Nvidia ces 2016 event

[kagui]

Just posting the link if someone is interested on seeing

time: 6 pm PST

http://blogs.nvidia.com/blog/2016/01/03/ces-las-vegas-event/

 

[96Firebird]

This is at 6pm PST, half hour from now.

 

[RaulF]

Pascal pascal pascal.

 

[n0x1ous]

just gonna be tegra and car stuff. nothing about pascal

 

[nvgpu]

Drive PX 2 announced, powered by Pascal GPU.

 

[xpea]

by far the most boring Nvidia CES event ever (first hour). They are going way too deep in this neural thing. Hey JHH, the C of CES means Consumer, its not a developer conference... absolutely out of the audience. very disappointing...

 

[jpiniero]

8 TF SP would be 30% higher than Titan X. GP102 maybe?

 

[tviceman]

Showed off next gen Tegra and Pascal GPU's, but did not announce either one. Boo.

 

[xpea]

8 TF SP would be 30% higher than Titan X. GP102 maybe?

Drive PX2 has 2 Pascal GPUs and 2 Pascal based Tegra SoCs
Each Tegra Pascal SoC has 4 Cortex A57 and 2 (next gen?) Denver cores

 

[Arachnotronic]

by far the most boring Nvidia CES event ever (first hour). They are going way too deep in this neural thing. Hey JHH, the C of CES means Consumer, its not a developer conference... absolutely out of the audience. very disappointing...

I was in the audience and completely agree that they went a bit too technical on that stuff. Was glad they at least demoed the new Tegra and Pascal silicon.

 

[Arachnotronic]

Drive PX2 has 2 Pascal GPUs and 2 Pascal based Tegra SoCs
Each Tegra Pascal SoC has 4 Cortex A57 and 2 (next gen?) Denver cores

Confirmed with Jen-Hsun that they are next generation Denver cores.

 

[xpea]

high res pic of PX2:
http://blogs.nvidia.com/wp-content/uploads/2016/01/ces-drive-px-2-square.jpg

Only the SoC side is available (GPUs are on the other side).
Between the 2 Nv SoCs, we can see an ALTERA Cyclone V SoC (ARM A9 MPcore variant):
https://www.altera.com/products/fpga/cyclone-series/cyclone-v/overview.html

And an INFINEON Tricore MCU (radar and vision specialized) is also spotted:
http://www.infineon.com/cms/en/prod...html?channel=db3a304342c787030142dbf31c2b151c

 

[frowertr]

This will excite my father. He has purchased quite a bit of Nvidia stock the last several years on the account that he believes their tech will pave the way for self driving cars.

But yeah, it as a bit boring to me...

 

[MrTeal]

Arachnotronic, were those MXM cards actually Pascal silicon or were they Maxwell placeholders? There seems to be conflicting information out there on that.

 

[MrTeal]

Arachnotronic, were those MXM cards actually Pascal silicon or were they Maxwell placeholders? There seems to be conflicting information out there on that.

NM, it appears that they're GM204.

I wonder what kind of size the die for the Pascal GPU planned for this will come in at? GM204 already gets 4+GFLOPS in the full mobile 980, so dual 980 MXMs would be at 8GFLOPS.

 

[StereoPixel]

Pascal its just Maxwell with FP64 support, said Nvidia.

 

[MrTeal]

Pascal its just Maxwell with FP64 support, said Nvidia.

On 16FF+. If those chips actually had been Pascal silicon, we'd be looking at a pretty large first die and a big performance increase over GP200. This doesn't really give us much more information of Pascal though, unfortunately. Hopefully we'll get some more info by CeBIT.

 

[Glo.]

IF we believe that PX2 is 250W and has dual GPUs with one Denver CPU, and assume that GPUs are getting 80% of power envelope(200W) that goes 100W for each GPU.

So it is 2048 core GPU clocked at 1000 MHz. 4 TFLOPs for each GPU at 100W.

Not bad at all.

 

[MrTeal]

IF we believe that PX2 is 250W and has dual GPUs with one Denver CPU, and assume that GPUs are getting 80% of power envelope(200W) that goes 100W for each GPU.

So it is 2048 core GPU clocked at 1000 MHz. 4 TFLOPs for each GPU at 100W.

Not bad at all.

Not really good, either. The specs are kind of strange for a Pascal product, since a Tesla M60 is already 9.7TFLOPS with a 300W TDP.

 

[StereoPixel]

On 16FF+. If those chips actually had been Pascal silicon, we'd be looking at a pretty large first die and a big performance increase over GP200. This doesn't really give us much more information of Pascal though, unfortunately. Hopefully we'll get some more info by CeBIT.

Tegra X1 its Maxwell as well, but on 20nm. Pascal its Maxwell as well, but with FP64 on 16nm.

 

[MrTeal]

Pascal its just Maxwell with FP64 support, said Nvidia.

Tegra X1 its Maxwell as well, but on 20nm. Pascal its Maxwell as well, but with FP64 on 16nm.

Got a source where nVidia said Pascal would just be Maxwell with FP64? That seems unlikely and contradicts most of what nVidia has hinted about with Volta and Pascal.

 

[nvgpu]

Best to ignore that person, he doesn't know anything and clearly extremely ignorant.

As seen from the leaked driver, Pascal is Compute Capability 6.0. Pascal is not Maxwell no matter how many times anyone tries to repeat that lie.

http://www.geeks3d.com/forums/index.php?topic=4247.0

Pascal

-D__CUDA_ARCH__=600

-D__CUDA_ARCH__=610

-D__CUDA_ARCH__=620

Personal attacks are not allowed.



esquared
Anandtech Forum Director

 

[Arachnotronic]

Pascal its just Maxwell with FP64 support, said Nvidia.

NVIDIA has said no such thing.

 

[StereoPixel]

Got a source where nVidia said Pascal would just be Maxwell with FP64?

http://cdn.wccftech.com/wp-content/uploads/2015/09/NVIDIA-Pascal-GPU_Compute-Performance.jpg

 

[Silverforce11]

http://cdn.wccftech.com/wp-content/uploads/2015/09/NVIDIA-Pascal-GPU_Compute-Performance.jpg

NV's claim of 10x compute performance of Pascal over Maxwell was in "Deep Learning" tasks that uses mix compute mode, something Maxwell is gimped at.

And NV does need a refresh of their Tesla lineup, which is still dominated by Kepler Titan.

So there's expectations that Pascal is a compute focused chip. Which should be good for the DX12 gaming era with the shift to compute-heavy rendering techniques.