Zen 5 CPUs CONFIRMED | RIP Hyperthreading - Arrow Lake LEAKS
ฝัง
- เผยแพร่เมื่อ 3 มิ.ย. 2024
- Whokeys 25% off code：RGT
Windows 11 Pro $23,2:biitt.ly/581eD
Windows 10 Pro $17,6: biitt.ly/9f0ie
Windows 10 Home $15:biitt.ly/XYm9o
Office 2016 $28,8:biitt.ly/cy7Vb
Office 2019 $48.6:biitt.ly/YInvw
www.whokeys.com/
Zen 5 CPUs CONFIRMED | RIP Hyperthreading - Arrow Lake LEAKS
Today we have some fresh leaks for AMD Zen 5 and Intel Arrow Lake CPUs, as a fresh set of leaks CONFIRM Zen 5 CPUs, and we have Intel Arrow Lake NOT having hyperthreading seemingly confirmed. Join as we examine the latest Ryzen 9000 leaks, thanks to yet more AGESA leaks thanks to MSI. Zen 5 processors are highly anticipated, and the leaks on the hardware specs and performance have been coming thick and fast - but what about Intel?
We have some fresh leaks on Arrow Lake CPUs, with the core count and clock speed seemingly CONFIRMED, as well as a confirmation of Arrow Lake NOT having HT / hyperthreading - does this mean it is RIP hyperthreading? We examine the Arrow Lake specifications, which again includes cores and clocks, and how it compares to what our sources have been saying and also discussing Intel's launch plans for the next gen CPUs.
Plus, we also have a small update for the PS5 Pro.
Subscribe - th-cam.com/users/RedGamin...
Join our Discord - / discord
Tip on PayPal - paypal.me/RedGamingTech
LINKS
videocardz.com/newz/intel-cor...
/ 1779159569933258760
/ 1779114599708868874
AMAZON AFFILIATE LINKS
US - amzn.to/2CMrqG6
Canada - amzn.to/2roF4XO
UK - amzn.to/2Qi6LMS
GreenManGaming Affiliate link -
www.greenmangaming.com/?tap_a=...
SOCIAL MEDIA
www.redgamingtech.com for more gaming news, reviews & tech
/ redgamingtech - if you want to help us out!
/ redgamingtech - Follow us on Facebook!
/ rgtcrimsonrayne - Paul's Twitter
/ redgamingtech - Official Twitter
0:00 Start
0:02 Intro
0:49 WhoKeys
1:39 Arrow Lake
8:22 Zen 5 Granite Ridge
12:01 PS5 Pro
Royalty-Free Music - www.audiomicro.com/royalty-free-music - เกม
Hyper Threading was introduced to make the P4 cores more efficient. These had several computational units for different types of operations. The math processing is done using a separate unit so with hyper threading one process might be engaging the execution engine while the other process is using the math processing unit. If both processes try to use the same unit at the same time one will have to wait for it to become available. The thing was that for less than 5% extra circuitry in the core it could handle two processes instead of one. But all the cases where processes had to wait meant that it wasn't able to process twice the number of instructions at the same time, but in general it was faster than a single core processor in total.
To allow for multiple threads the core need to have at least two sets of registers and be able to copy them to the execution units when the particular thread are to access them.
To improve the performance more units were duplicated allowing the processes to work more of the time without having to wait for each other. But that means that more circuitry had to be added making the chips larger and more complicated. This went on with each generation of processors becoming more and more advanced.
Now my guess why Intel is dropping Hyper Threading is because they have reached the point where making the HT more efficient meant that they got close to using the same area for circuitry that adding an entire core would do. At that point they could kick HT out and add cores instead. And without having to allow for all the weird things that is required for HT the cores can be simpler and more streamlined.
Now I'm not going to pretend to know that this is why Intel kicks HT, but it's a logical guess. The interesting thing will see what AMD does after this. My guess is that it depends a lot on what the results are for Intel. If they are able to streamline the cores, curb power draw and improves IPC while being able to raise the CPU clock then AMD will follow the single thread core path. There is nothing strange about this. It has everything to do with remaining competitive.
Thanks this was a damn good read.
