NVIDIA Maxwell Speculation Thread

fellix said:
Is it possible that the eventual ARM/Denver core would replace the command processor itself? Is such general architecture enough powerful to act also as a machine-state processor or some dedicated logic must be left on?
Click to expand...

I don't know, but I wouldn't think so, since the Denver-ICs are supposed to work on serial tasks and in concert with the parallel part of the GPU. If it sits outside the SMMs, it cannot efficiently work with them on the same task very good, I would guess.

My current guess (and it's just that, a guess), is that the Denver-stuff (which is probably reserved for the Big Iron) is doing the SFU's work inside each SMM similarly to AMDs scalar unit inside their CUs.
Could be completely wrong, of course.
 
fellix said:
Is it possible that the eventual ARM/Denver core would replace the command processor itself? Is such general architecture enough powerful to act also as a machine-state processor or some dedicated logic must be left on?
Click to expand...

I do not who started the rumor of discrete Maxwell GPUs having an ARM/Denver processor on die but I believe that is wrong.

My belief is the Denver/Maxwell combo is reserved for the Tegra SOC and the HPC/Server products and that the discrete Maxwell for add-in cards will only have Maxwell and no ARM processor.
 
A1xLLcqAgt0qc2RyMz0y said:
I do not who started the rumor of discrete Maxwell GPUs having an ARM/Denver processor on die but I believe that is wrong.
Click to expand...

Maybe someone's interpretation of the VMM in Maxwell? I also cannot remember exactly when or where that got started.

A1xLLcqAgt0qc2RyMz0y said:
My belief is the Denver/Maxwell combo is reserved for the Tegra SOC and the HPC/Server products and that the discrete Maxwell for add-in cards will only have Maxwell and no ARM processor.
Click to expand...

Not sure why they would diverge their silicon now between the markets.
 
If someone thinks that CPU cores integrated on a high end HPC GPU core would be redundant, then of course it's no use debating about it.

As for who started the "rumor"? NVIDIA itself for which Google is your friend for more than one links considering "project echelon" quite some time ago.

Then there were a lot of references from NV employees over time such as that one:

http://www.xbitlabs.com/news/cpu/di...enver_Will_Not_Compete_for_Micro_Servers.html

“Right now we are very focused on high-performance computing. Strategy for Denver, which is a fully-custom ARM core […], is to find a way into Tesla applications. It makes a lot of sense on the server side to put it into Tesla so that to have a differentiation on the HPC market. Today we are not looking to attack the general-purpose server market, that market is looking quite crowded […] it will be a tough place to be. So, we will focus on the market we already know […] and which is a multi-billion dollar opportunity,” said Chris Evenden, director of investor relations Nvidia, at Raymond James IT supply chain conference.
Click to expand...

http://www.xbitlabs.com/news/cpu/di...ave_Integrated_ARM_General_Purpose_Cores.html
 
I think that a CPU is contained if it is GM11x.
If it is GM10x, there will not be the CPU.

There would come to be difference in the function between even kepler generation in GK110 and GK10x.
 
LordEC911 said:
Not sure why they would diverge their silicon now between the markets.
Click to expand...
The Denver core is large so it will take up valuable die area so unless it makes sense it will not be on the silicon.

It makes sense for the HPC/Server and Tegra market but not for the discrete add-in market.

The HPC/Server and Tegra will only have the Denver ARM as a CPU whereas the discrete market will have other CPUs (Intel/AMD/IBM) as the main CPU so the Denver core would be wasted on those systems.
 
A1xLLcqAgt0qc2RyMz0y said:
I do not who started the rumor of discrete Maxwell GPUs having an ARM/Denver processor on die but I believe that is wrong.

My belief is the Denver/Maxwell combo is reserved for the Tegra SOC and the HPC/Server products and that the discrete Maxwell for add-in cards will only have Maxwell and no ARM processor.
Click to expand...

I think this interview, particularly:

"As well as licensing Cortex A15, we also have an architectural license with ARM to produce an extremely high performance ARM CPU, which be combined with NVIDA GPUs for super-computing," he said. When we asked for timescales, Rayfield revealed: "The Maxwell generation will be the first end-product using Project Denver. This is a far greater resource investment for us than just licensing a design."

is the origin of the Denver / Maxwell association. There's certainly nothing in there that definitely indicates to me that Maxwell will have general purpose cores on die, package, or board of a dGPU. Many sites seem to have interpreted it that way, e.g.:

NVIDIA Maxwell to be first GPU with ARM CPU in 2013 - Guru3D.com

Nvidia Maxwell Graphics Processors to Have Integrated ARM General-Purpose Cores. - Xbitlabs.com

which xbitlabs also contains this quote from JHH in 2011:

"Between now and Maxwell, we will introduce virtual memory, pre-emption, enhance the ability of GPU to autonomously process, so that it's non-blocking of the CPU, not waiting for the CPU, relies less on the transfer overheads that we see today. These will take GPU computing to the next level, along with a very large speed up in performance."

Which I don't think strengthens the case for Maxwell dGPUs integrating arm cores, because he specifically says, "between now and Maxwell", as in that this has apparently already happened, and it didn't require arm.
 
If the size per one core of denver is an equivalent for two cores of A15, the size is about 10 mm2.
Is it large? is it small?
Are they every GPU, every ? GPC, and every ? SMX?
 
iMacmatician said:
Did they skip CC 4.0?
Click to expand...

"You know, we were going to call it CC 4.0, but Maxwell's compute capabilities are just so much better, the performance improvements it enables with complex algorithms are just so amazing that it didn't feel right to call it that. So we went with CC 5.0 instead, which gives you a much better idea of what Maxwell can really do."

—Jen-Hsun Huang.
 
Alexko said:
"You know, we were going to call it CC 4.0, but Maxwell's compute capabilities are just so much better, the performance improvements it enables with complex algorithms are just so amazing that it didn't feel right to call it that. So we went with CC 5.0 instead, which gives you a much better idea of what Maxwell can really do."

—Jen-Hsun Huang.
Click to expand...

Compute Capability 5.0, only on the GeForce 8*10^9.
 
ams said:
The Kepler.M GPU in Tegra K1 supports CUDA 5...
Click to expand...

Kepler.M fully supports CUDA 6, as do all Kepler GPUs.

Computer Capability level is related to the hardware capabilities of an Nvidia GPU, and is separate from the CUDA software platform levels.
 
You must log in or register to reply here.

Similar threads

P
Speculation and Rumors: Nvidia Blackwell ...
16 17 18
Replies
351
Views
36K
troyan
T
Rys
  • Sticky
Hello again
3 4 5
Replies
91
Views
10K
homerdog
homerdog
M
AMD made a mistake letting NVIDIA drive Ray Tracing APIs
5 6 7
Replies
121
Views
15K
MfA
M
xpea
NVidia Hopper Speculation, Rumours and Discussion
7 8 9
Replies
168
Views
32K
pharma
P
N
Hellslave (Saturn Homebrew)
Replies
6
Views
3K
Nesh
N
Back
Top