Linux Prepares For Next-Gen AMD CPUs With Up To 12 CCDs

[skeevy420]

Yeah sure buddy, we "cant see the forest for the trees", it is not like you are a blatant AMD fanboi. E-cores of Alder Lake ARE high performance. They are the equivalent in performance of Core i10xxx. They are not ARM or something low power. So it makes more sense to have more of THOSE when you need 128-256 cores etc. Pure high performance cores that offer the ultimate per core performance aren't typically needed in that high numbers in the "clouds". There is a reason Intel went with that approach, and Intel has always been a leading force in x86 trends.... Intel's architectures are simply going to be much more efficient than AMD's, whether you are on a desktop, a workstation, or a server farm.

Funny you mention ARM. AMD actually has experience (and the patents) with hybrid ARM/x86_64 so they could blow Intel out of the water in regards to big.LITTLE if there was better software support for dual architectures. If they wanted to they could release a CPU that ran the OS on low powered ARM and ran high performance tasks on the x86. That CPU would be as efficient as a smartphone until you fired up X Plane on the high-powered x86 cores...and they could pair ARM with power efficient x86 e cores and do...hmm...let's call it little.LITTLE arm.X64.

For real world -- the PlayStation 4 used a similar setup. ARM did the OS and standby mode while x86 did the games. Intel isn't a leading force here in regards to efficient paired with non-efficient. They're playing catch-up and they're limited in strategy since AMD holds the dual arch patents.

To set the record straight with you -- I think the Intel e-core stuff is pretty neat for portables, but as we go from portables to appliances to desktops to workstations to servers I see them as less and less useful than a something like a scheduler that artificially limits a few cores and just turns them into e-cores or a systemd service that takes the first 8 cores and switches them to powersave with a 1 to 2ghz limit "systemctl enable generic-ecores". I say that because AMD has the dual arch efficiency solution that could really rock some socks if they staff their Linux and software divisions right and because e-cores seem like a good way to use cores that aren't up to par, don't pass the tests to be high-powered cores (doesn't mean I don't think they can't be emulated with software and schedulers). Upcycle instead of destroy and rebuild is probably more cost effective and better for the environment in that the CPUs are more efficient nor do they take 3-4x the power to make (source, build, test, destroy, rebuild, test, repeat if necessary uses a lot of power).

I'm more of a fanboi in regards to AMD GPUs because they're the shit on Linux and AMD APUs because they contain AMD GPUs which are the shit on Linux. When it comes to CPUs I'm partial to Xeons...but they don't have integrated AMD GPUs. It's not my fault that up until very recently integrated Intel GPUs have sucked or that dedicated Intel GPUs are non-existent or that Intel tick-tocks their goddamn CPU sockets too. Things like that tend to sway a person to use the competition. I'm gonna have to give Intel a few years to see how their dGPUs are for us, if they offer stable CPU sockets like AMD does, if their iGPUs get better, and so forth to see where I decide to spend my money on my next system...but so far it's looking like some sort of AM5??? APU with DDR5 in two or three years with maybe a new APU in my current AM4 box just because I'd like to have the last AM4 APU...no real reason other than why not.

[kylew77]

As for E-core only configurations, they already have a low-end line of SoCs which are exactly that. Currently, the latest of these consists of up to 4x Tremont cores. No doubt, they'll soon replace that with a version utilizing Gracemont. What'll be interesting to see is whether they stick with 4x as the max core config. I'm betting they will, given that Gracemont already offers a big improvement in terms of IPC, frequency, and AVX2 support.

Thank you for that by the way. The E-core only configuration you are referring to is that what Jasper Lake is? Are you referring to its successor or something I don't know about?

[jaxa]

1) E cores don't have to have the best absolute performance in any kind of load. 4 e-cores equal 1 p-core in die area. That means that instead of a 128 core Ryzen you can get in a theoretical scenario 512 Intel e-cores.

Absolutely wrong. AMD's big cores are about twice the size of Intel's E-cores, and the dense "cloud" versions like Zen 4c will be about the same size. Except they'll have hyperthreading from the start. Although to be fair the E-cores will probably get hyperthreading in a future generation.

128-core Bergamo is only a 33% increase in core count from 96-core Genoa, but the cores are denser and less chiplets are used. AMD will also put out 256-core, 512-core, etc. eventually.

Thank you for that by the way. The E-core only configuration you are referring to is that what Jasper Lake is? Are you referring to its successor or something I don't know about?

The current E-core-only configuration is Jasper Lake (and Elkhart Lake) using the older Tremont core. We actually don't know if Intel will do E-core-only product lineups for consumers again. I think most people who use mini PCs and low TDP netbooks/laptops would prefer to have at least one P-core, so Alder Lake ultra mobile (1+4 and 2+8) ought to replace Jasper Lake. 2+8 in particular would be a fantastic replacement for something like quad-core Jasper Lake Pentium Silver N6005. The processors for cell phone base stations ("Snow Ridge" w/ up to 24 Tremont cores) could get E-core-only successors.

