Mystery NVIDIA GPU With 124 Compute Units Spotted Online

A rather odd but surprising NVIDIA Graphics device has surfaced online. The mystery NVIDIA GPU follows the unusual pattern of very high Compute Units but rather low GPU Clock Speeds. Based on the limited amount of information available, it is quite possible that NVIDIA is designing and testing the next generation of Tesla or even Quadro GPUs, meant for High-Performance Computing (HPC) segments.

A recent leak from Geekbench indicated there were two NVIDIA graphics devices equipped with 118 and 104 Compute Units. The latest leak suggests there’s even a third variant within the same pipeline. The mystery NVIDIA GPU packs 124 Compute Units, making it one of the largest GPUs. The GPU is merely mentioned in a leaked database, and hence the specifications could change significantly.

NVIDIA Readying High-End GPUs For Professional Workstations And Server-Side Applications?

There are currently three mystery NVIDIA Graphics devices, equipped with an increasing number of Compute Units. After the GPUs with 104 and 118 Compute Units leaked online recently, a new one with 124 CUs has appeared online. The number appears slightly odd, and hence, it is possible the finalized specifications of the Big Core might pack a total of 128 CUs.

Assuming 64 Streaming Multiprocessors (SMs), the mystery upcoming high-end NVIDIA GPU might pack a total of 8192 CUDA Cores. Needless to add, these are exceptionally high numbers, and hence, NIVIDIA might not be able to achieve commercial-level scalability. In other words, due to manufacturing yield limitations, such silicon might not be available for the market with full specifications.

The most glaring example is of the NVIDIA Quadro GV100 Volta GPU and the NVIDIA Tesla V100 GPU Accelerator. The Quadro GV100 Volta GPU featured a total of 84 Streaming Multiprocessors. However, in realistic manufacturing settings, the GV100 GPU was available with only 80 Volta Streaming Multiprocessors enabled.

Experts are optimistic about the new mystery NVIDIA Graphics device with 124 CU (SM). Some claim NVIDIA might be testing a new and unannounced next-generation Tesla GPU. However, the device is clearly lacking just as many SMs as Tesla Volta.

The benchmarks do not offer many details, but based on the information available, the new NVIDIA GPU is aimed at the high-end workstation and data center. In other words, the unreleased NVIDIA Graphics device won’t be optimized or even intended for the gaming hardware. The graphics cards are allegedly based on NVIDIA’s upcoming Ampere microarchitecture using TSMC as well as Samsung’s 7nm production facilities.

Mystery NVIDIA GPU Specifications And Features:

According to the Geekbench 5 report, there’s 32MB of L2 cache on the graphics card. The graphics card finished the Geekbench 5 benchmark with a 1.11 GHz Base Clock. It is important to note that the GPU was most likely an early engineering sample. Still, data center GPUs usually don’t operate at high clock speeds anyways.

[Image credit: Primate Labs Inc.]
The graphics card was also listed with 32GB of ECC memory. The memory reportedly runs at 1.20 GHz across a 4,096-bit memory interface. This clearly indicates the new device is packing HBM2 memory modules. The testbench comprised of an Intel Core i7-8700K 3.71 GHz 6C/12T CPU, and an Asus Prime Z370-A motherboard.


Tags

Alap Naik Desai


A B.Tech Plastics (UDCT) and a Windows enthusiast. Optimizing the OS, exploring software, searching and deploying solutions to strange and weird issues is Alap's main interest.
Close