NVIDIA RTX 3080 Time Spy Benchmark Leaked – Blows Away The RTX 2080 Ti
An unknown NVIDIA Ampere GPU, likely the RTX 3080, has had its 3DMark Time spy performance leaked courtesy of HardwareLeaks. While the benchmark could be either of the RTX 3080, 3080 Ti/Super or the RTX 3090, we have reason to believe it is the RTX 3080. The gaming performance shown here absolutely blows away the last generation flagship and showcases that NVIDIA is going to have a spectacular RTX 3000 series launch later this year.
NVIDIA RTX 3080 gets Time Spy benchmark score - 31% faster than RTX 2080 Ti FE
The benchmark posted by _rogame shows NVIDIA's Ampere GPU scoring a very impressive 18257 points in Time Spy graphics test which is just under Kingpin's insane 2.38 GHz clocked RTX 2080 Ti variant. A traditional RTX 2080 TI FE variant clocks around 13939 points and the spotted Ampere GPU is roughly 31% faster than this. The question then becomes what exactly are we looking at? The NVIDIA RTX 3080, RTX 3080 Ti/Super or the RTX 3090? Well, I am fairly certain this isn't the RTX 3090/Titan because of several reasons (although I am not a 100% sure, just to be clear).
Before we get into why I think that way, let's look at the logged specifications of the benchmark. The GPU clock was detected at 1935 MHz and the memory clock, interestingly was only a meager 6000 Mhz. This is almost certainly a misread by the software due to the fact the card is using brand new GDDR6X memory (as memory clocks shown usually need to be multiplied by 8 to get the actual speeds but this is obviously wrong). This is indicative of an early engineering sample and performance as well as detection will likely improve as things go closer to launch.
So let's put on our [speculation] hats now and try to figure out what card we are looking at. NVIDIA is planning to launch three variants of the GA102 chip at launch. These are the SKU 10, 20 and 30. These will have 10, 12 and 24 GB of memory respectively and will likely form the RTX 3080, RTX 3080 Ti/Super/3090, and RTX 3090 / Titan cards. Considering NVIDIA is moving to a smaller process with the Ampere series of cards, we are expecting a big performance jump (from the improved clock speeds, core count and of course, IPC).
So here is what I did. I took the scores of the RTX 2080 Ti FE and Lightning Z and calculated a "3DMark Time Spy" variable that could roughly predict their score. You do this by taking their known stable boost clocks (as NVIDIA cards usually stay much higher than their advertised boost speeds) and forming the following equation = Core*CLock*3DMark Variable = Score. With the FE at a known stable boost of 1860MHz and Lightning Z at a known stable boost clock of 1995 Mhz, we can solve for the variable and get the following result = 0.001729.
Now if you assume that nothing changed IPC wise between this generation and that, you can solve backward for the core count using the clock speed and score and will arrive at the following estimated core count: 5455. You can also add in an IPC improvement of 20% to get a core count of 4546. Interestingly both of these rough core counts are close to leaked configurations of the GA102 with the SKU 10, 20 and 30 being 4352, 5248 and 5376 respectively. Considering there will almost certainly be *some* IPC improvement with Ampere, I am almost certain this is not SKU 30 or the top end card. In my opinion, what you are looking at, is the RTX 3080 or the RTX 3080 TI/Super/3090 mainstream flagship and not the 3090/Titan HEDT flagship. In other words, in order of probability - this is SKU10 or at a stretch SKU20, and probably not SKU30. [/spectulation]