Another 'leak' of the RTX 3080 Ti has arrived, this time from Chinese leaker "Big hardware player" who claims to have an engineering sample of a Colorful RTX 3080 Ti on hand. This is a theoretically upgraded and faster version of the Colorful iGame RTX 3080 Vulcan we just reviewed, which would be one of the best graphics cards if you could buy it, though it's more like the 3090 variant. The 3080 Ti should rank even higher on our GPU benchmarks hierarchy, but we're not convinced of the veracity of the images or the data, for a variety of reasons.
Before we go on any further, VideoCardz.com received a response from a Colorful rep regarding this leak, which says: "This 'rumor' doesn't even 'deserve' a statement...." Of course, any graphics card manufacturer would probably say something similar, whether a leak is real or fake. Still, there's a huge chance none of this is legitimate, and we'll get into the why in a moment. Basically, take the following details with an extra-large scoop of salt (as you always should with leaks).
The leaker shared GPU-Z results of this "RTX 3080 TI," but due to GPU-Z's unfamiliarity with an RTX 3080 Ti SKU, GPU-Z registers the card as a vanilla RTX 3080. However, the specs almost lineup with what the RTX 3080 Ti rumors had it to be, which is 20GB of VRAM, and the same amount of ROPs/TMUs as the 3090. The only minor difference is that this RTX 3080 Ti result shows that it has the same memory bandwidth as the vanilla RTX 3080. This implies the RTX 3080 Ti uses standard 19Gbps GDDR6X modules instead of the 19.5Gbps modules that were previously rumored for this GPU.
Let's note some problems with these GPU-Z screenshots, however. First, testing was apparently done months apart, based on the GPU-Z driver dates. Why would you run three different GPUs using three different drivers? Why use 456.16 on the 3080 and 456.71 on the 3090, if you're testing the cards right now? A lot of things have changed in the past four months!
Also, check out the "WHQL" drivers from December 3 being used on the 3080 Ti. And note how GPU-Z version 2.35.0 was used on the 3080 Ti, while 2.36.0 was used on the 3090 and 2.34.0 was used on the 3080. That would be the right version on the 3080 launch, but 2.36 didn't come out until November 27, while 2.35 was released on October 15. But let's continue.
The leaker supposedly threw this mysterious GPU onto a test bench and ran a bevy of gaming benchmarks and 3DMark. On average, the "RTX 3080 Ti" was 3% slower than the RTX 3090 (which he also benchmarked himself) in the gaming tests. In 3DMark, the leaked 3080 Ti was anywhere between 2-8% slower than the RTX 3090.
Overall, these benchmark results for the leaked engineering sample 3080 Ti seem very suspicious. In the gaming benchmarks, the testbed's CPU is clocked at different frequencies for different cards, and the RTX 3080 shows just a 3% average performance deficit compared to the RTX 3090. This despite the "3080 Ti" having a 23% performance deficit in memory bandwidth. Modern GPU architectures love memory bandwidth and will scale well with it, so all of this seems extremely specious.
There are plenty of other red flags as well. The Witcher 3 has a severely blurry game image but legible text, and Battlefield V — a game with no built-in benchmark, so any test sequence would involve restarting from a checkpoint, and the non-deterministic nature means enemies, flags, etc. would vary quite a bit between runs — looks to have nearly the same background on all three test PCs. Also, let's not get started on the 'leak' PC having the same clocks, voltages, and temperatures across all 16 cores. 24C in Shadow of the Tomb Raider? That's unbelievably impressive!
As in, we literally don't believe it. If the leaker actually has the cards in question and ran the benchmarks, then at the very least, the testing methodology is clearly flawed. Different drivers, different CPU clocks and settings, different versions of GPU-Z, and probably plenty of other discrepancies as well. But hey, RTX 3080 Ti is likely going to be a real card at some point. As real as any of the other recent GPUs that are virtually impossible to find, anyway.
Stay on the Cutting Edge
Join the experts who read Tom's Hardware for the inside track on enthusiast PC tech news — and have for over 25 years. We'll send breaking news and in-depth reviews of CPUs, GPUs, AI, maker hardware and more straight to your inbox.
Aaron Klotz is a freelance writer for Tom’s Hardware US, covering news topics related to computer hardware such as CPUs, and graphics cards.
It's not like we're going to be able to buy one anyways.Reply
The statement "the 'leak' PC having the same clocks, voltages, and temperatures across all 16 cores. 24C in Shadow of the Tomb Raider" is not true at all if you look at the original source. The CPU clocks, and voltages are locked for testing purpose, while the temperatures did vary during the testing.Reply
Yet another disappointing launch of a card that's inaccessible to the 90%Reply
Nope, this is most honest leak ever, it is as fake as availability of new graphic cards anywhere remotely close to reasonable price. :-DReply
Jokes aside, it is not like nVidia has any incentive to make better cards when they basically immediately sell anything they can make. Good problem for nVidia yo have, but bad for consumers who can't get new cards.
Two things.sw269 said:The statement "the 'leak' PC having the same clocks, voltages, and temperatures across all 16 cores. 24C in Shadow of the Tomb Raider" is not true at all if you look at the original source. The CPU clocks, and voltages are locked for testing purpose, while the temperatures did vary during the testing.
You're assuming the card was actually tested, which as we've noted in the text is highly unlikely given the number of discrepancies in the data and screenshots provided.
If the cards were tested, they should all be tested the same way, and screenshots of performance should be comparable. How do you test cards at the same settings (locked clocks, etc.) and then capture a bogus screenshot?
I can see from the images that temperatures "varied," but how exactly was a benchmark run where CPU temps were 24C across all cores for a screenshot? Answer: Either it wasn't run, or the screenshot was doctored. And if it's the latter, then what else was doctored?