Catch up on our recent newsletters

Here you will find an archive of our previous newsletters, bear in mind any offers or promotions may now be expired

When it comes to AI training workloads, any GPU-accelerated server needs to be performing at its optimum capacity to get the fastest results. But at the same time, high powered servers consume a lot of power and in today’s environment of rising energy costs, driving those AI projects can become very expensive. Whilst it is true that all of NVIDIA’s DGX appliances have great performance the evolution from Pascal GPU architectures through Volta, Ampere and onto the latest Hopper models, has stepped things up with each generation - so much so that an older DGX may be costing much more to deliver those results than you may think. The video below may surprise you…

As shown in the video, the DGX H100 is way ahead of the older DGX appliances in every test carried out, but perhaps most importantly it’s the performance per kilowatt and performance per pound comparisons that tell the greatest story. Despite its higher unit cost and power consumption, the DGX H100 is up to 14 times more energy efficient and 47 times better value for money than its predecessors. Older hardware may also lead to reduced data scientist and developer productivity and higher maintenance costs.

Introducing the NVIDIA DGX H100

As we’ve seen, the newest DGX is named for its use of H100 GPUs based on the Hopper architecture. This marks a significant departure from earlier DGXs which shared the same underlying GPU architecture as other contemporary NVIDIA GPUs. In contrast Hopper is designed exclusively for AI workloads, so is an order of magnitude faster. The DGX H100 features eight H100 Tensor Core GPUs, each with 80MB of memory, providing up to 6x more performance than previous generation DGX appliances and is supported by a wide range of NVIDIA AI software applications and expert support.

8x NVIDIA H100 GPUs WITH 640 GIGABYTES OF TOTAL GPU MEMORY 18x NVIDIA® NVLink® connections per GPU, 900 gigabytes per second of GPU-to-GPU bidirectional bandwidth

4x NVIDIA NVSWITCHES™ 7.2 terabytes per second of bidirectional GPU-to-GPU bandwidth, 1.5X more than previous generation

8x NVIDIA CONNECTX®-7 and 2x NVIDIA BLUEFIELD® DPU 400 GIGABITS-PER-SECOND NETWORK INTERFACE 1 terabyte per second of peak bidirectional network bandwidth

DUAL x86 CPUs AND 2 TERABYTES OF SYSTEM MEMORY Powerful CPUs for the most intensive AI jobs

30 TERABYTES NVME SSD High speed storage for maximum performance

Trade In Your Old DGX

There’s never been a better time to upgrade to a new DGX H100, as for a limited time only, you can receive a discount when trading in an older DGX model - especially as NVIDIA has recently announced the End of Service Life (EOSL) date for DGX-1 and DGX-2 being June 25th 2023. Contact the Scan AI team to find out more on 01204 474747 or at [email protected]

Great reasons to shop at Scan