Ryzen Master that's not form over function
Eh, Ryzen Master is already really good as long as you're not dialing extreme overclocks ten times in a minute.
they made the program idiot proof wich is ironic because those are the only complainers
That curve shaper thing seems very interesting
Still keeping my 14900ks. ❤️
Hopefully this one won't degrade like Intel Raptor Lake
Both have different architecture and different core layout. So there should not be similar issues with them. I would worry more about arrow lake having same issues as raptor lake.
I am eagerly awaiting on your follow up video to the Intel i9 stability scandal Steve. I really appreciate what you and Wendell have done by shining light on the situation and I really want you to keep it up and make Intel sweat.
I'm sorry to point this out, but 1s 61ms is actually 1.061s, so technically they did not break this WR.
It's funny when extremely cocky, condescending people (like the jerks on this channel) make an ass of themselves.
This is what I want to see! Lets goo! Looks like its going to be an absolute monster.
2:45 what he says does also affect the older 5800X3D. With curve optimizer at -30, my CPU gets a boost from 4,25Ghz to 4,45Ghz while keeping the heat and powerdraw the same.
On my 5950X -30 was not possible, since the idle voltage became too low and it crashed. With the curve shaper I hope that problem is gone.
I also hope there is a (boost) cap we can enable, from my experience the last 10% clockspeed/performance raises the powerdraw/heat output by at least 30%.
If I can tell the CPU to boost only to 5Ghz instead of 5,5Ghz, then it should run very efficient/cool. Right now I can only put in a manual clockspeed/voltage to achieve that, but that limits low idle powerdraw and higher single core boosts, since it is flat for all cores.
For example: I wanted to lower the powerdraw of my 5950X, since it used 80W just for watching TH-cam. It boosted stock one core to the max 5Ghz @1.5V.
Using CO or voltage-offset didn't help, since it lowered the idle voltage as well and crashed when lowering it too much. I only got -5W tops out of that.
The only method that worked was to put in a manual clockspeed and voltage. The problem: It only works per CCD and not per core.
So I had to set the clockspeed of all 16 core to 4,2Ghz@1.1V. Higher clockspeed was not really possible without raising the voltage and powerlimit, since it was constrained by the 142W stock PL.
All that only to set a limit on the singlecore boost.
In the end I lost 10% singlecore performance, but it lowered the powerdraw by 40% while surfing and 30% while gaming. And it lowered the max temp by 30°C!!! on a 280AIO.
But most games run on single core so it would be drop in performance.
you can set custom frequency boost limit in pbo settings, since am4
@@s-x5373 That setting is not in my bios.
@senti2175 Depending on the game a CPU rarely hits the limit.
Starting to sound like we are tuning a an ECU on a combustion vehicle! 😂 Love it!
3:03 "lots of stuff we find in extreme overclocking finds its way into the mainstream"
Yeah, like degrading intel cpus lmao
These two guys are seriously the best. Truly passionate in what they do.
Good for the records ... not so much for my summer room :)
If you have liquid nitrogen it sure is good :D
@@castome7pi
Don't forget to close every doors, windows, ventilation, etc. We don't want some heat leaking that'll lead to thermal runaway. Enjoy the breeze of pure Nitrogen.
@@fajaradi1223 XD
As someone who's also feeling the summer heat, I was glad to see the TDPs of these new chips are going to be lower. That's the real efficiency indicator.
@@benjaminoechsli1941 Literally XD my bedroom is a furnace in summer
Curve Shaper is what I'd like to tinker with! Had a ton of fun using Curve Optimizer, this one looks even more interesting.
9950X3D please!
Just buy the 5800x3d or the 7800x3d, at 4k you will be GPU limited, the 9x3d will be no different.
@@williamtoppingYeah, but people also want the cores for productivity work. People want the best of both worlds.
only if it has 3dcache on all 16 cores. come on amd, i dont want a cpu with different types of cache for different cores.
if amd does that, the most epic gaming cpu will have 16 cores and be the 9950x3d.
if its still only max 8 cores with 3dcache, the worlds most epic gaming cpu will still be stuck on 8 cores in 2024/5 and be the 9800x3d.
@@slopedarmor having cache everywehere increases cost, and decrease performance, because the x3d ccd is capped in boost mode
while the none x3d can go higher when cache is not as needed, allowing mixed workloads, which ALWAYS are there when gaming, because games dont use the 16 cores
so game will ALWAYS only use one ccd x3d cache fully
@@slopedarmor
extra latency traveling between ccds make 3dcache on both ccds useless for gaming. best is to have only one ccd for gaming
Thank you AMD for letting GN in and allow your engineers to nerd out while being filmed. I know a lot of companies would rather have some PR speaker instead.
I wonder what the 9950x3d would do? This is actually insane
Less because of the 3d cache limiting the clock speeds
it would do less, the 3d cache is fragile in that it's much more thermally limited. the extra cache also doesn't do anything positive in these kinds of tests (productivity tests), it's just not what that x3d cpus are designed to do.
if they do game performance benchmarks then the x3d would be relevant there but then they would also use the 7950x3d to compare it to, they wouldn't use the 7950x.
@@RadialSeeker113 No one cares about clockspeeds if Zen 5 IPC increase is just like Zen 3. Zen 4 IPC is mediocre, lower than Alder Lake.
@RadialSeeker113 Yeah, I know in terms of raw performance compared to a 9950x, I'm kinda surprised what amd have been up to as of late. Their chips keeps getting better & and better As for Intel, idk about them, but they could make a cpu faster than this, maybe? Actually, no, it would thermal throttle at 110°C xD
Depends on if they can solve the heat issue with 3d cache
I've been so unenthusiastic with the hardware cycle lately, but thanks to your content it's hard not to pay attention
Ahh yes, yes I understand all of this
Me too man, me too 😂
Coming in 2045 - Noctua RGB Diapers
bruh the cpu sips 241w at 6.6ghz
Intel sips 300w at a single core at 6.9 gigahertz
my xeon v4 sips 200w at 3.3ghz bclk
Awesome. These guys are always great.
These things probably could stay overclocked for months and yet last more than intel 14th and 13th gen fiasco.
Probably
If they're built right.
If grandma had a wheel, she would probably be a bike, and AMD cpus are immediately cooked past 6.6k under 250w. 14900k was boosted to 9.1k stable in May, on the other hand. AMD is a year behind what Intel did a year ago. Don’t tell anyone!
gotta need a transient load tool testing
as well as varying load tool testing so you can test stability at each point
i love the no bullshit design of ryzens. No ecore/pcore nonsens......its desktop PC, not mobile phones.
Curve shaper looks great, my 5800X can do -20-25 under load, but would crash randomly at idle, so I ended up with -10-15 as fully stable.
AMD should hire KingpiN as their overclocking PR guy.
Imagine getting kingpin boards from Radeon officially.
Kekw. Would be nice tho, even if I like these guys. They're cool too
Can confirm the liquid freezer III is one of the best aios on the market. My 420mm lf3 allowed me to daily a i9 7920x at 5ghz all core.
How is the i9 9960x so good at WPrime? That thing is an absolute dinosaur in tech terms. Releasesed over 5 years ago