Chiphell leaks another apparent AMD "Rome" Cinebench score, this time with some proof

This is a photo taken by Gtx9 that shows Cinebench actually rendering the image with a 256 thread CPU. (Source: Chiphell)

About a month ago, Chiphell user Gtx9 leaked what he alleged was a Cinebench score of AMD's "Rome" server architecture. As Cinebench scores are easily manipulated, the weight behind Chiphell's leak was very weak. However, Gtx9 has returned with more convincing proof.

Matthew Connatser, Published 10/08/2018

The last time Chiphell's Gtx9 claimed he had a sample of AMD's "Rome" server CPU, we were skeptical. While Cinebench is a respected benchmark for testing CPU performance on both AMD and Intel platforms, results are very easily manipulated and can result in some odd rumors, such as an alleged 10 core 2800X. However, it's a different story this time, as there is actual proof that this new score might actually be legitimate.

Gtx9 showed three photos of "an unknown dual-channel 128-core 256-thread CPU" which is likely not one but two Rome CPUs; rumor has it that Rome will be 64 cores, and a 128 core CPU today is just absurd, so if true this is likely to be two CPUs. Not only did these photos contain a fully rendered image, a score of 12861 points (close to the previous score), and a CPU (or two more likely) whose name was redacted but count of 128 cores and 256 threads was not, but they also showed the benchmark in progress. Alone, a finished benchmark means very little with easily faked scores and Photoshop, but it is far harder to fake Cinebench rendering with 256 threads.

Each individual box in the render represents one thread. It is impossible to force Cinebench to show more threads than a CPU has, and it's not feasible to Photoshop such a render, since each box is tiny, is working on producing that part of the image (meaning if none of the boxes have an image, it looks fake), and is working on a specific part of the image. It is clearly visible that several boxes have partially rendered its section and that they are all sized the way one would expect with so many threads. One of the photos even seems to capture a camera effect called rolling shutter: examine in the second image how the opacity of the outer rendering squares increases as the image progresses towards the bottom.

Whatever the case may be, these scores are far more convincing than what came before. As always with rumors, never fully commit to them; these still could be fake. However, it is worth acknowledging that they could be real and that if they are, then this Cinebench score is the highest in the world. The previous record holder was 10038 points achieved by 4 Xeon Platinums.