👌
Hyperthreading is also to try and mitigate pipeline flushing/latency. Some instructions need to totally flush the pipelines such as branch instructions(jump,call etc). If you stagger the instructions running in the pipe line then if you need to flush one of the instructions then you still have at least one instruction executing to completion while the cpu is fetching new data for the other thread. It means you don't need to waste hundreds of cycles while you reload cache and fetched new data etc from main memory.
paused the vid, just to read this comment and the preceding comments, sounds like very reasonable hypothesis'
Geez if I would have wrote this I would have had about 5 - 10 people read it at most.
This is the very thing I've said more than once, you have to have replication of at LEAST the registers because as you follow through an instruction sequence that's what's getting updated and then I had idiots try to tell me there's no replication of circuits, I had a good laugh, they didn't like it and that was that.
Yes if you don't have replication of anything you are CONTEXT SWITCHING which is the worst possible thing a core can do.
The reason why Intel is getting rid of HT is you don't get a 2:1 uplift depending on what the core is doing. It can be close to that, but if both threads want to use the same thing then one has to wait, and if you want to get rid of that wait time you either need to replicate more circuits or don't use HT.
I'd say the reason why they're doing it NOW is the thing you DIDN'T mention as a main benefit, which is getting two threads performed in the space that's a TINY bit larger than a single core, so the BEST thing about HT is the efficiency of SPACE requirements.
Intel is moving to their "Intel 4" process node which means they can shove a lot more cores in a smaller space. Eventually I could see AMD doing the same thing but I think their implementation of SMT is better than Intel's so I don't know.
If AMD moves to something like TSMC N2 AND can get a mixture of SMT-1 cores, whether they are regular or C cores and can get 32 threads running I could see AMD moving away from SMT-2.
The OTHER main reason I see for doing this is better IPC. That's just fact. If you can't get 2:1 in performance in single verses 2 threads being processes by a core, that's a hit to IPC.
So, those are the reason I PERSONALLY think Intel is moving away from HT is better IPC performance, plus they can shove more stuff on the die with their Intel 4 process node.
I love how it's always "bigger number better" until it's too big and they have to change the scheme so it doesn't look old.
it's bigger number better on the same process node
Well, back in the day I remember hardware unboxed testing the i7 8700K and 7700K with and without hyper threading. Benchmark apps would work about 25% better with HT, but many apps would work better without it.
For games, the 7700K was getting a big hit on average fps and low 1%, while the 8700K got lower waverage fps but kept the same low 1%.
And this is with a 4 and a 6 core processor. I guess at some point you get so many cores that HT barely helps, and it could still cripple many apps that take advantage of having dedicated cores.
I'm not surprised. Hyper Threading is basically now provided by e-cores. They offload work from the performance cores. Since they are very tiny on todays manufacturing nodes and manufactures can cram in a lot of them there is less and less reason with every generation to still devote silicon to support HT. It also means, that the execution pipelines on performance cores are utilised efficiently enough (given the demanding workloads, they are running), that HT isn't necessary anymore. It also means, that the system/kernel/compiler scheduling that takes place has matured enough, where the majority of tasks are assigned efficiently to correct core types, hence HT is not necessary anymore. Also as more e-cores are added they see less and less %usage in real-world scenarios, so it also warrants dumping of HT. Also let's be honest a lot of programs still don't utilise threading to the fullest due to their core code being carried over from a different epoch. It's one thing to spot optimise specific demanding workloads and it's a different animal to re-write your softwares core loops (core logic) with that in-mind. It's hard to engineer against Amdahl's law as well, so a lot of software utilises "braided parallelism" or "burst parallelism".
That doesn't fit with HT being in the plans, but dropped later as they had problems. True that many E cores reduce the proportional benefit of Intel HT. But basically heavy MT benchmark code isn't heavy on cache so the ¼ area E core is effective en masse, while all P core would lose badly even with HT.
But you're factually wrong on scheduling, the Thread Director is not controllable by programs but will tell kernel to push AVX threads onto P core then off if the mix changes, the latency issues reported on Intel are a sign that it's not solved, Alder Lake didn't even work as advertised putting "background" tasks onto E cores, that was demonstrated as an RPL fix.
