bit-tech.net

AMD unveils FirePro W9100 16GB GPU

AMD unveils FirePro W9100 16GB GPU

AMD's FirePro W9100 offers a massive 16GB of GDDR5 memory, connected to a Hawaii GCN GPU offering up to five teraflops of single-precision compute power.

AMD has announced its latest workstation-oriented graphics board, the FirePro W9100, which packs an impressive 16GB of GDDR5 memory - nearly three times that of its predecessor the FirePro W9000.

The AMD FirePro W9100 is based around a 28nm implementation of the Graphics Core Next 1.1 'Hawaii' architecture, an upgrade from the GCN 1.0 'Tahiti' of its predecessor. Although full specifications aren't due to be announced until early next month, the company has confirmed an increase from 2,048 stream processors to 2,816 and from 128 texture units to 176. The result: five teraflops of single-precision compute performance, or 2.67 teraflops of double-precision.

Now, to put that into perspective, Nvidia's latest GeForce GTX Titan Z board offers eight teraflops of single-precision performance, but requires two GPUs in which to do it - and each GPU has access to only 6GB of the shared 12GB GDDR5 memory. The FirePro W9100, on the other hand, has only a single GPU which has the entire 16GB to itself - and, the company has confirmed, the boards will support stacking of up to four cards via CrossFire for systems that require higher performance.

The board, AMD explained during its press conference, is designed for those working on ultra-high resolution projects. As well as the increasingly popular UHD and 4K resolutions, the company has claimed to be seeing demand from markets looking to work with resolutions as high as 8K - which needs a significantly larger framebuffer than the company's previous FirePro W9000 6GB could offer.

Official pricing for the FirePro W9100 has yet to be confirmed, but those interested in getting their hands on one - or four - can expect to dig deep: as professional products the FirePros demand a hefty pricetag, the FirePro W9000 launched in August 2012 at $4,000 and the W9100 should easily smash that figure when it formally launches on the 7th of April.

12 Comments

Discuss in the forums Reply
Maki role 27th March 2014, 13:04 Quote
Despite Nvidia's marketing, this is much more akin to the Quadro K6000 really (a 12GB workstation card with 5.2TFLOPS SP but 1.4TFLOPS DP).

As I've mentioned before, if only there were a 12GB Titan, sadly these top end workstation cards are a fair chunk over my budget :(
schmidtbag 27th March 2014, 15:07 Quote
Quote:
Originally Posted by Maki role
Despite Nvidia's marketing, this is much more akin to the Quadro K6000 really (a 12GB workstation card with 5.2TFLOPS SP but 1.4TFLOPS DP).

As I've mentioned before, if only there were a 12GB Titan, sadly these top end workstation cards a fair chunk out of my budget :(

Workstation graphics are getting increasingly difficult to justify. Aside from the fact that they generally offer more memory, the only thing that really separates them are the drivers. It is possible to buy a gaming GPU and through BIOS updates, convert it into a fully-functional workstation GPU, but, this doesn't work with all models and it wouldn't surprise me if doing this wrong could brick your GPU.

One thing I always found interesting about workstation GPUs is how much of a difference the brand makes. It kind of irritates me because nvidia tends to support the most popular programs, and therefore people assume nvidia is automatically the best choice, when there are some tests where they are noticeably behind the AMD alternative.

What I don't get is why intel doesn't make workstation graphics. They're a little too far behind to make a good gaming GPU, but workstation GPUs tend to be significantly weaker than their gaming counterparts, because they're heavily optimized for the software they're designed for. Intel's GPUs have very low clock rates, why couldn't they do the same?
Gareth Halfacree 27th March 2014, 15:14 Quote
Quote:
Originally Posted by schmidtbag
What I don't get is why intel doesn't make workstation graphics.
They do, except they position it as an accelerator board and call it the Xeon Phi. No, really: Xeon Phi, which packs 50 Pentium-class x86 cores on a PCIe board with a chunk of GDDR5 memory, started life as Larrabee - Intel's aborted attempt at competing with AMD and Nvidia in the high-performance discrete graphics market. Okay, so it's a competitor to Tesla rather than Quadro, but it's still basically a highly parallel GPU-like product - albeit one without any graphics outputs (again, like Tesla.)
Mister_Tad 27th March 2014, 15:48 Quote
And aside from Phi, they do workstation graphics "lite" on Xeons with onboard graphics. The HD P-series graphics are nothing more than the HD parts with certified drivers.
GuilleAcoustic 27th March 2014, 15:49 Quote
Xeons with a "5" at the end have an IGP with certified drivers. Intel HD4600 is called P4600 on xeon E3-12*5 V3. I do not know if it gives better performances, but the IGP gains ECC memory.

