AMD Talks Next Generation Coherent Interconnect Fabric Connecting Polaris GPUs, Zen CPUs and HPC APUs (AMDs NVlink)

[Mr_Troll]

AMD Developing Ultra-Wideband, Low Latency Coherent Interconnect Fabric To Rival NVIDIA’s NVLINK

 

Since the creation of RTG (Radeon Technologies Group), Raja Koduri has put a lot of focus back to GPUs. In 2012, AMD’s strategy was quite different to what it is today. The company focused on their APU (accelerated processing units) and SOC (System-on-chip) designs however that didn’t work well for the company. The GPU department on the other hand has been a main driving force behind for them and that’s where both AMD and RTG is specifically eyeing at after the restructuring of the company in September 2015.

 

According to Raja, there’s a constant rise in compute demand for the high-end PC market. Specifically speaking about compute performance where the GPU clearly shines against modern CPUs, there’s just constant demand for greater computing needs in the high-performance computing sector which consists of HPC, Servers and the Workstation machines. To feed that demand, AMD is coming back with a big bang in the GPU department with their Polaris GPU architecture but their aim doesn’t stop at GPUs alone.

 

Even with rumors going on around that discrete GPUs won’t stay in the market in the long term, the statement is simply dismissed by Raja who further says that the market for dGPU will always be on the rise as compute demand will never decrease. Raja explains that the market has two kinds of users, those who don’t demand the compute performance are the casual and entry level audience who either buy low-end PCs or smartphones and high-end users who eye better performance and compute. The high-end sector consists of users more than just gamers and like stated, the HPC, Server and Workstation market is also part of this sector who keep on demand for better compute and performance.

 

 

 

Both AMD and NVIDIA face a similar problem in the high-end sector which is to feed the high compute demand of their customers. According to AMD, not even their Radeon R9 Fury (Fiji based) graphics cards are powerful enough alone to meet those demands but by offering more scalability and stability, allowing multiple GPUs to run in parallel to one another to solve a problem, the necessary performance could be achieved.

In a perfect situation, all GPUs inside a workstation can be made to act as a single resource that aims to solve a specific set of problems. Currently, AMD only provides a software based multi-GPU framework added in their recent Boltzmann Initiative but moving forward, AMD knows that an update to their multi-GPU solution, also known as CrossFire, is needed to provide a hardware based multi-GPU solution that aims to deliver a robust scaling across AMD’s GPUs. For this matter, they are developing their own open Interconnect fabric that will not only allow AMD GPUs to run in tandem with one another but also allow third-party solutions (GPUs, FPGAs).

 

 

AMD’s Coherent Fabric To Clock in 100 GB/s Interconnect Speeds – Open Nature Design

 

PCI-Express is already seen as a bottleneck when connecting several nodes in high-performance sectors. AMD sees their current PCI-e and CrossFire solutions not working with next generation machines hence they have to design a new coherent fabric. The interconnect will offer speeds of 100 GB/s across multiple GPUs and APUs that are featured inside AMD powered compute machines and will deploy some open standards. Asking if the interconnect will also maintain memory coherency and sharing between the GPUs and CPUs, Raja stated that he can’t reveal that right now but will definitely have a detailed showcase of their coherent fabric later on as coherency between their several chip designs is being kept in mind.

 

 

Comparing to Nvidias NVLINK which is going to have interconnect speeds of up to 200 GB/s, the AMD solution does have a slight advantage as it works across x86 processors as well while NVIDIA’s NVLINK works across the NVIDIA based GPUs and IBM’s Power CPUs. NVIDIA is already deploying NVLINK in two next generation super computer so it will be interesting to see AMD’s own solution going in action to power some high-end spectrum devices.

 

AMD’s Latest Packaging and Integration To Help Development of HPC APUs

 

The letter “P” is very dear to Raja as it means four key components for RTG while designing next gen GPUs. The four “Ps” include Performance, Power, Price and last but not least Packaging. AMD is the firs graphics vendor to ship an HBM powered graphics card and they have got some experience from it in the packaging department. Tight integration of the GPU and HBM silicon on the same substrate (interposer) leads to some crucial learning that helps designing next generation solutions for compute hungry audiences.

