Next-gen Nvidia Ampere gaming GPUs are one of the most anticipated pieces of hardware in the PC and gaming scene this year. Over the course of the last few months, a number of leaks surfaced detailing various aspects of the much anticipated gaming GPUs, ranging from actual specifications to leaked pictures of Founders Edition coolers.
Most of the mentioned leaks point to GA102 being the die of choice for Ampere’s top-end of the stack which includes at least 3 cut-down versions :
RTX 3090 / Titan
RTX 3080 Ti / Super
Yesterday, I was able to confirm the existence of 3 GA102 variants that are being tested, each with a different memory configuration :
GA102 with 24GB of VRAM
GA102 with 12GB of VRAM
GA102 with 10GB of VRAM
These variants lineup perfectly with the previous leaks but, sadly, we still don’t have any idea about the targeted performance of these GPUs/variants. Today, I’d like to change that.
I’m sure some of you have questions about that memory clock. It doesn’t match any of the usual known memory clocks for GDDR6 of HBM2. For context :
14Gbps GDDR6 is reported as 1750MHz
16Gbps GDDR6 is reported as 2000MHz
2GT/s HBM2 is reported as 1000MHz
From my experience, this type of weird reported clocks is either the result of an early/internal driver or a new type of memory that is still unknown to 3dmark detection software.
Next is actual performance numbers. I compiled a chart of time spy graphics scores that might help us better frame the result in question.
So basically, this unknown Ampere variant is :
30.98% better than a stock RTX 2080 Ti Founders Edition
21.07% better than a stock MSI RTX 2080 Ti Lightning Z
22.14% better than a stock Nvidia Titan RTX
8.30% better than the best Nvidia Titan V result under LN2
2.18% less than KINGPIN’s overclocked EVGA RTX 2080 Ti XC
These are some very impressive results given the lower than expected clock speed.
For those who are still skeptical about the authenticity of this result, I was able to 100% verify that this test was done by an Nvidia employee who’s name shall remain anonymous for obvious reasons.
Lower than expected clock speed
Putting the weird memory clock aside, 1935MHz boost clock is not what a lot of people were expecting. It’s not that far off current 12nm Turing clock speeds and only raises more questions than it answers. Are Samsung 8nm rumors true? Is the GA102 die already so large that clocks can’t be pushed much further? Or perhaps are Nvidia aiming for an efficiency improvement first and foremost?
Let’ hope these clocks are only for an early sample and that final clocks will be much higher.
Nvidia might have some juicy hardware lined-up for its next GPU launch. Delivering 31% more performance over RTX 2080 Ti FE is nothing to sneeze at, especially if the tested GPU is not the top GA102 variant.
One thing is for sure, between this and AMD’s RDNA2, gamers and enthusiasts around the world are about to witness one of the most competitive GPU launches in the last few years.