Nvidia Pascal Announcement

[CSI PC]

Nvidia launching new Pascal dGPUs in the Quadro range, including a 'P100' variant:
The Quadro GP100 has a TDP of 235W, which is lower than the 250W for the PCIe P100, worth remembering though it is more general purpose than the P6000 that has higher FP32 TFLOPs.
The lower powered Quadros are interesting as well, models between 30W to 47W and then 75W model followed by 105W model, but still not quite as impressive as the specialist Tesla P4.
An interesting addition to the Quadro GP100 is the NVLink to be used for linking 2 of these cards together instead of SLI, this is not available on the Tesla PCIe P100 models and makes sense considering its primary purpose.
http://www.nvidia.com/object/quadro-graphics-with-pascal.html
http://www.tomshardware.com/news/nvidia-workstation-quadro-pascal-grahpics,33584.html
http://www.anandtech.com/show/11102/nvidia-announces-quadro-gp100

Cheers

 

[ImSpartacus]

Nvidia launching new Pascal dGPUs in the Quadro range, including a 'P100' variant:
The Quadro GP100 has a TDP of 235W, which is lower than the 250W for the PCIe P100, worth remembering though it is more general purpose than the P6000 that has higher FP32 TFLOPs.
The lower powered Quadros are interesting as well, models between 30W to 47W and then 75W model followed by 105W model, but still not quite as impressive as the specialist Tesla P4.
An interesting addition to the Quadro GP100 is the NVLink to be used for linking 2 of these cards together instead of SLI, this is not available on the Tesla PCIe P100 models and makes sense considering its primary purpose.
http://www.nvidia.com/object/quadro-graphics-with-pascal.html
http://www.tomshardware.com/news/nvidia-workstation-quadro-pascal-grahpics,33584.html
http://www.anandtech.com/show/11102/nvidia-announces-quadro-gp100

Cheers

Anybody else think it's kinda cool that they appeared to have lopped off a controller on the P2000 to "legitimately" get to 5GB of VRAM in GP106? Seems neat to me.

 

[CSI PC]

Anybody else think it's kinda cool that they appeared to have lopped off a controller on the P2000 to "legitimately" get to 5GB of VRAM in GP106? Seems neat to me.

I wonder why they did this though, unless part of the way to reach 75W with the smaller bus or the fact even reduced it is still improved spec compared to the Maxwell M2000.
To me in the group the P4000 seems rather special with its performance/spec at 105W, still mulling over how good the lower powered cards/profile are but nice to see 30W-47W range from Nvidia finally updated and they will have their place with quite a few I think.
Cheers

 

[Voxilla]

Nvidia launching new Pascal dGPUs in the Quadro range, including a 'P100' variant:
The Quadro GP100 has a TDP of 235W, which is lower than the 250W for the PCIe P100, worth remembering though it is more general purpose than the P6000 that has higher FP32 TFLOPs.
The lower powered Quadros are interesting as well, models between 30W to 47W and then 75W model followed by 105W model, but still not quite as impressive as the specialist Tesla P4.
An interesting addition to the Quadro GP100 is the NVLink to be used for linking 2 of these cards together instead of SLI, this is not available on the Tesla PCIe P100 models and makes sense considering its primary purpose.
http://www.nvidia.com/object/quadro-graphics-with-pascal.html
http://www.tomshardware.com/news/nvidia-workstation-quadro-pascal-grahpics,33584.html
http://www.anandtech.com/show/11102/nvidia-announces-quadro-gp100

Cheers

A Titan-X P quite easily does 14TF FP32, with a stable overclock ~2 GHz.

 

[CSI PC]

A Titan-X P quite easily does 14TF FP32, with a stable overclock ~2 GHz.

Any links where reviews managed to sustain 2 GHz?
Point is within normal boost clocks and that is 12 TFLOPs FP32 for the Quadro P6000 due to being a full core GP102, businesses/pro implementations do not tend to OC and this is not a consumer part.
But even in game it could be interesting as an OC Titan X will not always beat it, 17% is a lot to make up in Hitman for average and the 1% avg probably is a lot worst (below is normal clocks and unfortunately just avg/min fps).
http://hothardware.com/reviews/nvidia-quadro-p6000-and-p5000-workstation-gpu-reviews?page=6

This is their experience with an OC Titan X: http://hothardware.com/reviews/nvidia-titan-x-review?page=6
Thanks.

