[WCCF] AMD 14/16 nm Opteron and FX Processors with up to 20 Cores in 2016-2017

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
Yes, but for most people more GPU performance is also waste, unless you're gaming. In fact even 4 CPU cores is waste too, 2C is sufficient for most people. Yet they buy 4C desktop PCs... ;)

Most people buy 2C. I would be amazed looking on the SKU lineups if even 25% sold CPUs are quadcores.

Yes, but even 6C is more than 4C. :) And the games can still be designed for more threads if intended to be released on desktop too. If you design your SW well, it should be able to make use of more threads, as we've already seen with several games making use of more than 6 threads.

The important thing is that the consoles have started this trend of more threaded games. Once you're past the single/double threaded thinking, an important barrier has been passed. Then you no longer hard code what tasks should run on each CPU core. Instead you need to divide your work into separate parallelizable tasks and throw them onto the OS scheduler, regardless of whether 6 or 12 cores are available. I'm not saying it's easy, but having good frameworks and libs supporting that certainly helps. Like the TPL framework in C#/.NET, although for games similar lower level libs have to be used of course.

It's simply the way to go, since IPC and frequency increases do no bring much more CPU performance these days.

Consoles had 6 threads+ since 2004. Didnt change anything.

Also scaling is another issue. Doesnt help you much if 6 for example gives you 20% more performance. But the quadcore clocks 25% higher. Not to mention your inherit problem of of all the applications with 4 threads or less. not to mention the large dependency on the main worker threads performance.

So where does the Intel HEDT CPUs fit? They do not have any IGP, but are not generally regarded as server CPUs.

HEDT chips are regarded as workstation/server chips. Without workstation/server, no such chips would exist because the market is too small.
 

Fjodor2001

Diamond Member
Feb 6, 2010
4,149
559
126
Most people buy 2C. I would be amazed looking on the SKU lineups if even 25% sold CPUs are quadcores.
Note that I'm talking about the desktop, where there are lots of 4C CPUs sold.
Consoles had 6 threads+ since 2004. Didnt change anything.
Only the PS3, so it did not make sense to increase the parallelism then in most cases. Because the market would be too small for the extra effort needed. Now both PS4 and XBONE have 8C, and use 6C for gaming. That's a huge difference. And we're already seeing it in the next-gen games, which are becoming more heavily threaded.
Also scaling is another issue. Doesnt help you much if 6 for example gives you 20% more performance. But the quadcore clocks 25% higher. Not to mention your inherit problem of of all the applications with 4 threads or less. not to mention the large dependency on the main worker threads performance.
Correct to some degree. But the exact impact of this is hard to predict. Is it 5%, 10% or 20% higher frequency? With later process nodes (14/16nm) chances are that more CPU cores can be packed onto the same chip without reducing frequency that much. Intel will have up to 18 cores and 36 threads on the 14 nm Xeon CPUs for example.

So the higher frequency possible on a 4C vs 8C on 14/16 nm will not be able to compensate for the additional processing power provided by the 2x amount of cores if used correctly.
HEDT chips are regarded as workstation/server chips. Without workstation/server, no such chips would exist because the market is too small.
Primarily workstation I'd say, but the border is not strict. As an example, we used Xeon CPUs on our workstation computers at the previous company I worked for.

Anyway, if priced lower, the HEDT platform would be more widely used in other segments. But in order to maximize profit Intel has intentionally chosen not to price its CPUs based on die area, which is understandable.

If AMD provides pure 6/8C versions of these next-gen CPUs without iGPU at a reasonable price, I'm sure there's a market for it.
 
Last edited:

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
Note that I'm talking about the desktop, where there are lots of 4C CPUs sold.

Still small vs 2C isnt it? Celeron, Pentium, i3, half the Richland line etc is all 2C.

But we are derailing the thread and going in a loop again.

Anyway, Its about K12. So its ARMv8 cores only and not x86.

Up to 20 ARMv8 64bit cores in 2016/2017 for dense servers.

Also in case anyone needs a refresh on Anandtechs article on the subject:
http://www.anandtech.com/show/7990/amd-announces-k12-core-custom-64bit-arm-design-in-2016
 
Last edited:

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
So where does the Intel HEDT CPUs fit? They do not have any IGP, but are not generally regarded as server CPUs.

AMD could make similar ones, but priced lower. Then they can get the high end gaming crowd on the wagon who pair their CPU with a discrete GFX card anyway. And for all other use cases where HEDT is used today of course.

Intel HETD are rebranded server/workstation chips, they are not developing this specifically for the desktop market. Chips developed for the desktop market do have iGPUs.
 

Fjodor2001

Diamond Member
Feb 6, 2010
4,149
559
126
Intel HETD are rebranded server/workstation chips, they are not developing this specifically for the desktop market. Chips developed for the desktop market do have iGPUs.

There is not such fixed definition. The chips are designed for both server, workstation and desktop use. The border between workstation and desktop is not that strict by the way. A workstation is normally just a more powerful desktop computer.
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
There is not such fixed definition. The chips are designed for both server, workstation and desktop use. The border between workstation and desktop is not that strict by the way. A workstation is normally just a more powerful desktop computer.

