Two Leaked TESLA-grade GPUs Continue To Assert Nvidia’s Dominance
Noteworthy news comes from two sources, either directly delivered by official channels, or by leaks. At the end of February, an attentive observer of the industry, going by the Twitter handle @_rogame, discovered not one but two possible new Nvidia GPUs.
To clear the air first, if you are not an engineer, a scientist, or otherwise preoccupied with big data computing, you should rest at ease that your expensive, top of the line RTX 2080 Ti won’t be overshadowed by a newer version.
Leaked TESLA-Grade Nvidia GPUs
Both of the unknown Nvidia GPUs fall under the TESLA category of Nvidia GPUs. You can find them in supercomputer clusters and workstations for all-purpose parallel computing. The hefty price tag mirrors their purpose acutely. The basic version, Nvidia Tesla v100 16GB, would drain you of about $6000 USD. One could easily build three high-end PCs for that amount!
However, even if for a specialized purpose, the new tech reveals what leaps in performance can we expect in the future, as it trickles down to consumer-grade GPUs.
The lesser of the two has a GB5 Compute score of 141654 (Open CL):
- 6912 CUDA cores (108 CUs)
- 1.01GHz core clock
- 47GB of memory
The more powerful one is so by a wide margin, achieving a GB5 Compute score of 184096 (Open CL)
- 7552 CUDA cores (118 CUs)
- 1.11GHz core clock
- 24GB of memory
What Does CU Mean?
If you are wondering, CU just means Compute Units, which we get when we divide CUDA cores by 64.
To put things into perspective, the currently most powerful consumer-grade GPU, RTX 2080 Ti, achieves the same test score of only 128805, with 4352 CUDA Cores (68 CUs).
Still, given the price tag that is over five times lighter, that’s an outstanding GPU power to have.
When it comes to the current stock of TESLA GPUs, the Tesla V100 16GB scores at 139921, and Tesla V100 32GB at 153741, which means that the lesser of the two leaked cards sits right in the middle.
You may have noticed that they both clock at about 1GHz, which is much lesser than the base clock of RTX 2080 Ti at 1350 MHz, not to mention the much higher boost clocks of its various versions. This indicates a massive spike in power for the next reiteration of consumer-grade GPUs.
Unless AMD comes up with something truly ground-breaking like they managed to accomplish with the Ryzen series of CPUs, it seems that Nvidia will yet again end up competing with their own GPUs.
This is not good news for the end consumer one bit, as there will be no incentive for Nvidia to lower their prices. Departing from the Turing line, we will likely see the consumer version of this tech under the name Ampere. If not canceled due to the Corona pandemic, you can expect the official info at NVIDIA’s GPU Technology Conference (GTC), from March 23 – 26.