NVIDIA ‘Ampere’ 8nm Graphics Cards

[D.P.]

At nearly 3x the transistor density the 3000 series will be a massive leap forwards over Turing.

 

[TNA]

At nearly 3x the transistor density the 3000 series will be a massive leap forwards over Turing.

Yea but Panos says that is not possible

 

[KillBoY_UK]

https://www.anandtech.com/show/15801/nvidia-announces-ampere-architecture-and-a100-products

The GA100 has 8192 shaders,but the A100 has 6192 shaders and a 400W TDP,which is an increase over the previous 12NM V100 which was 5120 shaders and a 300W~350W TDP. The GPU is huge at 826MM2. Poor yields a problem??

new and different shaders/cores.......

 

[Troezar]

Man, that is not enough RAM ffs. They need to be going 12 for 3070 and 3080, 16 for 3080Ti and 24 for Titan.

Apparently the RAM gets restricted as there are somepsome applications that could make use of the RAM. Without sufficient RAM users have to buy the expensive professional cards.

 

[Skribla]

The DGX-A100 system being priced at $199,000 compared to $399,000 for previous gen could be a slight glimmer of hope for the pricing of the gaming cards that's a pretty drastic reduction in price.

 

[Rroff]

The DGX-A100 system being priced at $199,000 compared to $399,000 for previous gen could be a slight glimmer of hope for the pricing of the gaming cards that's a pretty drastic reduction in price.

Some of that probably comes from not being paired with a hilariously overpriced Intel CPU/board.

 

[TNA]

Apparently the RAM gets restricted as there are somepsome applications that could make use of the RAM. Without sufficient RAM users have to buy the expensive professional cards.

Que?

 

[Troezar]

Que?

Nvidia deliberately put less memory on their consumer cards so they are less able to run professional tasks, apparently. No idea if that's true but makes sense. In the same was desktop and server chips are segmented.

 

[Grim5]

It already is in cars right now. The latest Mercedes A-Class has nvidia powering the dual panel display and augmented reality.

https://blogs.nvidia.com/blog/2018/02/02/mercedes-benz-a-class-mbux-ai-graphics/

Audi has been using Nvidia powered in car systems since 2017, it's not new

 

[K1LaW8tt5]

Apparently the RAM gets restricted as there are somepsome applications that could make use of the RAM. Without sufficient RAM users have to buy the expensive professional cards.

I'm just quoting this to tell you my phone was driving me mad because if I clicked in the middle of the text box to add some words it would sometimes keep duplicating words and stick a letter from the line in the middle of it. Glad it's not just me. Im assuming that's what happened with the somepsome

 

[Phixsator]

$699 for a 3080 and I would bite.

I wouldn't, it would have to be the 3080 ti for that to happen and no I'm not a betting man.

 

[Gerard]

I wouldn't, it would have to be the 3080 ti for that to happen and no I'm not a betting man.

Even if the FE version is that price you can bet your ass the strix version will be about a grand.

 

[JediFragger]

bet your ass the strix version will be about a grand.

LOL FO, £1500 easy

 

[D.P.]

new and different shaders/cores.......

There are a lot more transistors dedicated to the Tensor cores, the standard FP32 CUDA cores have only a modest increase because they have less importance in HPC or ML workloads.

Actual benchmarks show a 7x performance increase in such work loads

 

[Sargatanas2511]

I wouldn't, it would have to be the 3080 ti for that to happen and no I'm not a betting man.

You may as well say you're not buying. Surely performance is more important than saying what you will or won't buy at X price point?

 

[Troezar]

I'm just quoting this to tell you my phone was driving me mad because if I clicked in the middle of the text box to add some words it would sometimes keep duplicating words and stick a letter from the line in the middle of it. Glad it's not just me. Im assuming that's what happened with the somepsome

Lol could be, OnePlus 7T here

 

[RusstheDroog]

Well i'm glad i didnt go and get all amped like he told me to.

 

[McGraw]

Well i'm glad i didnt go and get all amped like he told me to.

do I detect some resistance?

 

[Donnie Fisher]

It's there not chat about the GPU making heavy use of texture compression, so the effective ram could potentially be notably higher than the base figure. So 8 GB appearing something closer to 12 etc.

 

[Rroff]

It's there not chat about the GPU making heavy use of texture compression, so the effective ram could potentially be notably higher than the base figure. So 8 GB appearing something closer to 12 etc.

Doesn't help if the game already uses some form of texture compression though (depending a bit on the implementation).