Pondering The Hybrid CPU Future w/
HTML-код
- Опубликовано: 11 сен 2024
- Are Hybrid CPUs really the future for desktop PCs? In this video Gordon chats with Dr. Ian Cutress from @TechTechPotato about what the future holds for CPU design.
Buy merch in our Crowdmade shop: crowdmade.com/...
Follow PCWorld for all things PC!
--------------------------------
SUBSCRIBE: www.youtube.co...
TWITCH: / pcworldus
TWITTER: / pcworld
WEBSITE: www.pcworld.com
#intel #amd #cpu
![BAK Jay - Find Out (For Your Love) [Official Music Video]](http://i.ytimg.com/vi/10FJO_XBCvY/mqdefault.jpg)
This was a really intriguing discussion. I hope to see Dr. Potato on the channel more - if not in person, at least joining Full Nerd streams.
Ian is usually lurking in the live chat during Full Nerd broadcasts. A "friend of the show", as Adam would say!
Very good Interview. And it's always nice to have an expert such as Ian Cutress.
I am always amazed on how much Ian knows, and his ability to explain it really really well!
I understand more of his writing though. Maybe more to do with his accent.
He reminded me as a lighter version oh Kimi Raikkonen. Almost monotones and straight face. Sorry Ian 😃
Hey I recognize that guy! His name is Gordon, right?
Great to see the meeting of these minds. Cheers Gordon and Dr C
Ian is a freaking genius! This explanation was flawless and brilliant 🎉
Interesting discussion, and really enjoying PC World having only discovering the channel a few weeks ago. It did get me wondering about the how the SPARC server I used to use would compare to AMD/Intel X86 CPUs. The data sheet I googled shows the SPARC M8 has 32 Cores, 8 threads per core and runs at 5Ghz. Getting benchmarks that run on both platforms would be tricky.
This was very informative, every time I watch Dr Cutress have a discussion with others I always learn something new. For example, the tidbit about phones have cores operating at different frequencies despite being the same core architecture / type that I did not know and I suspect others didn't either. So, glad that Dr Cutress decided to go the analysist route as with his plain speak skillset he can really get the tech info across as well as being a massive point of reference for others when dispelling fud.
Dr. Ian Cutress!
Love this colab.
Ian looks at ease here. Great to see both of you in the same video!
Interesting discussion. It seems like Intels P/E core scheme is working out ok, letting them get that multi-core leg up, but it's just one strategy. Still, works better than I would have thought. However, in a somewhat related note, they just need to execute for the server market. Just come out with their product. That's the real meat and potatoes, and they are still losing market share to AMD. All these delays are bleeding them dry.
I think you are going to see accelerator chiplets that go along with the main cores. Very small cores that are super efficient/performant at very specific tasks.
Good explanation! I'm very advanced according to TechTechPotato, because I run only E-cores on my 14nm Ryzen 3 2200G (0 P-cores & 4 E-cores; say 1-16W/core, from suspended to max load).
It's great seeing Ian here again!
Seems from a programmer's point of view such extra cores are rather annoying because it requires special handling so extra code. Instead just splitting your data across the threads and be done with it. It is already bad enough that for some reason the compiler is unable to automatically split the data across threads. That is not only an issue with thread unsafe languages like C/C++ but also the case with thread safe languages like C# and Rust.
Yep, people ignore the inherent problem when a resource pool is split. The Ecores share L2 cache and slow L3 access.
W11 & Thread Director was buggy but it was even worse using other OSes.
The true efficiency core idea for OS tasks and managing accelerators makes more sense. But they needed E cores to score points in MT creator benchmarks so they were clocked faster but are weaker at fp. Under heavy loads they clock too fast in ADL and become power inefficient (analysis article by Chips & Cheese)
while it is leaps more complicated, hybrid core architecture has been around for a while (phones) so its not necessarily new unseen territory. its just new to US on desktop.
@@jake20479 it's not being used to save battery on desktop .. it's used in a power inefficient way and made a dogs dinner out of OS scheduling.
You can discover cache size in a program to tune it, but stupid E cores change the capacity and what the processor is capable of from under you.
@@jake20479 There is a mountain of desktop software that is not coded for hybrid chips. Give me homogeneous cores over hybrid cores any day of the week.
But everything it runs is javascript, or unity engine… like very few programs code that low other than games and even then this is abstracted away mostly
I think P cores and E cores makes sense for laptops but not for desktops.