Edit: Ninja-ed by Mister_Tad
schmidtbag 27th March 2014, 15:55 Quote
Quote:
Originally Posted by Gareth Halfacree
They do, except they position it as an accelerator board and call it the Xeon Phi. No, really: Xeon Phi, which packs 50 Pentium-class x86 cores on a PCIe board with a chunk of GDDR5 memory, started life as Larrabee - Intel's aborted attempt at competing with AMD and Nvidia in the high-performance discrete graphics market. Okay, so it's a competitor to Tesla rather than Quadro, but it's still basically a highly parallel GPU-like product - albeit one without any graphics outputs (again, like Tesla.)

I've heard of that, but as far as I'm aware that wasn't quite the same thing. IIRC, that was supposed to compete with CUDA and openCL rather than act as an actual GPU. In other words, you wouldn't use CAD software with it, though you might use it for rendering. I could be wrong though. Regardless, it was a stupid product idea. I REALLY wish Intel would realize that x86 isn't the answer to everything.
GuilleAcoustic 27th March 2014, 16:16 Quote
Quote:
Originally Posted by schmidtbag
I've heard of that, but as far as I'm aware that wasn't quite the same thing. IIRC, that was supposed to compete with CUDA and openCL rather than act as an actual GPU. In other words, you wouldn't use CAD software with it, though you might use it for rendering. I could be wrong though. Regardless, it was a stupid product idea. I REALLY wish Intel would realize that x86 isn't the answer to everything.

They do have the Intel P4600 /4700 IGP on Xeons E3-12*5 V3, but you can only use it with an Intel C226 based motherboard.
Mister_Tad 27th March 2014, 16:18 Quote
Quote:
Originally Posted by GuilleAcoustic
Xeons with a "5" at the end have an IGP with certified drivers. Intel HD4600 is called P4600 on xeon E3-12*5 V3. I do not know if it gives better performances, but the IGP gains ECC memory.

Edit: Ninja-ed by Mister_Tad

I'm like a cat!

There are pretty significant performance gains in apps for which the drivers have been tweaked, but as it's basically the same part as the HD with the same core capabilities, everything else is the same.

I've found the list before (I have a Xeon E3, so was interested), but can't find it now, just "Certified on 15 applications from Autodesk, Adobe, Solid Works, Bentley and Siemens"

EDIT: ah ha! http://www.intel.com/content/www/us/en/workstations/certified-applications.html
GuilleAcoustic 27th March 2014, 16:30 Quote
Thanks for the list. I'm tempted by an E3 too but that will be for Linux use :)
asura 27th March 2014, 19:29 Quote
Quote:
Originally Posted by schmidtbag
Workstation graphics are getting increasingly difficult to justify. Aside from the fact that they generally offer more memory, the only thing that really separates them are the drivers. It is possible to buy a gaming GPU and through BIOS updates, convert it into a fully-functional workstation GPU, but, this doesn't work with all models and it wouldn't surprise me if doing this wrong could brick your GPU.

With Nvidia this hasn't worked since the 7xxx series - yes an appropriate quadro BIOS can be flashed to a contemporary card (? I certainly experimented with 8800GTX&8800GTS versus a FX4600) the benefits are; minimal, non-existent or, in many cases offer a lower frame rate than the GPU before it's alteration.

Can't comment on ATI/AMD
schmidtbag 27th March 2014, 20:05 Quote
Quote:
Originally Posted by asura
With Nvidia this hasn't worked since the 7xxx series - yes an appropriate quadro BIOS can be flashed to a contemporary card (? I certainly experimented with 8800GTX&8800GTS versus a FX4600) the benefits are; minimal, non-existent or, in many cases offer a lower frame rate than the GPU before it's alteration.

Can't comment on ATI/AMD

If the hardware is physically different, it won't work. AMD has a tendency to just re-brand one of their gaming GPUs as firepro and just alter the drivers. I'm sure nvidia might make some small tweaks here and there to deliberately prevent people from cheating.

Performance has proven to be significant, but of course it heavily depends on what software you're testing. If you're playing games, performance will most certainly drop.
mi1ez 27th March 2014, 22:42 Quote
A 16GB GPU is a bit of a misnomer, no?

A 16GB graphics card maybe...
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.



Discuss in the forums