Is 4GB enough for a high-end GPU in 2015?

ToTTenTranz said:
And I guess DP performance is unlocked to the chip's 2:1 full capability.
Click to expand...

The S9170 is specially a DP compute part.. 5.24 Tflops SP / 2.62 Tflops DP for DGEMM.

The AMD FirePro S9170 offers unparalleled GPU compute performance and power efficiency with the following benefits:

  • With up to 2.62 TFLOPS of peak double precision performance, the AMD FirePro S9170 is the fastest single-GPU server card available for DGEMM heavy workloads, delivering up to 40% more performance than the competitive solution;
  • Support for 40% better double precision performance, while using 10% less power than the competition;
  • The AMD FirePro S9170 is the industry’s first server GPU with 32GB ultra-fast GDDR5 on-board memory and features a 512-bit memory interface for 320 GB/s of memory bandwidth;
  • Equipped with 32GB of GDDR5 memory, the AMD FirePro S9170 GPU can accelerate memory-intensive applications and process larger and more computationally complex workloads with ease; and
  • The AMD FirePro S9170 GPU features 33% more memory than the competitive GPU, helping to improve overall workload speed and system responsiveness, especially when working with large amounts of data. via AMD


“We have been developing a fully-parallel computational tool based on the AMD GPU heterogeneous computing platform and OpenCL,” said Omid Mahahadi, co-founder and director, Geomechanica Inc. “This tool accurately captures the complex physics of massive mines plus oil and gas fields rapidly and reliably. Thanks to the impressive 32GB of memory of the new cards, we expect to run computations on massive data structures containing tens of millions of data elements. The combination of rapid double-precision operations with the large memory capacity enables accurate, detailed, and reliable computations. A similar performance using CPUs would likely require much higher capital and maintenance costs. Moving forward, we plan to take advantage of the recent features of the OpenCL 2.0 open API to further enhance the performance of our software.”

Read more: http://wccftech.com/amd-launches-fi...enadas-524-tflop-compute-power/#ixzz3fPBg7nlc
Click to expand...
 
AMD's marketing compares the 32GB to K80's 24GB. But that's a poor comparison; the K80 has dual GPUs with 12 GB each. AMD's single memory pool is far more useful for memory capacity limited applications even if it wasn't bigger.
 
lanek said:
How is it possible they got half the ram usage than Nvidia gpu's ?( in gamess who dont go over 3000mb not limited by the vram capacity )... Thoses games are not new, so, cant be a developper work. or a special codes in the games who differ for Nvidia and Fury. If ofc the report of vram usage is not bugged.
Click to expand...
Just because your card shows 4GB being used doesn't necessarily mean you're hitting a limit. 1GB (or whatever arbitrary amount) of that could be stuff that get's flushed as necessary with no perf hit. You will often see cards that have more VRAM use more VRAM in the same game same settings as another card with less VRAM.
 
spworley said:
AMD's marketing compares the 32GB to K80's 24GB. But that's a poor comparison; the K80 has dual GPUs with 12 GB each. AMD's single memory pool is far more useful for memory capacity limited applications even if it wasn't bigger.
Click to expand...
How is AMD smacking so much GDDR5 on there? Is it clamshell mode?

Also why hasn't NVIDIA put 24GB on a GK210? Seems like they would be more than happy to if the demand was there, unless there is a technical reason why they can't.
 
homerdog said:
How is AMD smacking so much GDDR5 on there? Is it clamshell mode?

Also why hasn't NVIDIA put 24GB on a GK210? Seems like they would be more than happy to if the demand was there, unless there is a technical reason why they can't.
Click to expand...
Yep. 32 8Gb chips using clamshell mode. 8Gb GDDR5 chips are still very new.
 
Thanks Ryan. I totally thought those cards were using 8Gb chips but I guess those weren't available then.

Were there ever any NVIDIA cards that used clamshell GDDR5? Since they are by far the leader in the professional GPU compute market I thought they would've been all over it but from what I can tell not even the greatest Tesla hardware has used it. I don't think it should be an architectural limitation but it does make me wonder.

This will all be ancient history soon with HBM2 and HMC around the corner but it still interests me.
 
homerdog said:
Thanks Ryan. I totally thought those cards were using 8Gb chips but I guess those weren't available then.

Were there ever any NVIDIA cards that used clamshell GDDR5? Since they are by far the leader in the professional GPU compute market I thought they would've been all over it but from what I can tell not even the greatest Tesla hardware has used it. I don't think it should be an architectural limitation but it does make me wonder.

This will all be ancient history soon with HBM2 and HMC around the corner but it still interests me.
Click to expand...
Most of the Quadro/Tesla cards use clamshell, as do all of the Titan cards, and I believe the 4GB GTX 960s as well. It's been around for years and has been in many products.
 
Ryan Smith said:
Most of the Quadro/Tesla cards use clamshell, as do all of the Titan cards, and I believe the 4GB GTX 960s as well. It's been around for years and has been in many products.
Click to expand...
Interesting. I guess the smaller chips are less than half the price of the larger ones. Otherwise I can't see why Titan would use clamshell.
 
Are they compressing textures when loading them maybe? If it's true that because of Dx10 they are still not very compressed. And as game sizes are generally similar between condole and PC, does that mean console versions use uncompressed too? If so there's a significant bandwidth gain to be had there? (And download/disc size bloat)
 
You must log in or register to reply here.
Back
Top