PCworld is killing it with this duo and videos segments
I'm generally open to the idea of some lower freq OS task & accelerator management cores.
But I feel moving application tasks from different types of cores is inefficient. Effectively Thread Director 2 is making 2 resource pools, if you background a program they'll migrate and the foreground task has many faster cores serving it.
But I did that when analysing chess with a process priority utility that could also set affinity. There's not anything I heard I wasn't doing on Haswell.
I am slowly switching/upgrading all my machines to AMD, i see no point to e cores at all, gimme all powerrr
I went with 7800x3d to upgrade my i9-9900K as I didn’t want to handle all that heat of the 13th gen intels. I got a full tower case so it fits a 420mm AIO but now I am thrilled I can just slap on a air cooler. I’m sensitive to thermals and not necessarily the power consumption but 7800x3d gaming performance was so impressive, I wanted to give it a try…
I always value Dr. Ian's input!
2:50 "8 more skylake cores" not for gaming there they are bulldozer cores
03:00 E-cores are not the same performance as Skylake, they have the same (ish) IPC. You forget that they clock much lower. And the IPC part also has quite a few asterisks.
The problem with E-cores is not about performance. It's that they cripple the P-cores in terms of AVX-512. So it's not like they add free performance, they are the reason why Golden Cove had to be cut. And there are people and their work loads that would prefer uncut Golden Cove P-cores, more of them, in exchange to the E-cores any day.
Mentioning Skylake cores makes me feel funny because I am still rocking my trusty old 7700k.
I am on 6600K right now. So going to 13700K (or even 13900K) will be great, love the idea of getting 2x the processor I already had but then those crazy fast raptor cove cores as well. The Intel A750 will end up as a bottleneck tho.
Ian using some obscure video editing software is still odd to me.
that would be a crazy upgrade
Over the last 25 years, I've exclusively only bought Intel CPUs. The geek in me, was very curious to see Alder Lake's dual core design at its release and I do find it interesting, on a tech level. But when it comes down to it, I don't want 2 different core types in my own system, for many reasons. My next build, being planned now, will definitely be AMD, for the first time. I'd probably still be buying Intel, if they had offered a 10-16 normal CPU without any stripped down "E-conomy" cores.
Would say 95 percent of my builds/purchases have been Intel. Looking through my old machines: I did find an AMD Athlon and budget college laptop. I'm considering 7000 series for the next upgrade (might wait 7800X3D or similar) or AMD build 5800/5800X3D if current gen, have only considered Intel for the low budget end though honestly it will come down to thermals and gaming performance. Some of the cheaper Intel 12th/13th just have P cores, but I do run VM on occasion so having all P cores is appealing/not having to disable e cores for certain games with issues etc. I did remember some of the past hybrid architectures and feel uncertain about it. Am still on 9th gen, but I agree the option for all P cores would have been great.
i just watched techtechpotato's quantum computing video, so it's cool seeing him here too!
I'm still running a Skylake processor (6850k). I wouldn't feel a 'need' to upgrade, except my motherboard is dying.
One might think skylake at 10 watts is an achievement... then you realize Apple m2 has Alderlake kind of performance at the same power envelope then you realize atom cores are puny cores.
if you need Hybrid CPU, the cores dont need to be different, just the same cores half running 5 GHz and half running 4 Ghz
@10:15 yes that's exactly what I've been thinking, you imagine someone going to PCPartPicker because an enthusiast like us told them to go there, and they search for a 12 core, they'll get 12700/K/Fs and shop them against a 5900X/7900X that have 12 P cores not knowing that 4 of the cores aren't hyper threaded and aren't as powerful as P cores, but those consumers may not know or care if they don't do reviews, I'm sure that has worked and will continue to work "Price per core" for at least, a small segment of the market, that has to be right imo.
Gordon's shirt game has been strong lately.
that's all the Disney sponsorship money put to good use!
Hey! What do you think about big Pcores like really big, with 4 threads, not just one ore two. Is that the future? What is even the e-core? Isnt it already a one core with shared cache and 4 threads?
I would really want to know, why are they so "much" worse than P cores and have worse latency etc. are they missing some instructions? caches? how can they be so much smaller? Is it because of the ring bus design and it esentially makes one stop for all 4 ecores vs 1 stop for p core?
