As the days go on, we seem to be getting more and more rumors about Nvidia’s upcoming Ampere line of graphics cards. Not so long ago we saw images of leaked designs and leaked specifications. But perhaps the most exciting news is that we are now starting to see potential benchmarks crop up.
What Did The Leaks Show?
Rogame goes on to say that these three GA102 variants have been tested and each has a different memory configuration. These configurations are a GA102-400 with 24 GB of RAM, a GA102-300 with 12 GB of RAM, and a GA102-200 with 10 GB of RAM.
The next piece of interesting news is a graph produced by Rogame that shows the performance numbers of this unknown Ampere GPU compared to a few other graphics cards. This was found on a private listing that hasn’t been shared, so we only have this graph to go off for now.
For those who are looking to upgrade to an Ampere GPU when they release, this graph is a very promising piece of information. We can see some monstrous performance from this card - and we don’t even know what it is yet! Here’s to hoping it’s one of the lower models and there’s even more performance to come.
For those who don’t want to do the math, Rogame has included an interesting comparison at the bottom of his article. Overall, it seems this Ampere variant is around 31% better than a stock RTX 2080 Ti Founders Edition, 21% better than a stock MSI RTX 2080 Ti Lightning Z and 22% better than a stock Nvidia Titan RTX.
Incorrect Memory Clock Speeds?
When looking at the memory clock speeds for this benchmarked variant, some interesting points come up - the main one being the fact that the clock numbers don’t quite make sense. A private Time Spy result for this card showed some interesting clock numbers.
- GPU vendor: Nvidia Corporation
- GPU core clock : 1,935 MHz (Boost Clock)
- Memory clock: 6,000 MHz
With the memory clock sitting at 6,000 MHz, it doesn’t seem to match any of the known memory clocks for GDDR6 / HBM2 memory, which would be 14 Gbps - 1,750 MHz, 16 Gbps - 2,000 MHz and 2 GT/s - 1,000 MHz.
This could be explained by an early version of a driver being used or even just the fact that a new type of memory isn’t yet supported in the benchmarking software.
Putting this inconsistency aside, the 1,935 MHz boost clock is also a lot lower than a lot of people would be expecting. We don’t know whether this is due to the current GA102 die already being pretty large or maybe even Nvidia are focusing on efficiency this time around.
With this being a very early benchmark, a lot of these figures can (and probably will) change. We’ll make sure to keep an eye out for any more leaks and benchmarks, but for now, be sure to check back often for all the latest news.