 

[CSI PC]

Anyone else notice the interesting spec of the Quadro GP100 in that it is 235W and yet more similar to the NVLink P100 than the PCIe version in terms of performance and also having a form of NVLink?
Quadro GP100 235W with 10.3 FP32 TFLOPs and a partial NVLink
NVLink P100 300W with 10.6 FP32 TFLOPs
PCIe P100 250W with 9.3 FP32 TFLOPs no NVLink

Nvidia are usually pretty consistent with their specs when released at this point so interesting whether this is down to improvements of the TSMC node or to do with the GPU (seems more likely), so higher clocks for less power when compared to the PCIe P100.
I was thinking maybe they did similar to the original Titan where one had to enable the 'DP mode' to get full FP64 and while disabled one had higher clocks/better power demand, but would they do that for a modern dGPU I doubt it.
Cheers

 

[CSI PC]

Nvidia has a few interesting (or amusing when considering them at a high level compared to AMD) upcoming events at GDC 2017.
D3D Async Compute for Physics: Bullets, Bandages, and Blood: https://developer.nvidia.com/d3d-async-compute-physics-bullets-bandages-and-blood
NVIDIA Vulkan Update: https://developer.nvidia.com/nvidia-vulkan-update
Vulkan update includes further update on Vulkan Device-Generated Commands extension from December (I think): https://developer.nvidia.com/device-generated-commands-vulkan

Amongst several other very interesting ones in the list:
Wednesday 3/1/2017

• 9:30-10:30 AM | Introduction to Deep Learning
• 11:00-12:00 PM | Zoom, Enhance, Synthesize! Magic Image Upscaling and Material Synthesis using Deep Learning
• 12:30-1:30 PM | Photogrammetry for Games; Art, Technology and Pipeline Integration for Amazing Worlds
• 3:30-4:30 PM | Watch Dogs 2 - PC Version Success Story with NVIDIA
• 5:00-6:00 PM | DirectX 12 Case Studies

Thursday 3/2/2017

• 10:00-11:00 AM | D3D Async Compute for Physics: Bullets, Bandages, and Blood
• 11:30-12:30 PM | Game Physics on the GPU with NVIDIA PhysX 3.4
• 2:00-2:30 PM | Adding photo mode to your game with NVIDIA Ansel
• 3:00-3:30 PM | NVIDIA Aftermath: A new way of debugging crashes on the GPU
• 4:00-5:00 PM | NVIDIA GameWorks Animation Technologies in Unreal Engine 4
• 5:30-6:30 PM | Real-Time Rendering Advances from NVIDIA Research

Friday 3/3/2017

• 10:00-11:00 AM | NVIDIA Vulkan Update / Vulkan GPU Work Creation
• 11:30-12:00 PM | How to Stream Your Game to Millions with GeForce NOW
• 12:15-1:15 PM | Accelerating your VR Games with VRWorks
• 1:30-2:30 PM | The Witness on Android - Post Mortem
• 3:00-4:00 PM | VR Best Practices: Putting the fun in VR Funhouse

https://developer.nvidia.com/gdc17
Cheers

 

[iMacmatician]

From WCCFTech: "NVIDIA GTX 1080 Ti Rumored To Launch In March, Reportedly Now In Production."

The card will offer similar performance to the company’s flagship $1000 TITAN X but at a more affordable price for gamers sources claim. Furthermore, unlike the TITAN X, custom designed and aftermarket cooled versions of the GTX 1080 Ti will be available shortly after release. The newcomer is speculated to be priced between $750 and $900 at launch and to land sometime between March 20th and March 23rd.

The rumored specs for the GTX 1080 Ti place both the base and boost GFLOPS at 98.5% of those of the TITAN X (2016). That's similar to how the 780 Ti compares to the TITAN BLACK, although in that case the TITAN card was released after the GTX ## card.

 

[ImSpartacus]

From WCCFTech: "NVIDIA GTX 1080 Ti Rumored To Launch In March, Reportedly Now In Production."

The rumored specs for the GTX 1080 Ti place both the base and boost GFLOPS at 98.5% of those of the TITAN X (2016). That's similar to how the 780 Ti compares to the TITAN BLACK, although in that case the TITAN card was released after the GTX ## card.

Yeah, this seems like bad news for Vega 10 (if true).

Nvidia isn't stupid. They very deftly cut Fiji's legs out from under out with the 980 Ti and I think we all fully expect them to attempt the same with the 1080 Ti to Vega 10.