 

 

 

For some time, we have been hearing about HPC APUs which are simply put, high-end APUs that will come with a fast discrete graphics chip, several next-gen x86 cores and tons of HBM memory, all integrated on a package and all chips linked via the fast interconnect which has been talked about in this article. We have previously reported on extensive details regarding the High Performance Computing APU and Heterogeneus Processors from AMD. If all goes well, we will see an update in this regards when the Zen processors hit the market which is probably due for release at the end of this year.

 

 

 

Shame its probably not for consumers. 

 

Source:http://pc.watch.impress.co.jp/docs/column/kaigai/20160115_739108.html

http://wccftech.com/amd-coherent-interconnect-fabric-gpus-cpus-apus/

[SurvivorNVL]

That.. doesn't look bad, actually.  That's.. huh.  Well, I'm deeply interested.  Okay, AMD.  Wow me!

[Guest]

Looks like the Red Team is making a comeback!

[SurvivorNVL]

Maybe.  If they can make this more appeasing to IBM than NVlink.  IBM can buy AMD!  Wow.  What a team that would be.  Terrifying, even.

[Doobeedoo]

Good to see progress on AMD side this year.

[Trixanity]

Oh dear. I expect countless threads complaining (or asking) about having to buy new hardware in order to use new amd graphics cards just like with Pascal and nvlink.

It's great to see amd providing an open alternative to nvlink but I can't help but wonder at the gap in bandwidth, why it exists and what effect it will have both in performance and in adoption.

[TechGod]

AMD has been on a roll lately. GPUOpen, Zen coming out, Polaris with the demo regarding its efficiency, pushing Mantle and then giving Mantle to the Khronos group to use as the base for Vulkan and now this. Great job!

[azurezeed]

Those hpc apus could be used for next generation vr enable consoles, though I'm sure that could be expensive

[Master Disaster]

I honestly believe it's to late for AMD to make any significant impact on the prosumer and enterprise market places.

Intel are so well established, especially in the enterprise market, that I just don't see large data centres and server farms ever completey ditching everything and swapping over to an entirely new architecture.

At this point AMD would have to do something completely unbelievable to ever take that share away from Intel.

That said though, this actually sounds very interesting. I think were all hoping that Zen at least competes with Intel. No one is expecting them to come out and top the charts but if they come and are getting destroyed then AMD MIGHT end up taking a different path in the future.

[AlwaysFSX]

Oh but I thought something like NVLink was stupid. <_<

 

Hopefully AMD can keep on doing things to increase performance, this is yet another step in that direction.

[patrickjp93]

That.. doesn't look bad, actually. That's.. huh. Well, I'm deeply interested. Okay, AMD. Wow me!

It'll be a year late to the party and a step behind. Intel's 100GB/s Omnipath 100GB/s interconnect is already out, and NVLink 2.0 is a 200GB/s interconnect coming out with Volta. By the time it comes out at the end of 2016, it'll be soon ready for retirement.

[SurvivorNVL]

It'll be a year late to the party and a step behind. Intel's 100GB/s Omnipath 100GB/s interconnect is already out, and NVLink 2.0 is a 200GB/s interconnect coming out with Volta. By the time it comes out at the end of 2016, it'll be soon ready for retirement.

I didn't know that.  I guess it still goes back to the same point as before.  The best option would be their business going up in flames, and Intel / nVidia cannibalizing their assets and then going to kill one another in competition. 

[don_svetlio]

I didn't know that.  I guess it still goes back to the same point as before.  The best option would be their business going up in flames, and Intel / nVidia cannibalizing their assets and then going to kill one another in competition. 