When you're talking about parallelising loops that's compiler optimisation setting up SIMD vector processing which is a form of pipelining on a single core.
HT mitigates memory stalls, which is why disabling it caused capacity problems on servers.
Just like not enforcing power limits and adding E core while dropping AVX512, this is Intel taking decisions to cut corners due to execution problems, under pressure to deliver benchmark review scores. The real issue is the P core are too big and too power hungry to match Zen glued together dual CCD CPUs, which efficiently used SMT.
IPC is ALWAYS based on the workload. This also means improvement to IPC moving to a next gen will be based on the workload. Since one architectural change may benefit some workloads and not others there isn't a good way to ACTUALLY calculate IPC. You can get IPC if you have a routine written to do exactly that, get an IPC, but once again it's going to be based on the instructions that were run during that routine.
So, IPC uplift simply looks at how much faster next gen is over current gen, when both CPUs are locked at a certain clockspeed, to removed clockspeed from being able to affect the results since after all, it's "instructions per clock cycle (IPCC if you will)"
I think I'll go with what Tom has released for now, which is between 17 - 20% when the avg. is calculated. So in some apps the IPC will be higher than this and in others, lower than this.
AMD, as they have now for the life of Zen will pick a pretty good list of current apps and get an IPC uplift for each, and then calculate an average based on that. This will also include a few games.
Just for general info, since games are so data intensive, IPC tends to improve a lot when cache sizes are increase and it doesn't matter if it's L1, 2 or 3.
The core count is bumping so high the HT doesnt matter much. I would rather they focus on improving their e cores and getting those TDPs under control. I would gladly take a ~20% bump in PPW then take the crap shoots of if HT scheduling is actually helping you. There is a surprising amount of real-estate taken by HT now on AMD. Just add it to the cache budget or add another core.
Heck even the NPUs are starting to be useful in conferencing and other apps.
Good , And we don't even need E-ecores or hybrid architecture
I wonder if Asynchronous computing works better with out HT.
Thank you!
so, what happened to the rumored bartlette lake for lga1700? nothing new rumor wise?
From what I heard it's just going to be 13th gen again. With the possibility of a model with 12 p cores instead of 8-16.
I tend to think the IPC situation on Zen 5 is going to be complicated and I also suspect all your sources are correct from a certain point of view. The bigger increases in IPC probably relate to AI tasks, given all the focus on AI AMD has probably made sure the NPU on Zen 5 is very performant because that's what everyone is in the tech sector is prioritising right now and the lower amounts relate to more day to days tasks.
Intel’s biggest issue in my view is power efficiency and platform longevity.
Hopefully, the use of Intel 20A or TSMC 3nm with arrow lake tames power consumption.
Yeah "hotplug" mean thermal spare cycling. Boost to +6GHz for some ms. Interesting new trick to single thread.
I wonder what will happen with 12th-14th gen intel when games start using up all the p cores i9 only have 8
It's less about the naming scheme and more about if there's a leap in processing per watt over the last 3 generations.
Come on Intel, you can do it.
I think it depends on the transistor count per mm^2
Remember getting a pentium 4 single core ht lol...
Core triple number will be fine. Honestly I lost interest in Intel's naming scheme when they went to 10000. I figured they would have changed it then because of how silly it sounds. This also goes to prove that we will adapt, because when you say 14900k it doesn't sound silly even though it did when they dropped 10000 series :)
But then you have to make sure its not the same core triple number naming scheme as Nehalem or Westmere. Normally this is easily solved, like how AMD went from HD to RX on their GPU line. But in this case you couldn't have the exact same "Core" branding. Intel really cornered themselves with this one lol
Thanks for not taking betterhelp sponsorship
hey is this one of those hype train videos?
I believe there is another hyperthreading security floor Linked to Hyperthreading found and Fixes for previous flaws slowed down cpu cores a lot so remove hyperthreading and drop fixes that slow it down and core extra performance & die shrink alow more cores to aliviate issue plus by dropping hyperthreading scheduling of cores is easier in OS. AMD did have as many issues with security only scheduling issues are X3d picking right cores for work load ie game that likes fast clocks to fast clocked core and more cache to slower core with more cache.
