NVIDIA, here are the new A10 and A30 graphics cards for AI, HPC and professionals
NVIDIA
NVIDIA introduced the A100 graphics card to market last year, built to deliver cutting-edge data center and supercomputer performance, but the product is too powerful and expensive for less demanding workloads. So, at its GTC press conference, the company unveiled two new models: A30, for analytics and artificial intelligence servers, and A10, for mixed computing and graphics workloads.The A30 graphics card represents the true little sister of the A100 and is based on the same compute-oriented Ampere architecture. In fact, it supports the same features, a wide range of formats for AI and HPC workloads (FP64, FP64TF, FP32, TF32, bfloat16, FP16, INT8, INT4) and even multi-instance GPU operations (MIG) with 6GB instances. From a performance standpoint, the A30 offers just over 50% of the performance of the A100, so we are talking about 10.3TFLOPS FP32, 5.2 TFLOPS FP64 and 165TFLOPS FP16 / bfloat16.
Regarding memory , the drive is equipped with 24GB of RAM with a bandwidth of 933GB / s. The subsystem appears to lack ECC support, which could be a limitation for those who need to work with large datasets, who should upgrade to the more expensive and powerful A100. NVIDIA A30 will be available in a dual-slot FHFL (full-height, full length) form factor, with a PCIe 4.0 x16 interface and a TDP of 165W, higher than the A100's 250W. Additionally, the A30 supports an NVLink at 200GB / s (a decrease in consistency compared to the A100's 600GB / s).
Instead, A10 is a completely different product that can be used for graphics, AI inference, and video encoding / decoding. The unit supports FP32, TF32, blfoat16, FP16, INT8 and INT4 formats for graphics and AI, but not the FP64 required for HPC. This is a single slot FHFL graphics card with a PCIe 4.0 x16 interface that will be installed in servers running NVIDIA RTX Virtual Workstation (vWS) software and equip workstations that require both AI and graphics capabilities remotely.
To a large extent, A10 should be a viable proposition for artists, designers, engineers and scientists (who don't need FP64). NVIDIA A10 appears to be based on the GA102 GPU (or a derivative thereof), but since it supports INT8 and INT4, we can't be 100% sure this is physically the same processor found on the GeForce RTX 3080/3090 and RTX A6000 cards. Finally, the performance of the A10 (31.2 FP32 TFLOPS, 125 FP16 TFLOPS) is similar to that of the GeForce RTX 3080 and on board there are 24GB of GDDR6 memory capable of offering a bandwidth of 600 GB / s.
Regarding availability, NVIDIA expects its partners to start marketing machines equipped with its A30 and A10 cards by the end of the year.
NZXT C850 is the right power supply if you want to try try your hand at overclocking, with 850W of power and 80Plus Gold certification. You can find it on Amazon.
3 Reasons to Buy NVIDIA After Its Investor Presentation
In today's video, I look at my favorite things from NVIDIA's Investor Day presentation. The presentation showed various new projects and products for the company, which drove the stock price up over 5%.
Here are three reasons to watch NVIDIA (NASDAQ:NVDA) after its presentation:
Click the video below for my full thoughts.
*Stock Prices used were the Mid-Day prices of April 12, 2021. The video was published on April 12, 2021.
This article represents the opinion of the writer, who may disagree with the “official” recommendation position of a Motley Fool premium advisory service. We’re motley! Questioning an investing thesis -- even one of our own -- helps us all think critically about investing and make decisions that help us become smarter, happier, and richer.