Some weirdo on here tried to insist that the previous Atom cores aren't true and honest E-cores, not realizing that Intel rebrands things all the time.

[billyswong]

AMD officially never said something like that. But that has always been the unofficial approach their fanbois/shills/trolls whatever want to call it had. Remember for example when the first Ryzen was introduced. It was bad in terms of IPC, it lacked proper fast AVX2, and it wasn't that power efficient, but AMD's approach was to offer "moar cores for the money". This lead to the narrative that AMD was "punishing Intel" and were the "people's champions" for "breaking Intel's 4 core barrier". How easy it is for people to forget i wonder?

And it is not like Intel was actually greedy or "bad" or wanted to keep the desktop at 4 cores forever. Intel simply put didn't find any reason to stack the desktop with more cores at that moment in time while for the vast majority of desktop/gaming workloads 4 FAST cores were more than enough. Hell, even TODAY for most people in the desktop/mobile space 4 cores are enough.

But that was AMD's communication environment approach, whether officially declared as such or not. Of course when Intel stumbled upon 10nm delays and Zen 2 and especially 3 managed to bridge the gap with Intel's IPC and efficiency, they changed their tune. Suddenly they didn't offer "moar cores for the money" anymore, and they kept the desktop at 8 cores instead of providing more. But no one called them out on being greedy like they screamed at the top of their lungs about Intel a couple of years earlier....

And now that Intel's fabrication issues are becoming a thing of the past, suddenly AMD "remembered" the "moar cores for the money" approach. It seems AMD ALWAYS attempts to offer moar cores for the money when they lose at everything else. That explains the 12core CCDs

Ryzen 9 3950X have 16 cores and it works on a mainstream AM4 motherboard. Are we living in different parallel worlds? Else I don't see where your statement "they kept the desktop at 8 cores instead of providing more" came from.

[grok]

The current E-core-only configuration is Jasper Lake (and Elkhart Lake) using the older Tremont core. We actually don't know if Intel will do E-core-only product lineups for consumers again. I think most people who use mini PCs and low TDP netbooks/laptops would prefer to have at least one P-core, so Alder Lake ultra mobile (1+4 and 2+8) ought to replace Jasper Lake. 2+8 in particular would be a fantastic replacement for something like quad-core Jasper Lake Pentium Silver N6005. The processors for cell phone base stations ("Snow Ridge" w/ up to 24 Tremont cores) could get E-core-only successors.

Some weirdo on here tried to insist that the previous Atom cores aren't true and honest E-cores, not realizing that Intel rebrands things all the time.

I suspect lots of cheap laptops will use the die with 6+0 cores and so you don't have to think about these cores if you don't want to. 2+8 cores : premium ultramobile, perhaps fanless (or that's for 1+4), perhaps 5G modem, everything soldered. 6+0 : 15W CPU, RAM slots, M.2 slots, RJ45, HDMI, USB-A.

mini-PC : might use either.

[kylew77]

The current E-core-only configuration is Jasper Lake (and Elkhart Lake) using the older Tremont core. We actually don't know if Intel will do E-core-only product lineups for consumers again. I think most people who use mini PCs and low TDP netbooks/laptops would prefer to have at least one P-core, so Alder Lake ultra mobile (1+4 and 2+8) ought to replace Jasper Lake. 2+8 in particular would be a fantastic replacement for something like quad-core Jasper Lake Pentium Silver N6005. The processors for cell phone base stations ("Snow Ridge" w/ up to 24 Tremont cores) could get E-core-only successors.

So if you were looking to buy a new Chromebook it would be a good idea to wait for the product refresh in the new year and get something with at least 1 P core then? If I'm reading between the lines here right. I was thinking about pulling the trigger on a Jasper Lake 4 core Pentium Silver model this year but now I'm thinking about waiting.

[coder]

If they wanted to they could release a CPU that ran the OS on low powered ARM and ran high performance tasks on the x86. That CPU would be as efficient as a smartphone until you fired up X Plane on the high-powered x86 cores...and they could pair ARM with power efficient x86 e cores and do...hmm...let's call it little.LITTLE arm.X64.

Hmmm.... dosen't seem like you posted that at 4:20, where you live.


That's all I'm going to say about that particular flight of fancy.

For real world -- the PlayStation 4 used a similar setup. ARM did the OS and standby mode while x86 did the games.

LOL wut? Had to look that one up...

https://en.wikipedia.org/wiki/PlaySt...iary_processor

I think it's more like an I/O processor they probably use to keep from tying up one of the main CPU cores on slower I/O tasks. Given that this was introduced in 2013, it would've probably been on par with a Gen1 Raspberry Pi, if not even slower.