I had a fucking stroke reading your comment. Wtf did you say
I could care less if Intel wants to call Arrow Lake 1st gen of a new processor. They probably could have done it with 12th gen frankly with the beginning of their p and e cores.
But this is really completely new for Intel, moving to MCM.
Paul looks like he's been working out 💪
Roids
An extra £100 for just 10% increase in CPU performance??? MIGHT be worth if it also meant a 10% increase in GPU performance as well. I do expect them to charge £100 more for the Pro.
granite ridege vs harrow lake, ??? vs lunar Lake. when the second will output , i will have a lot a new competitor in the game work on NPU, just need to wait 2025 or 2026 for a nice competitive price for NPU, IA and other fresh stuff, and photon cpu and more, new spec new wafer base more stacking, i have 5 years to upgrade very fast in thec after will go to stabilaze.. gpu cpu apu npu chipset mb and memory ...
There is a persistent rumor out there, that there exists somewhere a real world workload where hyper threading turned out to be noticeably useful.
I only had a PS5 for a couple months before knowing i was ready to wait till the pro releases.
Pro for me will be a day 1 purchase.
😂over here the ps 5 got available like this year before that non hehe
Intel's new gen should be called Stupid Lake. I have I9 9900KS and done with their new E, P core shite with mega Watt consumption . Next CPU will be def a AMD Zen...
why did you buy the 9900KS?
"PS5 Pro. It's like Marmite."
- RedGamingTech
Keep your hands still.
I heard that hyper threading was going because Intel are replacing it with their new core scheduler which according to them allows for better performance than hyper threading. We will wait and see. Last year I also heard that meteor lake was a flop, having a lower IPC and max core clock than previous gen. Hopefully that’s not true or is fixed for arrow lake.
Zen5 x3d is my next upgrade. Gonna be a year from now
same, on 5800x3d now. I don't need Zen 5 x3d... But I want it.. Maybe FSim with VR will be better.
Why, oh why, would someone wanting to keep their identity secret be sending out slides with their identity watermarked on them?
Why do so many people make such a big deal over the name of something? Sheesh.
harder to understand performance tiers
HT takes up space on a silicon die and increases costs, so I guess Intel just figured it's better to invest that resources into E cores instead. Perhaps all models will now have at least 8 E cores.
SMT in fact improves power and area efficiency. So, it actually can save costs because you need less silicon for the same performance.
Either SMT is broken in Arrow Lake (some rumors were saying that). Or Intel deliberately takes a step back for now. Because they had a lot of security issues due to SMT. Professionals even recommended to turn off SMT on Intel processors. And Intel tries to compensate it with e-cores in the future.
@@avynian I think circumventing speculative execution vulnerabilities inherent to current Intel implementation of SMT is the real answer, too. Other people are saying purpose of removing SMT is to gain IPC, but that doesn't add up since p-cores now do heavy lifting while SMT provides greater efficiency and was achievable even with added e-cores.
The downside of course is that with these additional e-cores, newer Intel CPU's are WAY less power efficient and run much hotter. People used to make fun of Piledriver on the AMD side for its inefficiency but the current Intel lineup is even worse.
2 extra big P cores
And
64 E cores
When intel?
15th gen
Ive been here for a very long time, and Bro!!! your getting Jacked! very proud of how far this channel has come!
It’s never worth it to buy the pro versions of the consoles because the problem is the games themselves. The console will deliver for power, but it’s up to the developers to actually offer enhanced game modes for those pro machines. Which they never did.
30 fps gaming ☹️😓 sad for them
Didn't intel drop hyper threading in the past and then they brought it back a year later?
Nope they dropped it for most the 9th gen with only the i9 getting hyper threading. That's the only time they've come close to dropping it before.
They did when they went to the core 2 duo/quad. Pentium 4 went to Pentium 4 with HT, Core 2 duo/quad dropped HT, then i series brought back HT.
@@matthewtremain683 i was meaning the iseries line up but yes you are right they didnt have ht then when they did on the p4
@@matthewtremain683 should of remembered that as at the time i built i p4 system with ht felt so furturistic xD
R.İ.P Intel
Gains are starting to show, are you natty? :D
I wouldn’t mind if intel just simplified their lineup and copied Apple. They should offer a core 2, core 2 pro, core 2 max, and core 2 ultra.
