Update: The chart and the text have been updated.
Nvidia’s next generation Ampere based gaming GPUs are shrouded in mystery, despite all the apparent leaks that have indicated memory capacity, core count, and process used. While I am personally skeptical of these leaks, I am not skeptical of this new leak, because it comes from two good sources: 3DMark and Rogame. Rogame has discovered a leaked Ampere benchmark on the Time Spy benchmark and has verified its authenticity. The Ampere GPU ran at a clock speed of about 1935 MHz and beat the 2080Ti by 30%.
For what is likely an early sample of whatever GPU this is (likely based on GA102), it’s a very impressive score. If the rumors are true that the RDNA2 powered Big Navi die is roughly twice the size as the 5700XT, and thus about twice as fast (enabled by RDNA2’s increased efficiency), then this Ampere GPU is just a little slower while running at an unusually low clock speed. That bodes very well for Nvidia, unless this isn’t a low clock speed at all. However, we have no reason to believe that clock speeds will regress this generation, even when Ampere for gaming is rumored to be on Samsung’s fabs, not TSMC’s.
Additionally, the memory could not be read properly and had an extremely high clock speed of 6000 MHz. Perhaps this is a new type of memory (GDDR6X is rumored by some but this is not an actual JEDEC specification for memory, unlike GDDR5X) or there’s some driver issue preventing it from being read properly. It would not be surprising if the memory needed an upgrade over the 2080Ti given the kind of performance we’re seeing here. Overall, this is a very interesting and very likely valid leak, but it still doesn’t totally pull the curtains away.