I always try to think of E cores as cores that just operate at the most effiecient frequency while the P cores push it to really stupid levels. Also they dont bother the ring bus design that much right?
I cannot believe how little power my 12900k/3090 take. Playing my favorite game (FAF/Flight Sim 2020) it's usually 100-180 watts from the wall. Holy cow. My 2009 pc idles at 120 watts and that's with a Pci x1 graphics card.
In an ideal future, both the chiplets design and hybrid architect should exist hands in hands for every party I think, that's how we get both the performance of the hybrid and the energy effieicent of the chiplet and open up a level playing for the compeition - which is a win for every consumer
Intresting technology whit many possibilities
But my guess is 10 years from now we have core architectures that's scale and is variable that we don't need different cores
That the core simply run in efficiency mode til its need the performance
Or even its scales whitou steps from ultra efficient to ultra power
Nice video. Like the studio. Like people who can listen. good subject.
In theory the P-core and E-core CPU design sounds a lot like a twin turbocharger for car engines.
Conventional turbos can suffer from turbo lag. Pressure from exhaust gasses must be built before the turbine can spin around and suck air into the engine. Turbo lag and other parasitic losses are par for the course with forced induction.
With a twin turbo design, you can use a small turbo to quickly build up boost pressure at low RPM and let the big turbo deal with the high RPM range. Similarly P-cores can be used for more single-threaded tasks and use the E-cores for spreading the load on multi-threaded tasks. It's not like you'll be able to run at max clock frequency with all cores on a conventional multi-core CPU. Why not have a bit of both that high clock frequency single threaded performance and multi-threaded performance simultaneously?
Just as with twin turbo setups, the engineering can get really complicated. It's probably why it hasn't been done until recently. It probably also explains the close partnership between Intel and Microsoft on Windows 11. Thread management probably had to be redone from scratch in Windows to accommodate hybrid CPU designs. A completely hardware based solution for thread management might have simply not been viable.
When all is said and done a hybrid design doesn't sound too bad for gaming. Video games rarely fully utilize your chunky 16 core Ryzen processor anyway. Why not?
I would take a 8 core CPU with the area and cost of a 16 core CPU with all the spare area filled with L2 and L3 cache any day of the week. A supercharged X3D version of a 8 core CPU would reign supreme for games.
I was expecting a conversation of x86+FPGA or x86+RISCV style hybrid chips, not the P+E conversation. Still a great discussion!
More more of these pls! Invite framechasers! 🙏
Overclocker vs stock debate
@@roythunderplump More like tuner vs stock debate
It's interesting where at lower price points there are product segments where you get p and e cores for the same price as just p cores. Think it does make comparing products more complicated. I'm not sure if it's pure performance cores might be best for gaming. The hybrid cores is interesting, we're seeing in games where e cores might have to be disabled so the game doesn't drop on to an e core also. OS' will need to handle the cores and programs, Windows 11 or similar versions of Linux for the hybrid cores, where I can see P cores being better for my gaming case usage in theory.
Love the content atm. Cheers!
Great, now my Cinebench score drops in half when I watch youtube.
Intel in servers is very much the right direction of the market. AMD can have 128 cores in server market but having accelerators for specific tasks that can do well efficiently and at low cost is the future. You can cram as many cores into chips you want but itll only increase costs heat and power consumption.
Honestly they should go for a special gamer skew KX where they only have 12 Pcores highly binned and running at higher Clock out of the box. The KS can be replaced by this since people who want e cores don't really need the extreme OC capabilities this KX chip would offer!
Even gamers wants E-cores so the the background apps use E-cores and leave more thermal room, power budge, and just generally more performance to the P-cores that matter for FPS.
12 p-cores on a monolithic chip is physically impossible.
Intel maxed out at 10 core with the 10900K.
@@saricubra2867 Xeons with 30+ P cores exist and they "accidentally" showed off a 34 P core Raptor Lake (which probably is a prototype/experiemental part to compete with Threadripper)
@@modash1231 Nah, Intel has nothing against Threadripper and Epyc since Ryzen launch.
Just wait for AMD 3d 7000 series I don’t think Intel will be able to compete. Unless price comes into your decision making.
Randomly i called this in the p3 into e6600 era. With speeds not jumping as predicted, the answer was clear then as now: massively multi core credit card sized chips where cores are recruited as required. AMD has since shown how to run multiple chips will together.
