Exascale Heterogeneous Processor. 32 x86 Zen cores. 32GB HBM.

The thread and title is misleading.
AMD didn't "reveal" anything.

A researcher made a theoretical paper.

That is it.
No roadmap.
No announcement.
No PR.
Just a rumor-site...feeding a starving "population" of AMD-fans, that seem desperate for any kind of news.

But watch the AMD-PR-SPIN machine and forum warroris go into action...this is hillarious. (and so "#¤#¤ing sad at the same time)...
 
Darakian said:
In all fairness the researches do work at AMD.
Click to expand...

Dosn't alter the facts...it's already (in what I would call PR FUD manner) being tried to lied into "official" AMD talk.
Look at the title of the OP posts.
Look at the content.

Go ahead...set yourself up for a another dissapointment.
Of face the facts...this is worse than PR...this is FUD PR.
 
I've read the paper and if you read my post above, you'll see I'm not expecting a 32 core APU with a monster GPU. That said, the paper begins with

Achieving Exascale Capabilities through Heterogeneous Computing
Michael J. Schulte, Mike Ignatowski, Gabriel H. Loh, Bradford M. Beckmann, William C. Brantley,
Sudhanva Gurumurthi, Nuwan Jayasena, Indrani Paul, Steven K. Reinhardt, Gregory Rodgers
AMD Research
Advanced Micro Devices, Inc.
Click to expand...

I suggest you read the paper and the thread before you assume that people read anything out of it.
 
Factum said:
Dosn't alter the facts...it's already (in what I would call PR FUD manner) being tried to lied into "official" AMD talk.
Look at the title of the OP posts.
Look at the content.

Go ahead...set yourself up for a another dissapointment.
Of face the facts...this is worse than PR...this is FUD PR.
Click to expand...

In all fairness it does not sound anything like FUD it is just something that AMD thinks would be plausible in say next 5 years or so ?

HBM was also in the process of being developed several years ago but if you asked 5 years ago you could have said the same thing and now there is HBM.

That the practical implementation of this project is prolly not consumer based is trivial.
 
Anyone else who needs to try and divert from the facts with wish-full thinking?

This is NOT something AMD revealed...this is worse than some posters who search LinkedIN high and low for little clues to their confirmation bias.
Shall we take a look at the failure rate of research papers...in this field? ^^

Or do more people need to come on a sugarcoat/remove focus from that this thread misrepresents the facts (read ¤#%¤#%¤#% rumors)?!
 
Man, wouldn't it be great if AMD would die, like Factum wishes, so Intel and Nvidia could become the hardware developer equivalent of Comcast! I would love to have companies in a monopoly standing so they can offer fair prices on cutting-edge hardware!
Oh wait, no competition breeds items that are barely better than their predecessors, at 200%+ prices... (case and point any Comcast service package)
Good call Factum!!
 
People there is an hpc apu on amd's 2016- 2017 roadmap this is jsut a rough out of what such a monster would look like to give you the idea of where that future hpc apu is heading is the graphic of it accurate some parts yes some no but its not meant to be that its to generate buzz and interest when all amd has for us in 2015 is to give us a sliver of hope.
 
Factum said:
The thread and title is misleading.
AMD didn't "reveal" anything.

A researcher made a theoretical paper.

That is it.
No roadmap.
No announcement.
No PR.
Just a rumor-site...feeding a starving "population" of AMD-fans, that seem desperate for any kind of news.

But watch the AMD-PR-SPIN machine and forum warroris go into action...this is hillarious. (and so "#¤#¤ing sad at the same time)...
Click to expand...

The nerve of us, to be talking about future amd processors on a section of a hardware forum entitled amd processors!

I think we owe you an apology.
 
It'd be amazing if something like this happened.. With DX12 paired with this, I'd love to see how this would run games haha. One can dream.
 
All of us know AMD put themselves in the IPC shitter with that ridiculous module core methodology. A shitter they have yet to craw out of. However there is no reason to believe they can never be competitive again. We all need AMD to be in the fight. As others have said on this thread a one player market would be bad for us all.

In all reality other than for gaming AMD is just fine. The average person would never tell the difference. I still use AMD for nearly all my non-gaming builds.
 
If the Zen core 40% boost is real, along with AMD's high frequency designs, it's going to be good times for everyone. Real competition leads to great pricing for us.
 
Government contracts are worth their weight in gold to a company. I cant blame them for vying hard for it.
 
Hey, If AMD can snag a gov't contract and get some stable income, We could be looking at an ACTUALLY competitive industry*. We may see Intel finally give a damn about making an improvement to their desktop CPUs.


*That's taking into account the unlikely event that AMD management gets its head out of its own ass and starts managing the company like adults.
 
Government contract are worth not even in gold but in diamonds but don't you think AMD went over the top by announcing the exascale processor while the 2025 is realistic exascale goal?

Comparing this processor to the soon coming Intel Knights Landing chips with up to 72 cores and 3 teraflops FP64 general processing power is not favorable to AMD.
 
OFaceSIG said:
All of us know AMD put themselves in the IPC shitter with that ridiculous module core methodology. A shitter they have yet to craw out of. However there is no reason to believe they can never be competitive again. We all need AMD to be in the fight. As others have said on this thread a one player market would be bad for us all.

In all reality other than for gaming AMD is just fine. The average person would never tell the difference. I still use AMD for nearly all my non-gaming builds.
Click to expand...

Not sure where this is coming from but IPC? Who cares on this kind of machine you think that IPC will be the deciding factor in a user buying this. If anything this would be bought who need computing power of the GPU.
 
Oniigumo said:
It'd be amazing if something like this happened.. With DX12 paired with this, I'd love to see how this would run games haha. One can dream.
Click to expand...

Probably like shit, considering the low clocks that would be necessary to accommodate so many cores within a reasonable power budget. Server/HPC CPU does not equal a good gaming CPU. Just ask anyone with a 2P system how it runs games. They will tell you it will bench slower than a 1P system with 4-8 cores and higher clocks. Titles are just now starting to reach into more than 4 threads.
 
To be fair he said DX12 he expects that to make it work a lot better. Then again DX12 does not mean that by default all of the cores get used using multiple cores is an engine based functionality that is programmed by game developers.
 
Factum said:
Dosn't alter the facts...it's already (in what I would call PR FUD manner) being tried to lied into "official" AMD talk.
Look at the title of the OP posts.
Look at the content.

Go ahead...set yourself up for a another dissapointment.
Of face the facts...this is worse than PR...this is FUD PR.
Click to expand...

Whilst I agree with the premise of your post (the widespread reporting of this as if its an actual product being misleading), please stop misusing the term "FUD PR". Nothing about this is literally spreading fear, uncertainty, or doubt in any shape or fashion. Actual FUD PR is what 95% of these shit-tier "tech sites" publish on a daily basis, yet most people either eat it up or don't notice it because it's so subtle. Like how next to no one is giving Nvidia a hard time for their laughably bad Win 10 driver situation yet will be the first ones to talk shit on AMD if they were the ones in the position. The fact that Nvidia has been having more driver problems than AMD has for quite some time now and the popular "fact" amongst forum-goers is still that AMD has "bad drivers" says a lot.

Don't get me started on the scumbaggish lies and other shit Nvidia still gets away with because they've bought out most of the sites. Intel is hardly any better with their blatant lies, like how they tried to say "Broadwell can clock to 5ghz on air" at one point but the reality is that it's hard-pressed to clock beyond 4.2ghz. Not trying to make AMD out to be saints here, but the double standards in the forums and "tech" sites are at such an all-time low that it's honestly embarrassing to read/witness a lot of the time now.
 
Factum said:
A researcher made a theoretical paper.
Click to expand...
This looks like a potential product that fits in with the custom silicon division. I imagine this is just an idea being floated out to see if there's any interest, and may be an attempt to leverage a few things AMD has leftover from SeaMicro like fabric connect technology. Intel has done the same thing with its custom silicon strategy.

I agree it's nothing to get immediately excited about.
 
AMD is doing a HPC APU which is projected for 2017, but it's nothing on the scale of what people are purporting right now. Exascale products probably won't be here until around 2020 or even beyond that, and even then it's unlikely those designs will scale down 100% to client.
 
I am a huge fan of AMD, so let's just get that out of the way. I love their design philosophy, I love their market segmentation philosophy, and I love their pricing.

BUT, you have to wonder when you see this shit coming out of AMD's research labs. OK, so they are dropping CMT and finally accepting SMT with some form of HT (probably) is really the way to go. Fine. But why, oh WHY are they insisting on bumping up the core count and on-die GPU for ALL of their products in the future? Don't they realize that, even as gaming and a bunch of other stuff becomes increasingly multithreaded, there will always be a small number of programs and users that really just need a fast single thread? Don't they think they can compete on price/perf with Intel there, if not on actual raw performance? Don't they even want to try?

One great little chip they released on the FM2+ platform was pretty telling - you take an Athlon X2 370K, which runs at 4.2ghz, and it's a ~$50 chip on the Piledriver core that can handle most modern games admirably for an entry-level build, or do a credible job at other single-thread tasks, even if it does look a lot less shiny next to the Pentium G3258 (theoretically anyway; no one bothered to actually bench the two chips together for some stupid reason).

Getting a 16 core Zen chip is awesome - but we already know AMD caters to large core count users. What about a budget Zen quad core that's unlocked, sits on a 5ghz base, and maybe has an R9 GPU w/ 2gb HBM sitting on it? That's the sort of thing that would make the market cream its pants for AMD.
 
Where did you get this 16-core Zen chip from? The Zen that will be on AM4 next year is 8 core, 16 thread, because of 2-way SMT. The 16 core parts will be Opterons that are MCM, like in the past.
 
NaroonGTX said:
Where did you get this 16-core Zen chip from? The Zen that will be on AM4 next year is 8 core, 16 thread, because of 2-way SMT. The 16 core parts will be Opterons that are MCM, like in the past.
Click to expand...

My mistake, but the point still stands, just to a lesser degree.
 
Not really.

Don't they realize that, even as gaming and a bunch of other stuff becomes increasingly multithreaded, there will always be a small number of programs and users that really just need a fast single thread? Don't they think they can compete on price/perf with Intel there, if not on actual raw performance? Don't they even want to try?
Click to expand...

AMD has already stated that the focus of Zen from day-one was on per-core performance, i.e. single-threaded perf, because they know the Bulldozer project didn't work out so well. Top-end Zen on AM4 will be a CPU-only part which is up to 8 cores, 16-threads for the highest bin. Then in 2017 there will be the APU variants which will use an undisclosed number of Zen cores with the newer GCN 2.0 graphics IP blocks and quite possibly on-package HBM2.

They're not focusing purely on multi-threaded performance with Zen. Zen is already finished (it's taped out a while ago) and they're testing it now. The new GCN 2.0 GPU's are also finished and they've taped and are being tested. The only thing we don't know is what the final clocks will be on the Zen parts and what the performance will be. I've heard whispers from insiders that the single-threaded performance will be really good, but of course I've no way of verifying it or proving it myself.

So I guess we'll just have to wait and see until then, lol.
 
NaroonGTX said:
Not really.

AMD has already stated that the focus of Zen from day-one was on per-core performance, i.e. single-threaded perf, because they know the Bulldozer project didn't work out so well. Top-end Zen on AM4 will be a CPU-only part which is up to 8 cores, 16-threads for the highest bin. Then in 2017 there will be the APU variants which will use an undisclosed number of Zen cores with the newer GCN 2.0 graphics IP blocks and quite possibly on-package HBM2.

They're not focusing purely on multi-threaded performance with Zen. Zen is already finished (it's taped out a while ago) and they're testing it now. The new GCN 2.0 GPU's are also finished and they've taped and are being tested. The only thing we don't know is what the final clocks will be on the Zen parts and what the performance will be. I've heard whispers from insiders that the single-threaded performance will be really good, but of course I've no way of verifying it or proving it myself.