So if you assume that the 1080 Ti will be enough to compete with Vega 10, then things aren't going well for the red team.

 

[Lightman]

Yeah, this seems like bad news for Vega 10 (if true).

Nvidia isn't stupid. They very deftly cut Fiji's legs out from under out with the 980 Ti and I think we all fully expect them to attempt the same with the 1080 Ti to Vega 10.

So if you assume that the 1080 Ti will be enough to compete with Vega 10, then things aren't going well for the red team.

I would say it is a very good news for consumer! I want Vega (if good) and I want it at reasonable price. If Vega is only as fast as GTX10180 then I'm afraid 1080Ti will cost arm and leg to buy.

 

[Kaotik]

https://www.io-tech.fi/uutinen/vahvistus-geforce-gtx-1080-ti-esitellaan-28-helmikuuta/

We at io-tech have been able to confirm that GTX 1080 Ti will indeed be launched at the end of this month in GDC, and that it'll come on sale as custom AIB models in March

 

[ImSpartacus]

https://www.io-tech.fi/uutinen/vahvistus-geforce-gtx-1080-ti-esitellaan-28-helmikuuta/

We at io-tech have been able to confirm that GTX 1080 Ti will indeed be launched at the end of this month in GDC, and that it'll come on sale as custom AIB models in March

So if custom stuff is coming in March, then does that mean no founder's edition?

Seems like there wouldn't be a window of time for the fe to exploit.

But I recall interviews with falcon northwest folks that said they appreciated the founder's edition because it provided a consistent card/cooler/tdp that could be designed against. So Nvidia might still have an excuse to make a founder's edition.

 

[DavidGraham]

NV teases a TI card on it's Geforce site and Facebook page: It's almost Time, with a 6 days countdown.
http://www.geforce.com

Obviously a reference to 1080TI, IMO if this isn't the full GP102, then they are shooting themselves in the foot.

 

[Kaotik]

NV teases a TI card on it's Geforce site and Facebook page: It's almost Time, with a 6 days countdown.
http://www.geforce.com

Obviously a reference to 1080TI, IMO if this isn't the full GP102, then they are shooting themselves in the foot.

Latest rumors suggest it will have even less SMs active than Titan X, which isn't full either

 

[DavidGraham]

Latest rumors suggest it will have even less SMs active than Titan X,

Yeah, I know that, but they are advertising the 1080Ti as the ultimate geforce, don't think a heavily cut down Titan Pascal applies for the name ultimate. If so, then NV is stupid.

 

[Kaotik]

Yeah, I know that, but they are advertising the 1080Ti as the ultimate geforce, don't think a heavily cut down Titan Pascal applies for the name ultimate. If so, then NV is stupid.

Technically Titan X (Pascal) isn't GeForce, so they just might

 

[lanek]

Yeah, I know that, but they are advertising the 1080Ti as the ultimate geforce, don't think a heavily cut down Titan Pascal applies for the name ultimate. If so, then NV is stupid.

Marketing... vist any sites of brands ( whatever it is AMD, Nvidia, motherboard brands, tv brands etc ).. "the ultimate... the world faster.... the next step in future... " well choose your marketing formula.

 

[lanek]

NV teases a TI card on it's Geforce site and Facebook page: It's almost Time, with a 6 days countdown.
http://www.geforce.com

Obviously a reference to 1080TI, IMO if this isn't the full GP102, then they are shooting themselves in the foot.

If you analyze the page, you will see 1080TI writen in full letters.

...........................................
(sorry double post ... )

 

[DavidGraham]

Technically Titan X (Pascal) isn't GeForce, so they just might

Marketing... vist any sites of brands ( whatever it is AMD, Nvidia, etc ).. "the ultimate... the world faster.... the next step in future... well choose your marketing formula.

Historically, the ultimate Geforce chip has always sported the full count of CUDA cores available: 8800GTX, 285, 580 ,680, 780Ti, 980, 980Ti .. etc, the only exception was the 480, and that was the exception due to heat and power consumption and then quickly got rectified with the 580. The 780Ti even scored more cores than the original Titan Kepler (not counting Titan Black). So I can't see a solid reason why NV would go out of it's way to announce a half assed cut down chip as the ultimate Geforce chip, when everyone in the enthusiast community knows there is a more complete "ultimate" chip. Unless they become morons of course!

 

[DuckThor Evil]

980Ti was not a full chip. Interesting to see the final config and if there are any updates to the Titan X, which has been out of stock for a while.