I'd like to see an expansion card that can slot a desktop cpu in the future
Fun fact: that’s how some CPU’s were installed back in the 20th century! I remember my first AMD Athlon system back in the 90’s was like that.
Buena charla, se me antojo una Pepsi ya terminando de ver el vídeo...
e cores aren't Skylake cores. They're a modified Skylake core. They don't run SMT, so all the logic needed to run two threads at the same time is removed. So, it's based on Skylake, but not the same as what Skylake was on desktop. It's like at one point the rumor was AMD is working on a big-little design where the little core is based on Zen 4. Sure, Zen 4 with different sections of the ISA ripped out and no SMT so that 4 of those cores fit in the space of a single Zen 5 core. And that's really the issue. If you produce a big little design you want a ratio of about 4:1, so you're doubling the thread count for the little cores in the same space as a big core.
And thank you for saying the very thing I knew as soon as I heard Intel is putting out a big-little architecture, which is it's based on their limitations at the current time with "Intel 7". The ONLY way they were going to compete with AMD, on TSMC N7 and then N5 was to produce this hybrid design so they could boost thread count in the available space they have, and two, within an acceptable power budget because Intel 7 uses a lot of power at higher frequencies, so doubling the thread count in the same space of a single p core with 4 e cores, and then clocking them slower allows Intel a higher thread count and keeps the power consumption manageable. If Intel was always ahead of AMD on process nodes and THEY were the ones who could stay ahead with core count, they may have never made that big-little design.
If Intel could drop straight to Intel 20A, they wouldn't need e cores. They could put out a 32 core desktop part very easily and their p cores are excellent. They could even boost cache some.
To clarify. We didn't say they were skylake cores - we said they have the same performance as skylake cores. And you're wrong in saying they're modified skylake cores. The e-cores are Gracemont cores, updated Atom designs with actual performance behind them. I've gone deep into the microarchitecture with my coverage to understand the differences, and speak with the engineers who designed it.
Hybrid won't bring power saving at least for combination of Intel + Windows. But, Hybrid will bring a better MT perf at lower cost. Consumer Market is complex, some do video editing, some do gaming and some only watch youtube. Given Intel HEDT line will come back, it should satisfy P core only customers. Thus, hybrid will continue to cover most of the situation.
I have a bit of an objection to Gordon associating PC with Windows. Linux is even more PC than Windows because you are more in control of your hardware and system and what is more PC than that? Being the master of your system. Check, Gordon. :p
I hope that the future isn't about cramming more cores (by stacking them) because I already can't saturate my 3700X's 8 cores. Most applications don't rely on multiple cores, so unless we start multitasking a lot more in our computing habits or games magically get a new design paradigm that makes them more multithreaded, 256 cores in a consumer grade chip just isn't useful...
Basically, I hope we have more frequency wars than multicore wars
Basically its really intriguing hearing an ex-overclocker (and good one) saying that ,now all i care is stability.. not performance.
At some point it all comes to that.
It’s a big difference in use case too. It’s fun to overclock, and it wouldn’t shock me if Ian is still doing some overclocking as a hobby on the side. But his work demands that his system is operational and stable 24/7, so it shouldn’t be a huge shock that that’s what he now values as his income depends on it.
Ian rocks!
If you had an all E-Core design, wouldn't it just be a less powerful but efficient CPU? Basically like the 10th Gen "T" CPUs?
Good report but how about a Thread Directed compiling course, specifically how E cores can be coded for SIMD and/or parallel operation at any single task as simple as encode, decode otherwise relying on the E cores as an array. This has nothing to do with core task assignments but how integer operations can be accelerated in parallel. And Dr Ian you missed that Intel class at Innovation Days? Way too much trade floor time for you. mb
I definitely care about efficiency. Not so much because of the power consumption, but for the inevitable heat output. I think a lot of gamers do.
How was Intel able to make the 10th generation 10 cores and they can't with the new Raptor Lake cores?
Great interview, thanks.
Well the big picture is that Intel and AMD need to create power efficient SOC’s to compete with the ARM chips and they have never got this right or in the low wattage envelopes for all ranges of devices from small tablets to laptops to gaming PC’s with the same core design. ARM started small and just keeps moving up market and at their efficiency they will eventually push out everything else.
Except that from the processor side both Intel and AMD have architectures and cores that easily compete with the best ARM chips on performance/watt, except that they can also scale up to higher power and energy usage than the ARM chips when wanted.