No,they aren't. For example, you don't need qpi links on desktops and thats's what you have on hedt. The fact that you can use these chips on desktops doesn't mean they were designed for these environments.
 

redzo

Senior member
Nov 21, 2007
547
5
81
First gen bulldozer was a big disappointment and a lot more hype. They ditched single core performance and efficiency for high clock power hungry many slow cores. It would have been a good decision for them only if there wouldn't still be high demand for single core performance.
If the rumors are true, now AMD seems to take a big bet on heterogeneous computing. It is pretty risky if you ask me. History might repeat itself. If the market is flooded with intel non hsa hardware, nobody is going to support it, market adoption will be too slow, and they will end up with the same bulldozer fiasco.
 

Conker10k

Junior Member
Oct 14, 2013
13
0
0
It will be 20 real cores. They got a new top tier engineer from the successful days of the Athlon 64. I'm sure its possible. Windows 9 and Directx12/Mantle will take advantage of the core count better. The future will be multicore. Clock speed is over with. You can't get much faster anymore without generating too much heat. Only way now is to increase the amounts of CPU's aka cores in the system to increase the processing power. IPC optimizations will only get so far. I'm sure we are at a point where the focus now is at core count. They got G34 Opteron 16 core cpus already. I wouldn't be surprised if they upgraded the memory controller to DDR4 and added 4 more cores with the latest tech they could hit at least 3-4ghz. They can already hit 3.2ghz at 115watts right now with 16cores. AMD is making a secret weapon mostly. AMD might come out on top this time :sneaky:
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
Even AMDs and Intels 1P server chips got IGP today.

Only 2P+ server chips are without IGP.

No desktop or laptop chip is developed without IGP anymore.
 

pTmdfx

Member
Feb 23, 2014
85
0
0
Even AMDs and Intels 1P server chips got IGP today.

Only 2P+ server chips are without IGP.

No desktop or laptop chip is developed without IGP anymore.
Technically speaking, there are 1P SKUs of chips without IGP. Yep. So this is just the matter of product binning and positioning.
 

NTMBK

Lifer
Nov 14, 2011
10,421
5,715
136
Technically speaking, there are 1P SKUs of chips without IGP. Yep. So this is just the matter of product binning.

Those are 2P chips with the QPI link fused off, or IGP chips with the IGP fused off.
 

Fjodor2001

Diamond Member
Feb 6, 2010
4,149
559
126
Even AMDs and Intels 1P server chips got IGP today.

Only 2P+ server chips are without IGP.

No desktop or laptop chip is developed without IGP anymore.

So you agree there is no such strict single common definition that says a desktop CPU must have an IGP?

Also, it depends on what you mean by a desktop chip. I consider both FX8350 and 4930K to be desktop/workstation CPUs, and they do not have IGPs.

Sure, there may be parts of the die that probably has most use for server duty. But that is expected if the design shall cover all the server/workstation/desktop segments. I.e. it may be cheaper to have a single common design, even if not all parts of it definetely needed for all use cases. Same as with CPUs with IGP, where the IGP is waste e.g. if paired with a discrete GPU or used as a file server without any display attached.
 

pTmdfx

Member
Feb 23, 2014
85
0
0
Those are 2P chips with the QPI link fused off, or IGP chips with the IGP fused off.
Of course I know, but this doesn't matter much as the SKUs are the real end product. Ah, we have also the new Seattle and its friend Avoton on this list, and they are real man without IGP in this case. The point is that one can't generalize it this way.
 

pTmdfx

Member
Feb 23, 2014
85
0
0
Anyway, Its about K12. So its ARMv8 cores only and not x86.
There is a new x86 core in addition to K12, and even the slide you quoted has a line of words about it. Why would people lie to you when this was the information from AMD? Don't even mention that Anandtech had an short article about "OMG AMD is also developing new x86 core" immediately published after that linked K12 article.
 
Last edited:

pTmdfx

Member
Feb 23, 2014
85
0
0
EDIT: I don't fully understand these rumours, so maybe it is 20 full cores, I'm not sure.
Probably it is because people like weird things, so people made weird rumours to satisfy the demand for those. Just expecting a proper SMT core is fine enough IMO, and it is always better to set your bar of expectation lower - at least not 20 core in the first wave of products. Eww.
 

inf64

Diamond Member
Mar 11, 2011
3,884
4,691
136
There is a new x86 core in addition to K12, and even the slide you quoted has a line of words about it. Why would people lie to you when this was the information from AMD? Don't even mention that Anandtech had an short article about "OMG AMD is also developing new x86 core" immediately published after that linked K12 article.
Indeed there is a new x86 core (ARM K12 "sister core") and it will operate closer to ~4Ghz range according to Keller.
 

pTmdfx

Member
Feb 23, 2014
85
0
0
Indeed there is a new x86 core (ARM K12 "sister core") and it will operate closer to ~4Ghz range according to Keller.
"They will be closer in frequency to the 4GHz of AMD's latest Kaveri x86 SoC than today's 2GHz ARM chips, he said." So mathematically speaking, any numbers greater than 3.0 is closer to 4 than 2, while as a reminder we have BRCM's on-paper Vulcan at the 3 Ghz mark so far. By the way, note that the subject is they, which refers to both new cores by context.
 
