NVIDIA is joining Intel and AMD with a keynote at the virtual event this year. We are going to be covering the NVIDIA Computex 2021 keynote live ahead of a big NVIDIA GPU server review a few hours later. Please excuse typos as this is being done live for the keynote to end a long Memorial Day holiday weekend in the US.
NVIDIA Computex 2021 Keynote Coverage
NVIDIA started this keynote off by, delaying about 15 minutes. More to come as the presentation starts.
Since we were pre-briefed on the enterprise news, there is a good chance that this is going to be a GeForce-heavy presentation. Kicking off, some fun numbers in terms of NVIDIA’s gaming TAM:
As expected, the NVIDIA GeForce 3080 Ti just made its debut.
NVIDIA says that the new card is roughly equivalent to 1.5x GeForce RTX 2080 Ti‘s.
In terms of key specs, we get 12GB of GDDR6X memory and NVIDIA says it will be “available” on June 3 for $1199.
Beyond the 3080 Ti, there is a lower-end card, the NVIDIA GeForce 3070 Ti which the company says is around 1.5x a RTX 2070 Super at $599 with 8GB of GDDR6X memory.
For many STH readers, this is going to sound great, except it will do little to help availability. We are very excited for the RTX 3080 Ti with 12GB of memory as that may be a very useful SKU for our lab. We can also see that NVIDIA is keeping the 3000 series design language with the coolers which means that these will be less useful in higher-density scenarios with the Founders Edition cooler so we will have to hope for blower-style coolers to hit the market.
On the enterprise side, NVIDIA is pushing more into its model with support offerings. Here is an interesting slide with some of NVIDIA’s AI ecosystem.
NVIDIA is discussing the DPU. Since we just covered A Quick Look at Logging Into a Mellanox NVIDIA BlueField-2 DPU and DPU vs SmartNIC and the STH NIC Continuum Framework we are going to point our readers there.
NVIDIA has its Base Command tool to make the management of AI clusters and jobs easy. This is similar to the Inspur AIStation for AI Cluster Operations Management Solution that we took a look at last year, but NVIDIA is offering this and will be bringing it to the DGX along with AWS and GCP.
The NVIDIA DGX SUPERPOD is not available as a subscription model. NVIDIA will host the DGX Superpod infrastructure and then lease the infrastructure to customers, similar to its public cloud customers that also resell hosted chips. NVIDIA’s goal is to help let customers try before buying.
This is a nice slide mapping GPUs to target markets.
NVIDIA has a slide on the DPU. We have a lot of content listed above.
We also have a What is a DPU A Data Processing Unit Quick Primer article.
NVIDIA is showing its DOCA BlueField ecosystem.
IBM Red Hat is building a NVIDIA Morpheous based developer kit with BlueField-2.
We have covered BlueField-3, but effectively these get even bigger for the next generation. We will have more compute resources as one would expect to go along with faster fabrics.
Another big announcement is that NVIDIA is extending NVIDIA Certified to Arm servers.
Overall, there is a lot that NVIDIA is announcing today.
Quite a smart, or deviant depending on your perspective, move by NVIDIA with the limited GPU memory on the 3080 Ti. The 2080 Ti used to be quite popular for AI servers, with 8 of these cards put e.g. into a Supermicro SYS-4029. With the increase in memory-demand to run modern machine learning models this seems to now get less and less attractive and push those customers towards the actual datacenter products…