Leaked Patch Confirms AMD Zen Will Have 32 Cores Per Socket?

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
The crew at TechFrag have been digging around in what is believed to be a leaked Linux patch and have come to the conclusion that AMD Zen based processors will feature up to 32 physical cores.

A leaked Linux patch on LKML.org, first spotted by The New Citavia Blog, suggests that AMD Zen based processors will feature up to 32 physical cores. The patch also hints at the similarity of parts of the “Zen” and “Zeppelin” codenames. The Zeppelin codename was first mentioned back in August last year, and parts of the patch identify it as a “family 17h model 00h” CPU.
 
I think this is just setting the future upper limit to the number of cores per socket for the server socket. And certainly not that in 2017 you will be able to purchase a 1GHz 32 core / 64 threaded AMD server chip costing over 10 thousand dollars.
 
are people running AMD in virtualization? If not, what would you need so many addressable cores for?
 
I think this is just setting the future upper limit to the number of cores per socket for the server socket. And certainly not that in 2017 you will be able to purchase a 1GHz 32 core / 64 threaded AMD server chip costing over 10 thousand dollars.

Same thought went through my mind as well theoretical number rather then 32 core Zen when AM4 arrives :) .

Since 16 cores for the desktop already makes little sense (due to software not written well enough) it would only end up on the server side of things ....
 
This seems to suggest there will be server versions of Zen.

I was under the impression that Zen was the desktop/mobile model, and K12 was going to be the next server arch...
 
Zarathustra[H];1042119106 said:
This seems to suggest there will be server versions of Zen.I was under the impression that Zen was the desktop/mobile model, and K12 was going to be the next server arch...

K12 is ARM ...
 
are people running AMD in virtualization? If not, what would you need so many addressable cores for?

Yes and AFAIK is actually quite popular for that. We currently have 8 AMD virtualization servers. 8 cores/cpu are quite good for the price.
 
Seems about right. I read somewhere (rumors) that the mainstream CPUs were going to be focused around 8 cores but they might have some monster special APU with CPU/GPU with 16-32 cores. Not sure if this article is just regurgitating that same info or if this is a second source.

I personally still use the 8 core 8350 and I like the extra cores because I do a lot of work from home that involves virtual machines so its nice to have the extra cores to run my VMs and not have it bog down my system in general.
 
And I'm still waiting for that one too.

But that is not going to take of until they are able to run some engines and even then game development cycle will have some impact on this as well.
Vulkan is really good but the thing is that the Khronos group takes their time with things...
 
Yes and AFAIK is actually quite popular for that. We currently have 8 AMD virtualization servers. 8 cores/cpu are quite good for the price.

I ran my home ESXi server on an FX-8120, later upgraded to an FX-8350 for years.

great low cost application of these chips, especially since they support ECC (provided your motherboard BIOS does)

Biggest limitation - for me - and why I eventually switched to actual server hardware was the fact that 4 DDR3 slots effectively limited me to 32GB RAM (unless you get some ultra rare, and very expensive 16GB unregistered ECC DDR3 sticks), and if virtualization likes anyhting more than many cores, it is lots of RAM. The power use also wasn't stellar. My current dual 6 core (with HT) Xeon board uses about the same amount of power as my octacore AMD FX chip did.
 
I'm not the only one thinking it.

aRPw3xM_700b.jpg
 
Here's hoping for 8-core consumer desktop chips.

I'm not opposed to this, but what they really need to do is get 4 core chips with competitive single core performance out the door.

Once they have accomplished this THEN they can start thinking about many core designs.

If they don't stop the whole "let's make up for poor single core performance by throwing cores at the problem" then Zen will be a disappointment, and they probably won't be around for long...
 
I'm not the only one thinking it.

aRPw3xM_700b.jpg

No. No you're not.

Intel has double-down on increasing the efficiency of their architecture. And it's paid some fairly HEFTY dividends for them.

Meanwhile, AMD keeps trying to brute-force performance, throwing clocks and cores at everything. And it's, patently, NOT working.

Sure! There's a COUPLE of niche markets where these kinds of components make sense.

But that's all they are. Niche markets. And that won't keep them afloat forever.

iStock_000005443998XSmall-200x300.jpg
 
I really hope AMD isn't trying to launch a 32 core processor for the consumer market.. general task/gaming on an intel xeon 14C/28T is pretty abysmal due to lack of proper software. Virtualization on the other hand..

Though on the Opteron side, I can barely give away the darn things. :(
 
This proves that their will be a 32 core chip. YAWN Without performance numbers this is basically a waste of a leak. Unless the performance matches the chip count this leads to nowhere.
 
This proves that their will be a 32 core chip.

To me it does not. It shows that 32 is the maximum possible # of cores per socket for a server socket that has not been officially discussed. I wonder what the limit is on lga2011-3 (since Intel currently has 18 core processors)? Must be 32 or 64.
 
Last edited:
Hoping one day the software gods will make use of more than 1-3 cores? It could happen.

I think AMD was working on something that would enable multicore work as a single core. I wonder what happened to that.
 
I think AMD was working on something that would enable multicore work as a single core. I wonder what happened to that.

I believe they officially stated that reverse hyperthreading was never in the plans.
 
Related: 48-bit memory addressing in AMD64 means all new Athlon 64 motherboards coming out in 2002 will support 256K gigabytes!
 
Not sure if more cores is needed for everyone, but I just bought a Supermicro MBD-A1SAi-2750F-O with the 8 core C2750 processor for a new NAS box. It should be able to run a router, plex server and basic media server / nas operations with plenty of headroom. I will know more once I finish building it. I looked at AMD, but they didn't have anything with server grade motherboards and similar performance. However, to be honest this a pretty niche market.
 