Last edited:

redzo

Senior member
Nov 21, 2007
547
5
81
Well, 4Ghz seems a little too high.
From my understanding, high frequency is achieved by sacrificing IPC.
 

SOFTengCOMPelec

Platinum Member
May 9, 2013
2,417
75
91
It will be 20 real cores. They got a new top tier engineer from the successful days of the Athlon 64. I'm sure its possible. Windows 9 and Directx12/Mantle will take advantage of the core count better. The future will be multicore. Clock speed is over with. You can't get much faster anymore without generating too much heat. Only way now is to increase the amounts of CPU's aka cores in the system to increase the processing power. IPC optimizations will only get so far. I'm sure we are at a point where the focus now is at core count. They got G34 Opteron 16 core cpus already. I wouldn't be surprised if they upgraded the memory controller to DDR4 and added 4 more cores with the latest tech they could hit at least 3-4ghz. They can already hit 3.2ghz at 115watts right now with 16cores. AMD is making a secret weapon mostly. AMD might come out on top this time :sneaky:

I really like and agree with what you say, but unfortunately there are two major obstacles to solve first.

In summary the laws of physics (diminishing returns, overly expensive chip plants, ever increasing energy (Electricity) prices, limited practical TDP and customers not willing to pay three times as much, to get 4.6 GHz rather than 4.5 GHz (Exaggeration!) mean that sooner or later, a massive number of (probably) relatively low clocked cpu cores, is the likely future direction, especially in the very long term.

The two (or more) major obstacles are:

(1)..There would need to be massive improvement/updating/innovation/invention with the existing software base, and especially future software, to properly/efficiently/usefully utilise many core processors (of the future).
For various reasons, this could take a long time, and be extremely difficult. It is not simply a matter of latter versions of windows (9), Linux or similar (FreeBsd etc) to improve themselves, as mostly it is the applications themselves that would need to be considerably revised.

tl:dr
We have had this problem for about the last fifty years. (i.e. mainframes, and about 30 years ago the Transputer). There has been some progress, but most programming languages in common use today, have little/poor/incomplete multi-core/multi-thread support.

(2)..It would need to sell in huge quantities (i.e. most/many people/businesses would need to choose 20 core cpu computers over dual/quad cored ones.
So we would need "KILLER" applications.
But by and large, no such applications really exist to any extent yet.

E.g. Gaming is still mostly about how expensive/good a graphics card (or on chip IGP) you get.

There is some stuff like Chess/Video editing and such, which can already use multi-cores/threads with great efficiency, but it is not a major thing for most people.
Even a low cost dual/Quad core cpu, can do a decent enough job of playing Chess or video editing, that the massive extra cost of the current hex cored (socket 2011) stuff (soon to be replaced by Haswell-E[EP]) is just not worth the massive extra cost.

So even if AMD brought out a decent 20 cored (FX/APU) chip, it would be difficult to sell it at a significantly higher cost over a decent dual/quad cpu, with "TODAYS" software. Except the massive server market, which would gobble them all up in a milli-second, if they were competitive with Intel, overall.
(Enthusiasts would still get it, I might even by 2 or more, lol).

Probably it is because people like weird things, so people made weird rumours to satisfy the demand for those. Just expecting a proper SMT core is fine enough IMO, and it is always better to set your bar of expectation lower - at least not 20 core in the first wave of products. Eww.

Yes, I'm a big sucker for any rumour, of new affordable high core count cpus.

I'm still crossing my fingers on someone bringing out, e.g. 32 core Arm Cortex A57, at reasonable cost, available on the mass market.
Maybe AMD might do just that, eventually (as an 8 core one, Opteron A1100, is suppose to be coming out reasonably soon).
 
Last edited:

pTmdfx

Member
Feb 23, 2014
85
0
0
I'm still crossing my fingers on someone bringing out, e.g. 32 core Arm Cortex A57, at reasonable cost, available on the mass market.
Maybe AMD might do just that, eventually (as an 8 core one, Opteron A1100, is suppose to be coming out reasonably soon).
You know, there was a plan of 16-core Seattle. But it was scrapped. God knows why.
 

pTmdfx

Member
Feb 23, 2014
85
0
0
Well, 4Ghz seems a little too high.
From my understanding, high frequency is achieved by sacrificing IPC.
The original source of it (which was an indirect speech quoting Jim Keller) did not mean 4 GHz but closer to 4 Ghz than 2 Ghz. It needn't immediately mean to sacrifice IPC either. Otherwise, any attempt to clock higher in the history should have brought the IPC of any core down to a negative value.
 
Last edited: