32 Zen Core Opty Details (rumor)

Seems like this needs to be posted again:


7180aRPw3xM_700b.jpg



But seriously as someone pointed out in the comments on the thread, sticking on twice as many cores as Intel uses on their high end Xeon's means that the cores will be comparatively tiny, without the necessary circuitry for advanced branch prediction, etc... that give Intel cores their massive IPC advantage, meaning the IPC will suck on these. Which will be fine for many server applications, SQL, web serving, many VM applications, but on the desktop side would be like sticking a cut down Xeon Phi in the CPU socket, i.e. it would suck ass.


But seriously it sounds like AMD is trying to copy the Phi but instead of an addon board they're just going to stick it straight in the CPU socket. I'm not getting a particularly good feeling on Zen-FX...
 
^I don't get how you came to that conclusion. The die-size on the Intel Xeon's with lots of cores are pretty damned big, and this hypothetical APU (this FUDzilla article is a rumor made up by Fuad himself) is targeted at the HPC sector and would have a higher TDP. This would NOT be a consumer chip, so the whole "durrr moar coarz loh eye pee see!!1!" nonsense wouldn't even apply here.

Lisa Su has stated on more than one occasion that they're aware that their partners and customers want (i.e. they've been requesting) high-performance successors to the outdated Piledriver Opterons that came out three years ago. I don't think AMD would've gone through the trouble to poach people like Jim Keller and Raja Koduri just to make mediocre products. AMD needs their R9-300 series to be beast and they need K12 and Zen to be great products to gain back the marketshare they've lost over the past few years, to bring profits up and to try to score more design wins across the board with this new IP.

Besides, a 16 core, 32 thread chip wouldn't even be "twice as many" as a high-end Xeon -- actual high-end Xeons have 18 cores and 36 threads and a die size of roughly 662 mm^2.
 
Because the article stated 32C/64T not 16C/32T, that last part was just what the OP wanted, 32/64 is close enough to double 18/36
 
But seriously as someone pointed out in the comments on the thread, sticking on twice as many cores as Intel uses on their high end Xeon's means that the cores will be comparatively tiny, without the necessary circuitry for advanced branch prediction, etc... that give Intel cores their massive IPC advantage, meaning the IPC will suck on these. Which will be fine for many server applications, SQL, web serving, many VM applications, but on the desktop side would be like sticking a cut down Xeon Phi in the CPU socket, i.e. it would suck ass.

FYI, the new top-end Xeon E5 CPU is going to have 18 cores, so everything you just stated is pretty much bunk. :/
The problem with AMD wasn't "moar cores", it was the fact they used a CMT design, and banked on the probability that most desktop applications and programs would be multi-threaded by this point, which to be fair, most are still single-threaded.

There is nothing wrong with more cores, assuming the architecture is designed right (CMT was a poor implementation, imo).
Intel's CPU core performance ratio is 2:1 compared to AMD, so you can do the math there, not even counting IPC.
 
Because the article stated 32C/64T not 16C/32T, that last part was just what the OP wanted, 32/64 is close enough to double 18/36

Damn, I thought this was a repost of the earlier rumor that Fuad posted on his site.

As for this, who knows, I still won't believe anything until AMD themselves or at least a reputable source leaks/reveals it. A 32c/64t processor would have a huge die and would never in a million years come to the consumer market, and it wouldn't be enough information to assume that the cores themselves would be small/narrow or what have you.
 
Fudzilla said:
Just like the 16 Zen core high performance market APU, each core has 512KB of L2 cache and four processors share 8MB L3 cache. The highest end part will come with eight clusters of 4 cores and if you do the math this server oriented CPU will come with 64GB of L2 cache and 16MB of L2 cache for its CPU cores.

I'm starting to think that people over there don't proofread...
 
Damn, I thought this was a repost of the earlier rumor that Fuad posted on his site.

As for this, who knows, I still won't believe anything until AMD themselves or at least a reputable source leaks/reveals it. A 32c/64t processor would have a huge die and would never in a million years come to the consumer market, and it wouldn't be enough information to assume that the cores themselves would be small/narrow or what have you.

Or in other terms they must have had some major major major breakthroughs on thermal design of the chip. There are some things floating around on different forums about the size of the cores if they are really small in a different segment (Bulldozer core vs Jaguar core) this could make sense. Then again what is the outlook on such a product on the x86 server market
 
Can AMD stop sharing resources and dedicate it, perhaps then it can compete with intel...
 
and still sucks

WOW! I wish most things that "sucked" performed as extremely well as my AMD FX 8350 and FX 8320 do. :) Oh, and there is my XFX HD 6770, XFX R9 290, Asrock 990FX Extreme 9 and Extreme 4 boards that also perform extremely well.
 
The HMB will likely give AMD a game changer.
16 cores, that will change things with DX12, next year.
 
The HMB will likely give AMD a game changer.
16 cores, that will change things with DX12, next year.

This has honestly been the problem for a while now. It is not just DX12, but most applications in general, even the operating system. Multi-threaded workloads should be more prevalent in the market than they are by this point, but the software simply hasn't caught up, especially getting beyond 2 - 4 threads, hence why you start to see a rapid decrease in the rate of performance increase, even with Intel CPUs, that have the ability to process 4 or more cores/threads simultaneously.
 
The HMB will likely give AMD a game changer.
16 cores, that will change things with DX12, next year.

I agree but, since I have 2 computers I want to upgrade, I think only two 8 core processors will fit into my budget. :) However, I would love me so 16 core goodness but my bank will so, "HEY NO!" :D
 
What you said does not make sense. :confused: AMD does not share resources, everything stays in the company.

He's referring to their chip design. Bulldozer had 'modules' where every 2 integer cores shared an FPU and other components.

I think his opinion is misguided. The shared components dont seem to impact performance very much, running an FX chip in one-core-per-module mode does not substantially improve performance. The issue is that the integer cores themselves are slow.
 
He's referring to their chip design. Bulldozer had 'modules' where every 2 integer cores shared an FPU and other components.

I think his opinion is misguided. The shared components dont seem to impact performance very much, running an FX chip in one-core-per-module mode does not substantially improve performance. The issue is that the integer cores themselves are slow.

I believe the problem not so much about the shared components but more about the reduced # of processing units for each core. AMD took only 125% of the transistors of a Phenom II core to make a bulldozer module.
 
Back
Top