There has been lots of talk about graphical power lately thanks to the upcoming gaming consoles and the new Unreal engine that are bound to take graphics and our debt to the next level. As fancy as all that graphical power is though, it’s small fry to what is actually possible and if you don’t believe me, well then take a look at Nvidia’s new GPU which will make even the most expensive PC kit cry at its inadequacy for many years to come
Nvidia has just announced the release of its new Ampere A100 GPU that includes 19.5 teraflops of FP32 performance, 6,912 CUDA cores, 40GB of memory, and 1.6TB/s of memory bandwidth. Yes, that’s correct, this GPU is capable of more flops than even Crysis knows what to do with. To be fair though this is not your ordinary GPU and not meant to power your gaming PC at home (though no doubt some rich and crazy person may try), but is actually designed to power massive servers in data centres built for scientific computing, cloud graphics, and data analytics. Or mining Bitcoin.
It gets better though because Nvidia isn’t just talking about one of those wonderful GPUS at work though, Instead, according to The Verge, Nvidia is combining these GPUs into a stacked AI system that will power its supercomputers in data centres around the world. The entire DGX A100 system promises 5 petaflops of performance, thanks to these eight A100s, and they’re being combined using Nvidia’s third-generation version of NVLink.
Combining these eight GPUs means that there’s 320GB of GPU memory with 12.4TB/s of memory bandwidth. Nvidia is also including 15TB of Gen4 NVMe internal storage to power AI training tasks. This will enable researchers and scientists using the DGX A100 systems to split workloads into up to 56 instances, spreading smaller tasks across the powerful GPUs.
Needless to say, I want one, even though I won’t be able to do anything with it. Nvidia doesn’t give a price for what this system will cost, which is probably a good thing because I don’t think I would want to be tempted to even pretend I would find the money to be able to afford this.
Last Updated: May 18, 2020
HairyEwok
May 18, 2020 at 08:38
All that computing power for research and we still don’t have a solution to fanboy-ism 😛
Iskape
May 18, 2020 at 09:53
If you could put a price to it, what would it be? I’d say in the region of R200,000 – R250,000, considering Titan’s are about half the specs and you can get one for around R80,000 in SA.
HvR
May 18, 2020 at 12:47
Since it can only be used on the Nvidia specific server motherboards via Nvidia interlink etc costing will probably be very dependent on the completed sets you buy as a supplier.
The DDX A100 U2 integrated boxes have price tag of $199k
HvR
May 18, 2020 at 12:47
Since it can only be used on the Nvidia specific server motherboards via Nvidia interlink etc costing will probably be very dependent on the completed sets you buy as a supplier.
The DDX A100 U2 integrated boxes have price tag of $199k
HvR
May 18, 2020 at 12:47
Since it can only be used on the Nvidia specific server motherboards via Nvidia interlink etc costing will probably be very dependent on the completed sets you buy as a supplier.
The DDX A100 U2 integrated boxes have price tag of $199k
HvR
May 18, 2020 at 12:49
Interesting that Nvidia seems to able to produce the Ampere architecture for cheaper than Turing and they predicts RTX30xx cards to come in a cheaper price than the RTX20xx cards.