Intel please don't mess up Arrow Lake as well.
who cares
@@royboysoyboy fine or even great? How? In performance they were mostly good enough to compete with AMD, but they were years behind in efficiency, had no upgrade paths and cost way more.
It's not fine if the more expensive 14900K can tie the 7950X in performance while consuming double the energy and being on a dead end platform.
The 7800X3D even outperforms the 14900KS in gaming while costing not even half as much and using less than one third of the power.
Intel needs to achieve a huge jump in performance and at least cut power consumption in half to become competitive again. But considering their latest duds "14th" gen and MTL that's very unlikely.
@@nipa5961the idea that Intel is less power efficient than AMD is very misleading by the fact that benchmarks are putting heavy load on CPUs. But in real world usage even for gamers the CPUs are not going full time at 100% instead most of the time are underloaded and even idling. Even if AMD is more efficient at heavy load, Intel is more efficient when the CPU is not on heavy load and way more when idling. In my case 80% time is productivity, internet browsing, TH-cam, movies and only 20% is gaming. So definitely in my case the Intel is more energy efficient than AMD. Obviously both, Intel and AMD, need to further decrease the power consumption for their CPUs, but is kind stupid to go full over Intel for something that is presented in a misleading way.
I was curious to search info about market share and surprised by the fact that Intel is the master when comes to laptops platforms. And laptops are very used in productivity in companies with thousands of employees. Even my international employer with over 200 thousands employees is using only Intel platforms, desktop and laptops. When it comes to only laptops there you need to have a less power hungry CPUs at tasks that don't require so much CPU usage and Intel does this very well.
So benchmarks from TH-cam are not a representation of what real world usage looks like.
@@royboysoyboy 7950X is 520€, 14900K is 580€.
No motherboard you bought to run a 12100f is also capable to run a 14900K.
There is also no current Intel CPU under 100€ too, the 14100f is 122€.
The 7800X3D is already 31% faster in games than the 7500f. AMD offers the same upgrade uplift within one single generation already as Intel in "three" generations. And there are still most likely 2 more generations to come on AM5.
Why do think "The average PC user" cares about single core performance so much?
I think people care about single core, multi core and gaming performance in the same way.
Even Intels current 5% advantage is more than evened out by AMD's 10% advantage in gaming.
@@royboysoyboy The 7800X3D is better in games on average by 10% compared to the 14900K, I never even mentioned the 30+% outliers.
The 14600K is 41€ cheaper than the 7800X3D but also much slower in games and consumes way more energy. Strange you focus so much about multi core performance this time since you said single core is the most important thing last time...
Also interesting you mention poor performance and instability because j2c switched publicly to AMD because he newer managed to get his previous 12900K system to run stable. Also there are tons of news right now about Intel 13th and 14th gen CPUs black screening and not working anymore after just a few months of use.
I also never said there is a universal best CPU for every single workload, but considering all the aspects I mentioned, performance, efficiency, prices, platform, upgrade paths and running costs there is sadly no usecase a Intel CPU would be the better option right now.
In gaming the 7800X3D is by far the best and way way cheaper choice.
In multi core scenarios the 7950X is the obvious choice.
A 5% single core advantage is a joke if you factor in all other Intel downsides right now.
And really? You would recommend Intel for web browsing and office... well, maybe it's important for someone if a page opens in 0,0020 or in 0,0019 seconds...
But I hope Arrow Lake will finally catch up to Zen5, because I want to have a choice!
I want AMD and Intel to be neck in neck every generation. It's sad Intel is so far behind like AMD was 10 years ago.
Games only use 16 threads. So if they are trying to compete with Ryzen they might as well make the 16 best threads they can and that means 16 cores. Then throw in some extra cores for the desktop and background programs.
That's exactly what Intel is not doing anymore since Alder Lake.
Intel needs to drop the HT to give them more power budget for more crappy e cores.
estrogen cores
egregious cores
HT seems redundant with estrogen cores. Pretty sure a single E-core is worth one thread on a fat core anyway. Windows is cocking it all up because if it dedicated a solution AMD would probably break.