Don't be quick to think that. Patrick is, after all, the forum's expert "Crystal Ball"

[SurvivorNVL]

Don't be quick to think that. Patrick is, after all, the forum's expert "Crystal Ball"

I've thought it for awhile.  It is one of those things where it would probably be the quickest path to us getting back to actual competition.  nVidia has a great deal of revenue and capital flowing in to them, aside from lawsuits hampering that every now and again.  Intel is a monstrosity of assets, wealth, and capability.  The two of them with AMD's assets going head to head would be nothing but good for us.  Intel's pretty good on the open source end, and nVidia is--nVidia.

[Trixanity]

I really don't think nvidia and intel devouring amd is a good scenario. A third party providing a platform for amd to thrive is, in my mind, a better scenario for all, although the scenario I envision is pretty unrealistic because nobody would provide that without steering amd with an iron fist. Why? It's all about the money.

[don_svetlio]

I've thought it for awhile.  It is one of those things where it would probably be the quickest path to us getting back to actual competition.  nVidia has a great deal of revenue and capital flowing in to them, aside from lawsuits hampering that every now and again.  Intel is a monstrosity of assets, wealth, and capability.  The two of them with AMD's assets going head to head would be nothing but good for us.  Intel's pretty good on the open source end, and nVidia is--nVidia.

I for one would prefer to at least wait and see Zen first.

[patrickjp93]

I for one would prefer to at least wait and see Zen first.

So would we all. I just don't see it being the miracle everyone expects. I don't suspect Arctic Islands will be a saving grace either.

[don_svetlio]

So would we all. I just don't see it being the miracle everyone expects. I don't suspect Arctic Islands will be a saving grace either.

What about Athlon 64?

[patrickjp93]

What about Athlon 64?

Back in those days performance gains were easy to attain. Nowadays instruction latencies are down near their theoretical lows on Intel's side, and there's only so deep you can realistically make the execution pipeline (since iirc statistically every 7 lines of code is a branch/control hazard) to improve performance in the bulk of cases. And you can only make it so wide to take advantage of data and instruction parallelism since most applications are not programmed well enough to make good use of those features. Further, Intel has plenty of patents on vector ALU designs. AMD has an uphill battle to match AVX 256/512 performance to matter in the datacenter. Further Skylake E/EP/EX is moving to 6-channel memory whereas AMD designed AM4 as quad-channel and will be using it for everything from a quad-core A/C-PU up to the 32-core Zen flagship. It's much tougher to beat Intel when it's this far ahead and every performance gain is quadratically more difficult to obtain after the previous and much more engineering-intensive. It's why ARM is still far behind x86 as well even to this day. It's easy to improve when you're still way back there at circa-2004 performance. Nowadays, it's not as easy, and Jim Keller is no silver bullet.

[zMeul]

So would we all. I just don't see it being the miracle everyone expects. I don't suspect Arctic Islands will be a saving grace either.

I guess you about about the recent video where AMD talks trash about BAPCo

wondered why exactly is AMD doing this now, their CPUs aren't exactly new; then it hit me .. it's about Zen  <_<

[TechGod]

I guess you about about the recent video where AMD talks trash about BAPCo

wondered why exactly is AMD doing this now, their CPUs aren't exactly new; then it hit me .. it's about Zen <_<

Ah yes. The resident Intel fanboy troll.

[marldorthegreat]

Ah yes. The resident Intel fanboy troll.

But he has a point, there complaining about bias the year there new architecture is launching.

[marldorthegreat]

Maybe.  If they can make this more appeasing to IBM than NVlink.  IBM can buy AMD!  Wow.  What a team that would be.  Terrifying, even.

AMD will only be bought if it goes under. And probably split, intel might buy the gpu division, and nvidia the cpu division

[cesrai]

Ah yes. The resident Intel fanboy troll.

What a great counter argument would use every time /s .

[TechGod]

But he has a point, there complaining about bias the year there new architecture is launching.

And why should they not? This product is pretty much their last chance and if there is bias that shows their "do or die" chip in an unfavourable light, they're screwed.