The extra efficiency of SOC systems comes fromm having the RAM on the chip as well, which is great when you don't need to much RAM. There is a reason why you just don't see high RAM Socs, because it gets insanely expensive and difficult to do.
@@reappermen the apple m1 basic scales to I believe 32GB of ram as an option, max to 64 and ultra to 128
The m2 brings a 50% increase in max ram capacity as well
For consumer hardware these numbers are enough, but no idea how costly these are to Apple since their margins are so stupid high on even the basic products
@@aravindpallippara1577 I know the M1 and M2 are real outliners in general for ARM designs, but from articles and interviews I saw since the m1 came out, it was strongly hinted that the M1 chip alone without the ram already costs Apple more to manufacture than AMD spends for a Threadripper (both manufactured by TSMC). Presumably RAM falls into a similar problem, supported by the fact that you can see a very exponential growth of RAM prices with normal ram when upping the density of the modules.
So yes, you can build SOCs with good ram, but it iis noteworthy that they exclusively appear in products well above 1k$ in price range, mostly aimed at the near or above 2k range. For comparison, as far as I could find the average price for a laptop or pc in Europe and Northern Amerika seems to be somewhere between the 650-800$ mark depending on source, so for the vast majority of people it is just not an option to buy that expensive of a processor.
I just fanboyed a little here. I swear!
I've seen a lot of scoffing at and dismissal of high power consumption concerns, but I'd like to raise a different reason for why you might care. This one personally affects me.
If you live in a country with hot summers, you have to rely heavily on air conditioning to game. If your PC pours out heat from a 100+ watt CPU and a 450 watt GPU (or even a lot less) then your room becomes literally unbearable to be in very fast during summer, and if your air con isn't in the same room as your gaming pc or if it's just not very good, it seriously limits how much use you can get from the gear you paid for. That's before we even consider the power costs of running air con constantly. So the power concerns aren't necessarily just people being pedantic.
I forgot to mention I noticed a massive difference in room heating between when I got Ryzen 2700x and my very old intel 6 core (they made a hot and power hungry 6 core i7 briefly before they decided they were going to stick to 4 cores for the next decade almost)
An all e core (8 cores) SKU recently got leaked; i3-N300
I want non hybrid cpus, thats why I went AMD.
about the PC vs Mac discussion, I find it very similar to the iPhone vs Android debate. I will always choose a PC over a console for gaming, I like to personalize my experience, play with the controller I want or keyboard & mouse etc, and I've had friends ask me why I use an iPhone and not an android. in terms of PC vs Console, Android is certainly far closer to the PC experience, being easier to run different software and not be locked down to Apple's ecosystem. I feel like the Mac is really the console equivalent of a PC. I don't really use my iphone for much besides youtube, communicating with friends, and looking the occational thing up on google. it is a device that functions, and fairly well. also I spent a tax return paying it off so until it's useless this is my phone.
Edit: this comment is all over the place. but there's something in there intelligible I think
When is Thread Director 2 coming out? Soon is such a vague answer^^
Talking about Hybrid cores, do we have any clue on what's AMD next move for Zen 5?
I've seen people said it would be hybrid cores too, but different from Intel.
LOL Hybrid architecture is like 3 steps forward then 2 steps backward.
Ecore = THREAD to function range core Pcore = Main Thread to function range core they float around total amounts there are more then specified on all ranges have fun.
They should be adding features to the cores, not cutting them down. Intel still hasn't learned from Itanium and i860
imagine if intel going for the more cores route and at the time they reached their thermal limit and can't do more cores for general market and had to redo all of the things, and amd just still had much lower amt of cores but had better perf, thermals and efficiency, what would intel do while amd can just move to the more cores route or just use the same route
I'm still trying to understand how 8P + 8E cores in a 12900K is a better design than just using all P-cores, e.g. 12P + 0E (assuming it fits in the same die and same power envelope). Is it because E cores are more efficient with multicore loads, and 8P + 8E gets more work done than 12P if the load is heavily threaded? This is my best guess, but I hope someone could clarify.
PS. If the answer is yes, then why wasn't this hybrid arch done in earlier generations?