Elderly cores
dont forget the side channel vulnerabilities. they're skipping a proper fix by dropping it entirely
253 watts? Jesus, remember when they made fun of AMD for releasing the 190w FX-9590? This is just getting sad, hell even De8aur had to go liquid metal and delidding to keep his 14900KS from thermal throttling with a monster of a liquid cooler and it was still hitting over 83c.
Hyoerthreading was a solution to a problem that doesn't really exist any more. Specifically how to schedule multiple tasks on limited processor cores.
I know some still consider that hypertheading is some sort of core adding magic. But its not. As more real cores are added it offers Increasingly diminishing returns and adds Increasingly onerous complications. IMO at 24 actual cores, the utility of hyperthreading is vanishingly small and the drawbacks outweigh the benefits.
but are they abandoning it on p series xeons?
@@jimecherry That's a fair question. However, I would say that in server workloads it makes more sense (just a tiny bit) as they are often dealing with the inputs of dozens, sometimes hundreds of users simultaneously. Each with a tiny request.
An ERP like SAP needs servers that divide the total run-time by as wide a margin as possible and damn the consequences. 400 users all making queries now and then. Tiny little requests that hardly move the dial of CPU usage individually. But SO MANY of them.
But even there I think MT has almost reached its nadir
No. SMT always offers a noticeable uplift. At least 20-30% on modern x86 processors. Diminishing returns are only an issue if the uncore logic cannot scale well enough. But this will affect both, SMT and physical cores.
Who care about more E-cores. We want more P cores. Not that E cores are useless but they are not P cores
mb
I think sony was doing a pro but I think they've changed course
HT is insignificant. Performance over time and per W is significant.
I will consider once chips are released and reviewed by reputable independent tech enthusiasts. All this leakage and hair splitting is kind of useless.
An Intel 14 gen CPU actually need less power than Ryzen when only doing lite loads. But as soon as you start to go outside this envelope the power consumption goes up on an exponential curve very quickly
Does infinity fabric hold the idle power consumption higher for AMD ? Or they simply don't care about it ?
@@harshivpatel6238 I think they do not care it is from the top of my head around 15 watt ish depending on CPU. As soon as you start to do some real work. Intel start tu guzzle energy
@@afre3398 for my 5th gen laptop intel i5-5200U, it's less than 5W idling. I'd like to see that for AMD laptop chips, that can guzzle 55+watts when needed.
@@harshivpatel6238 Dude we are talking about desktop processors here. I thought that was obvious, jeez!
@@afre3398 why should desktop CPU be any less efficient tho?
Paul hyping ps5 pro for a year and now that it's coming out it's meh 😂😂
funny how intel fanboys are now saying
hyperthreading is insignificant and does not matter
i would wait and see how it performs first before you decide if it matters or not
because intel has been producing absolute garbage these past few years so chances are this will garbage too
it'd need more cache if they want to keep the core fed & not waiting
Absolute garbage except for their 12th gen tbf
Also I bet that Arrow Lake will also be lacking gaming stability. Yah, lets go blue, no hyperthreading, no AVX-512 and no gaming stability. Gonna be one hell of a CPU. lol
second >.
कुणी कुणी म्हणायलेत की E-core म्हणजे estrogen core आहे. ते Efficiency Core असते. Estrogen बायकांचं हार्मोन असतं.
50 seconds ago
Who cares about vanilla Zen 5? It will be slow and will be lucky to match 13900k/7800x3d in gaming. Most people only care about X3D variants. ArrowLake is much more interesting. Intel really needs it to be good.
Arrow Lake needs to be extremely good to compete with Zen5. I hope it won't be just another disappointment like Meteor Lake and "14th gen"
Does Arrowlake socket design different? So there is no point of upgrading from say 14900K and instead build a new system (new mobo/CPU combo).
@@Hyperion1722 Arrow Lake is on a new platform. 14900k is a dead end.
@@mimon6706 OK. So am5 owners are in a better/ cost effective position to upgrade to Zen 5 compared to an arrow lake upgrade due to mobo replacement.
@@Hyperion1722 Yes. I built an AM5 system last year and I plan to upgrade to Zen6 in a few years. All I'll need to do is changing CPUs, no further upgrade costs.