Nvidia Speeds up Data Center Graphics Offering With Tesla V100s

Nvidia Tesla V100

Nvidia Tesla V100 (Image credit: Nvidia)

Nvidia already touted its Tesla V100 as the world's most advanced data center graphics card. But now it's kicking things up a notch with the brand new, speedier Tesla V100s detailed today by TechPowerUp

The Tesla V100s is physically identical to the V100. It adheres to a similar dual-slot, PCIe add-in card (AIC) design and employs Nvidia's reference cooler. It's unknown if Nvidia will offer the V100s in the SXM2 board form factor, however.

Like its predecessor, the Tesla V100s is based on Nvidia's current Volta microarchitecture and utilizes the gigantic GV100 silicon. The GV100 die, which comes out of TSMC's 12nm oven, measures 815 mm² and bears 21,100 million transistors. The V100s continues to sport 5,120 CUDA cores and 640 Tensor cores. Nonetheless, the new V100s does pack more performance than both variants of the V100, which is probably attributed to higher core clock speeds. 

Nvidia Tesla V100s vs. Tesla V100 Specs

Swipe to scroll horizontally
Header Cell - Column 0 Tesla V100s PCIeTesla V100 SXM2Tesla V100 PCIe
Architecture (GPU)Volta (GV100)Volta (GV100)Volta (GV100)
CUDA Cores5,1205,1205,120
Tensor Cores640640640
Double-Precision Performance8.2 TFLOPs7.8 TFLOPs7 TFLOPs
Single-Precision Performance16.4 TFLOPs15.7 TFLOPs14 TFLOPs
Tensor Performance130 TFLOPs125 TFLOPs112 TFLOPs
Texture Units320320320
Interconnect Bandwidth32 GBps300 GBps32 GBps
Memory Capacity32GB HBM216GB / 32GB HBM216GB / 32GB HBM2
Memory Bus4,096-bit4,096-bit4,096-bit
Memory Bandwidth1,134 GBps900 GBps900 GBps
ROPs128128128
L2 Cache6MB6MB6MB
TDP250W300W250W
Transistor Count21.1 billion21.1 billion21.1 billion
Die Size815 mm²815 mm²815 mm²

The V100s delivers up to 17.1% higher single-and double-precision performance than the V100 with the same PCIe format. It also has 16.1% better Tensor performance. Nvidia has clocked the memory on the V100s a bit faster as well.

In terms of memory, it appears that Nvidia might only offer the V100s with 32GB of HBM2. It's unknown if the chipmaker will sell 16GB variants like it did with the V100. The V100s maintains the 4,096-bit memory interface but provides 26% more memory bandwidth in comparison to the V100.

Fortunately, the upgrades on the V100s have no effect on the graphics card's TDP (thermal design power). The V100s is still rated for 250W. As a result, the power requirements, a pair of 8-pin PCIe power connectors, remain the same as well. 

Nvidia didn't reveal the pricing for the V100s. The current 16GB and 32GB models of the V100 are selling for $5,855 and $7,200, respectively, on Amazon.

Zhiye Liu
News Editor and Memory Reviewer

Zhiye Liu is a news editor and memory reviewer at Tom’s Hardware. Although he loves everything that’s hardware, he has a soft spot for CPUs, GPUs, and RAM.