First thing im pretty sure it was stated as 4 ecores uses the same area as 1 p core so it would have been 10+0 instead of 12+0. E cores provide nearly 1/3 of MT perf, so replacing it with 2 p cores would prob meant alder lake getting crush in MT compared to Zen. As for the why not do it earlier part, i dont have definative answer but my guess is that the amount of development needed for hybrid arch on x86, alder lake isnt the first intel cpu to use hybrid back like 2-3 years there was this ultra low power cpu intel released that had 1+5(or 6, i forgotten) config.
Intel was forced to do E cores because their process nodes are very outdated vs TSMC nodes so adding cores gives them a big area and power penalty vs AMD. And because Intel was late to the party with chiplets as well, there is a limit to the die size for Intel products to have if good yields want to be maintained.
Thank you both for the answers. So it seems that they came up with a very clever, but expensive, way to squeeze the most out of their process node. As a consequence they needed to remove AVX512 support, a new scheduling system (Thread Director) and a new version of Windows.
PC power draw burning cables and cause fires needs to be toned down. Sooner or later something has to give.
PC Master race 4eva!!
Well I wouldn't mind some better efficiency numbers for regular desktop users. One can use integrated graphics when not doing anything high demanding then use efficiency cores for low demanding tasks.
Issue of course with schedule various tasks to specific cores so you don't get horrible performance either when you need it.
So far it has worked out for Android and Apple with these types of architectures. Then it will make more sense for enterprise to have one specific core type for the server they are running, may it be lots of efficiency cores or fewer but plenty of performance cores.
What Intel is doing is not about great low power performance but just getting better multicore performance out of the specific package. That sense I don't see AMD would follow.
I think that a 3rd competitor in the CPUs market is needed!
(Now that Intel got into GPUs, maybe we'll get another brand into CPUs market. Nvidia?)
windows power options Maximum processor state 90% and minimum at 5% on ryzen 7 2700x runs at 85w that's how I make my own E Core
Intel needed the e cores to compensate for not being able to produce 7nm... It seems to work well though. Its not new, smartphones all have different cores and speeds. Thats why i do think this wil be the future for pcs too. It makes sense to only use laptop kind of cores for browsing etc to keep power usage low. These e cores really use little power.
Go back to 2x system types.
1 for lots of cores..for professional use.
1x system for gaming (say 10 core rastest clock, with big cache).
Optimise the 2 different cpus for different uses...
Keep the gaming core cheap, n the heavy cores ,xxxxx cost.
It feels pointless the 13900k plays better than the core7k cause it runs faster in games, not the extra cores....
I disagree with this guy and agree with Gordon. Alder and Raptor Lake should have ten perfromance cores. It would future-proof the chips for gaming. Two more P-cores and half the E-core count would be better.
This guy is saying that it's down to power why we can't have ten P cores. However that is wrong as AMD have shown, with multi-chip-modules. I still doubt that ten P-cores would not run.
Intel are wanting tocompete with AMD on multi-threading though. Plus they don't want to give us too much future proofing, and we know they think that way.
Ian drinks something weird. The can says "Ke". Any clues what is it?
If your gaming on r9 7950x disable 8 core then overclock the other core to 6ghz
Man it was hard to watch this video because the volume difference between Ian and Gordon was such that I wanted to turn down the volume whenever Gordon was speaking because it was blasting my ears, and then Ian would speak and I could hardly hear what he was saying.
Audio? It's important to get it right. I stopped watching about 2/3 the way through because of it.
In the discussion of Hybrid Cores for AMD, would that possibly mean a combination of say Zen 2 with Zen 4 Cores? Or all Zen 4 Cores but with the "P" cores being the binned cores and the not so "okay" ccd be the "E" Cores side which will have power limits set to them by default at say 65/105watts max (for desktop) and not be allowed to throttle up to 170w? would that be a feasible option for AMD to go forward with? or will that make the processor unstable?
I am just curious if that is the way forward AMD could be looking at in light of the Zen 4/Zen 7000 series having ecomode now.
AMD could technically do that, but that would kind of be a waste. It would be more power efficient to design e cores on the newest architecture as well, since the architectures are always improving performance per watt.
It would not be to difficult to basically first design the architecture and the p cores, then make another design pass to find the ptimum point for a weaker core on the same arch to get the e cores.
And AMD already switched to chiplet design, so they can technically combine nearly anything in a fairly stable platform.
Zen 4 cores are the e cores that will get mixed with Zen 5 cores. I think the rumours say 16MB cache, 16 cores on a chiplet mixed with at least 8 Zen 5 p cores
I am a gamer and I care about efficiency, appearantly I do not exist... Either that or Gordon's notions are slightly outdated.