So is it four 8 core dies on one chip or two 16 core dies?
Neither. It's based off assumptions made from a faked slide before AMD introduced the Zen architecture (incorrectly that Zen is based around a 4 core module as a unit) and an incomplete understanding of what was being masked off in the shift. The original source is "Dresdenboy", an AMD fan who often gets things totally wrong.
 
No. No you're not.
Intel has double-down on increasing the efficiency of their architecture. And it's paid some fairly HEFTY dividends for them.
Meanwhile, AMD keeps trying to brute-force performance, throwing clocks and cores at everything. And it's, patently, NOT working.
Sure! There's a COUPLE of niche markets where these kinds of components make sense.
But that's all they are. Niche markets. And that won't keep them afloat forever.
iStock_000005443998XSmall-200x300.jpg

And how come the mobile market is dominated by ARM , then again while you are posting this where is the proof of AMD brute forcing can you link to something Zen that proves this ?
 
I am still waiting to see the next revolution in Mantle technology
:D:D:D

Vulkan, DX12, and LiquidVR all say hi! :D:D:D

edit: And as usual the trolls come out spreading their headache-inducing ignorance in the thread. Anyone who honestly thinks this chip would target the consumer market needs to just leave every tech forum ever, for the rest of their lives. Maybe get reincarnated as someone who actually has a modicum of common sense.

This chip is something that has been known about for almost two years or so already. If anyone bothered to look at the diagram they'd see that it's blatantly a HPC APU design.
7e42c9447e754167c85105ffe1a1d866_L.jpg


This shows a GPU die, HBM stacks and a CPU die all on an interposer, connected with GMI links (guess this replaces HyperTransport) and also 4 DDR4 channels. Which alone is enough to discern that it's not intended for socket AM4 because AM4 tops out at dual-channel.

Summit Ridge tops out at 8 core 16 thread. This was shown in leaked roadmaps (real ones, not fake ones) a year ago but people keep denying it so I'm not sure why I'm reiterating it again.
 
Last edited:
Vulkan, DX12, and LiquidVR all say hi! :D:D:D
None of those are "the next revolution in Mantle technology".

Khronos used parts of Mantle as a starting point for Vulkan, but it is not Mantle. Anandtech has a good overview: http://www.anandtech.com/show/9038/next-generation-opengl-becomes-vulkan-additional-details-released

AT said:
In fact Khronos has confirmed that AMD has contributed Mantle towards the development of Vulkan, and though we need to be clear that Vulkan is not Mantle, Mantle was used to bootstrap the process and speed its development, making Vulkan a derivation of sorts of Mantle (think Unix family tree). What has changed from Mantle is that Khronos has gone through a period of refinement, keeping what worked in Vulkan and throwing out portions of Mantle that didn’t work well – particularly HLSL and anything that would prevent the API from being cross-vendor – replacing it with the other necessary/better functionality.

DX12 accomplishes many of the same goals as Mantle, but it is very different from and not based on Mantle.

LiquidVR is a DX11 based SDK in AMD's GPUOpen.
 
They're not "revolutions" but they are indeed evolutions of Mantle. Mantle was never intended to be a full-on API with longevity (at least not in terms of being used for games past its introductory period.)

DX12 uses Mantle code in it. Go look at the SDK's, much of the DX12 stuff is literally copy-pasted from Mantle, just slightly re-worded in various cases. Vulkan is based off Mantle, it is Mantle and it isn't at the same time.

LiquidVR indeed also uses Mantle, as AMD themselves stated publicly that Mantle was halted in terms of being utilized as a public API but continues on behind the scenes.
http://semiaccurate.com/2015/03/03/amd-breaks-new-ground-liquidvr-sdk/
http://www.extremetech.com/extreme/200286-not-dead-yet-amds-mantle-powers-new-vulkan-api-vr-efforts

I honestly don't know how you came to the conclusion that LiquidVR is DX11-based when some of the things LiquidVR does are literally impossible to be done under DX11 :confused:

Not sure why you thought I was implying DX12/Vulkan were literally the same thing as Mantle. They are derived/evolved from Mantle but they are still pretty different under the hood today for obvious reasons. Even then it'd be best for Vulkan to "win" over DX12 but most likely it won't happen because it seems certain entities enjoy stifling process in that sector.

edit: I see about DX11. AMD said LiquidVR leveraged Mantle for much of the engineering but at that time it was DX11. They did say that LiquidVR will be implemented under all "relevant API's" which is where I got my line of thought from. Point still stands for me, they said numerous times LiquidVR was Mantle powered and for much of what they want to accomplish, it's easy to see why and how.
 
Last edited:
This proves that their will be a 32 core chip. YAWN Without performance numbers this is basically a waste of a leak. Unless the performance matches the chip count this leads to nowhere.

No it doesn't. Windows 7, 8, and 10 Pro/Enterprise x64 can use up to 2 physical processors with a max of 256 cores. Just because a Linux kernel can use a max of 32 Zen cores doesn't automatically mean that AMD will actually make a 32 core Zen. Are there any 128 core processors out there to please the theoretical Win 7/8/10 max?
 
They're not "revolutions" but they are indeed evolutions of Mantle.
No, they're not. The only thing semi-directly related to Mantle was when parts of it were used to speed up Vulkan development, but Vulkan is not Mantle or a descendant of it. glNext had been in development for years, and Vulkan retains many parts from that process. Read the link I posted. It's a very good overview of Vulkan.

DX12 is not based on Mantle. As AnandTech's coverage pointed out, while there are some superficial high level similarities, the details are very different: http://www.anandtech.com/show/7889/...level-graphics-programming-comes-to-directx/2

You're completely confused about LiquidVR as even the very front page of AMD's documentation clearly states it's a DX11-based API.
 
Back
Top