So I guess we'll just have to wait and see until then, lol.
Click to expand...

First of all, this stuff about a CPU-only part in 2016 on the AM4 socket is news to me! Just did the required reading... somehow I had missed that they will not all be APUs. I can't tell you how much of a relief that is to me.

Eight cores is a configuration that is finally coming into its own; I have doubts about uses in the consumer space for anything higher than that, but I'm glad to see that the worst parts of my fears are basically invalid.

There are heat and transmigration issues on the 22nm node from Intel when it comes to overclocking; do you think there will be similar issues on 14 and/or 16nm finFET nodes from AMD?
 
From what I currently understand, AMD's highest TDP given for the AM4 socket is 95W, and that's for the top-end 8 core parts, so I guess they're pretty confident on the heat levels and power consumption. In terms of overclocking, I have no idea how well the 14nm LPP/LPE nodes would handle that yet.
 
As always, AMD stays hot at 95W on a tiny node, lol. I hope they have a good TIM between the chip and its lid...
 
TheEschaton said:
As always, AMD stays hot at 95W on a tiny node, lol. I hope they have a good TIM between the chip and its lid...
Click to expand...

Ahh yes cause Intel parts run so cool, that they recommend turning off the integrated gpu to be able to overclock it. Also if you knew anything about AMD you would know they dont use TIM they solder their heat spreaders on. 95 watts on a 8 core part is pretty darn good.
 
Gideon said:
Ahh yes cause Intel parts run so cool, that they recommend turning off the integrated gpu to be able to overclock it. Also if you knew anything about AMD you would know they dont use TIM they solder their heat spreaders on. 95 watts on a 8 core part is pretty darn good.
Click to expand...

So quick to bite, you didn't have enough time to consider what you were saying, perhaps?

Intel chips run on less power, and they are as hot as they are because they are running on a smaller node. So my implication is that 95W on a small node is likely to be pretty hot.

Also, the many of the most recent AMD chip designs use TIM under the lid: http://www.techpowerup.com/197051/amd-a10-7700k-kaveri-de-lidded.html

In fact the only two recent AMD CPU designs I know of that do not have TIM under the lid would be the A10-7870K, which is verified soldered, and the resultant Athlon X4 870K part, which I haven't seen out yet.

In the most recent offering from their FX line, AMD rebranded the FX-8320 to the FX-8320E and the FX-8370 to the FX-8370E. Both of these chips are eight core chips offering the most recent turbo technologies (allowing them to turbo up to or above the original clock speeds of their parent chips), and which run eight cores with a 95W TDP, down from 125W TDP. Presumably these are soldered as well.

If AMD follows suit from one generation of FX to the next, then these chips will have solder under the lid. If, however, they instead follow the example of their APU cousins, they will have TIM, and the CPU-only parts will simply be the APU parts with the GPU die disabled - which would also make sense, given that both the APUs and the CPUs will be running off the same socket.

The jury is still out on that, but they will almost certainly be running hotter than the i7s they (hopefully) will be competing with either way.
 
APU is not the same as a CPU, so for now I have no reason to believe they will stop soldering the heat spreader on a CPU. ZEN as far as I know has a cpu only build for the enthusiast market, so no disabled gpu core, unless I missed something. Either way I wouldn't mind if they left the disabled gpu on, will add more surface area to allow for more heat transfer for cooling. Also Intels current cpu is only a 4 cores this is eight, so at 95 watts thats pretty darn good. Guess will find out one way or another next year on whether it runs hotter or not.
 
TDP =/= power consumption or even how hot the cores can get. The quad-core Llano that I have was rated at like, 100W and CPU-Z showed 112W for some reason, but the cores only get up to about 30*C at full load, lol. The chip always ran cold as ice, and that's without a crazy cooler.

So we don't know anything about how Zen will do in terms of heat and power consumption, so it's too early to jump the gun based solely on the TDP rating given. 95W for an octocore with SMT which will presumably be clocked fairly high out the box (>3Ghz) is pretty damned good so far.
 
You must log in or register to reply here.
Back
Top