ไม่งั้นก็เอาโทรศัพท์มือถือของบริษัท apple บางรุ่นมาทำใหม่คือเราไม่เน้น CPUระดับกลาง หน้าจอเป็นสำคัญกินไฟน้อยสแตนบายไม่นาน
Hybrid foods like seedless grapes, cronut and Ginseng have now become commonplace so why not Hybrid CPU.
The same p/e argument could have been made about multi core cpus 15 years ago. Multi core complicated cpu/software design back then, we didn’t go back to one giant core to meet SLAs. Who knows but p/e may stick.
I’m not sure how relevant that is. As of right now only the consumer market is getting any sort of hybrid design CPU - whether that ARM chips in mobile devices or what Intel is doing in laptop/desktop markets.
Where the SLA’s come into play, nobody is even attempting this and it’s not on any public roadmap either. The risk is too great (from a customer’s POV), and the amount of work required by the silicon designers (AMD/Intel/Arm/Apple/etc) is immense. Right now the task scheduler only works in a GUI environment and is making decisions in a ‘dumb’ way: foreground task gets 100% of P-cores, everything else goes to E-cores. How does that work in a virtualized environment where there’s no GUI and (almost) every load is a high priority task? The answer, which is what both AMD and Intel are developing, is different machines with different performance/watt characteristics. Each machine is a monolithic design and optimized around a set of workloads suitable for that performance/watt envelope. It’ll then be up to the customer to schedule their workloads to the appropriate system based on their desired performance level for that application.
Hybrid design is good for mobile chips, but as long as AMD still making all P cores design i will stick with them till All P cores intel chip comes out.
From what I'm studying about computer's architecture hybrid design is the future.
AMD is keeping up with Intel because they have a really good core design, not because an unified core design is better.
Think about the slide that AMD has shown about Zen 4 vs Alder Lake Golden Cove: half the area for similar performance, this is why AMD is keeping up.
TSMC kinda helping here to me thinks.
@@FakeGordonMahUng Surely it is.
Except that the size of the cores doesn't directly matter for that. AMD's cores equal or beat Intels p cores single threaded, which is still the main work/bottleneck factor for most CPU limited scenarios.
And if you got a use case for MT important enought to consider while buying, you can just get a dedicated super high core CPU like and Epic or threadripper or Xenos chip.
@@reappermen The size of the cores is extremely important. If AMD had cores the same size of Intel Alder Lake P cores they wouldn't be able to pack 8 of them in a CCD.
👍
From my point of view, people shouldn't care about whether something "just works".
If they rely on stuff "just working", it means that they don't know what they're doing and what their devices are doing without telling them. I don't want a device from the bittenfruit company - it wouldn't tell me what it's doing and it would hide everything under a smooth GUI. I have pretty old hardware and it just works too. The difference is that I installed all the software myself and that I know what makes it work, meaning that if something didn't want to work, I would immediately know what's going on. Actually, I have encountered more problems with proprietary software my friends are using because you can't see the code and you have to rely on the developers that they made it to "just work" - and then my friends are angry when a printer doesn't work with their windows device. My PC never has these issues, because I USE ARCH and I couldn't be happier.
It is great video, my type of humor :D
P cores are a bit tastier. E cores require salt.
better without thr lollie water product placement
Nvidia calls their cores Job Threads
gordon muscle car person. didn't he say on record that coupes are silly and he'd never own one? most muscle cars were coupes gordon. Nice shirt though.
Ecore's are efficient underload but when there overload they produce more heat and consume more power than a pcore's and pcore's are efficient in overload and less effient underload than ecore's
I had high hopes for Alder Lake regarding gaming but it's not yet there to integrate E-cores cleverly on lower load tasks yet. Benchmarks suffered as a result, threads should be assigned properly first to have a future in gaming.
It works surprisingly well and will only get better.
@@BatTech I wouldn’t count on it. Right now the scheduler is ‘dumb’ because nobody has thought of a good/better way of scheduling tasks. Foreground task (limit of 1) gets all the P-cores. Background tasks go to E-cores. Foreground task can also utilize E-cores if those cores are idle and the workload requests them (eg video rendering).
Where this helps with gaming workloads is that the game gets full use of the P-cores (including 100% of the cache!) and all the bloat that is Windows background tasks get relegated to e-cores.
RIP cloud