Having dedicated I/O processors in computers is nothing new. You wouldn't context-switch between them, so it doesn't matter what ISA they use. In fact, the PS3 actually embedded a PS2 CPU for similar purposes, I think (as well as backward compatibility with PS2 games).

[coder]

Thank you for that by the way. The E-core only configuration you are referring to is that what Jasper Lake is? Are you referring to its successor or something I don't know about?

Yeah, I should've posted a link to Jasper Lake. There's also Elkhart Lake, which is more for industrial embedded applications:

• https://ark.intel.com/content/www/us...sper-lake.html
• https://ark.intel.com/content/www/us...hart-lake.html

I'm not having any luck finding leaks of what the generation based on Gracemont will be. However, it'll almost certainly be Gracemont-based, since Intel is even reusing those for next year's Raptor Lake mainstream CPUs.

[coder]

Absolutely wrong. AMD's big cores are about twice the size of Intel's E-cores,

I didn't find this figure, when I looked for it. However, the die size comparisons would suggest otherwise, given that Alder Lake's die also contains GPU.

Do you have a good source for this?

Although to be fair the E-cores will probably get hyperthreading in a future generation.

We shouldn't assume that. With a couple exceptions, ARM never went down that path. And Intel quickly walked back from the SMT-4 they put in the original Atom.

128-core Bergamo is only a 50% increase in core count from 96-core Genoa,

OMG, maths! 128/96 = 4/3 -> 33.3% increase.

I can see how you might arrive at "a 50% increase", but the usual way that people will interpret that statement is "Bergamo has 50% more cores than Genoa", which is clearly wrong.

AMD will also put out 256-core, 512-core, etc. eventually.

I'm wondering how long this core-count race will continue. Are you aware that Epyc already has a way to segment the CPU into up to 4 NUMA domains? It's called the NPS setting. And the reason they do it is because SMP scaling starts to break down, when you reach that many cores.

https://downloads.dell.com/manuals/c...pyc-2ndgen.pdf

So, at NPS4, you're treating a 64-core CPU like 4x 16-core CPUs that happen to share the same package. I guess this could continue up to as many memory channels as it has, but I wonder if certain economies of scale will break down, first. For one thing, scaling to even more cores in a socket requires more expensive thermal solutions.

ARM is already planning around 192-core CPUs, based on their N2 cores. So, it looks like they'll be at the forefront... although, since they lack SMT, AMD will still have the most threads/socket.

The current E-core-only configuration is Jasper Lake (and Elkhart Lake) using the older Tremont core.

Intel has put E-cores in their Atom-branded embedded server CPUs for about a decade, now. The current Snow Ridge line has up to 24x Tremont cores:

https://ark.intel.com/content/www/us...now-ridge.html

...okay, I see you mentioned it later. Anyway, I'm leaving that, in case anyone wants the link.

We actually don't know if Intel will do E-core-only product lineups for consumers again.

With E-cores as strong as Gracemont? Of course they will! Jasper Lake hasn't even been out for 1 year, and the "Intel 7" production line is probably backlogged on just Alder Lake and Sapphire Rapids. However, once some capacity starts to open up, there's no doubt they'll refresh their entry-level SoCs.

I think most people who use mini PCs and low TDP netbooks/laptops would prefer to have at least one P-core, so Alder Lake ultra mobile (1+4 and 2+8) ought to replace Jasper Lake.

The Ultra Mobile they announced has a 96-EU iGPU. It's clearly a premium solution, not entry-level. Even Lakefield (pairing 1 P-core + 4 E-cores) was reserved for the highest product tier of MS Surface.

The only question is how competitive 4x or even 8x Gracemont cores will be against the competition from Mediatek. However, Intel can't sell these at a loss, and Mediatek's chips are probably fairly cheap. So, that's going to tie Intel's hands on pricing, which will in turn tie their hands on die size.

[coder]

So if you were looking to buy a new Chromebook it would be a good idea to wait for the product refresh in the new year and get something with at least 1 P core then? If I'm reading between the lines here right. I was thinking about pulling the trigger on a Jasper Lake 4 core Pentium Silver model this year but now I'm thinking about waiting.

Depends on how urgently you need it. Given that I found no leaks of a Jasper Lake successor, my guess is that it'll be a ways off. You could be looking at the better part of a year, before it's in machines that are available to order.

Still, if I didn't really need an upgrade, I'd probably hold off for Gracemont or the next thing from Mediatek. Don't assume you'll get a P-core by sticking with x86, but you would get a decent IPC & frequency boost + AVX2 (256-bit). Currently, Tremont and before have just SSE-family (128-bit) instructions.

Edit: also, the GPU should be upgraded to Xe. Jasper Lake seems to use Gen 11 (Xe is Gen 12).