Leaked RTX 4080 12GB Benchmarks Seemingly Justify Nvidia Retreat
Performance only matches the RTX 3080 12GB
According to a post in the Chiphell forums, one of its users managed to benchmark Nvidia's now-cancelled RTX 4080 12GB graphics cards. Unfortunately, performance was very poor for an RTX 4080-class GPU, with 3DMak Fire Strike and TimeSpy scores that only matched Nvidia's Ampere-based RTX 3080 12GB. That performance is disappointing, considering that the RTX 4080 12GB was technically supposed to replace the RTX 3080 12GB.
The original source never disclosed the exact RTX 4080 12GB model that was tested. However, the poster still showed us the complete GPU monitoring data, including core clocks, thermals and power, so we have a good idea of how these cards behave.
To refresh your memory, the RTX 4080 12GB was Nvidia's lowest-end Ada Lovelace GPU originally announced during GTC, alongside the 4080 16B and 4090. The 4080 12GB was armed with an AD103 die, running on 9728 CUDA cores, a 256-bit wide bus and 22.6Gbps of GDDR6X memory, for a total of 717GB/s of bandwidth.
But after severe criticism by the community, Nvidia canceled the RTX 4080 12GB model due to its very confusing name compared to the RTX 4080 16GB. The 4080 12GB features significantly altered core specifications compared to the 16GB version despite the names only suggesting there would be a capacity difference between the cards.
However, Nvidia's "unlaunch" came after RTX 4080 12GB cards started rolling off production lines; this is how the Chiphell user managed to snag a card.
According to the GPU-Z image from the Chiphell forum post, this unnamed RTX 4080 12GB model was operating at a maximum real-world boost clock of 2,820 MHz, and a 1,313 MHz memory clock. Thermals were in the 72.3C range at their max, with hot spot temps hitting 91.2C. Fan speeds at their max were at 43%, but the RPM is unknown. Power consumption can be seen hitting 261W at its peak.
This monitoring data proves to us that the RTX 4080 12GB was severely cut down compared to the RTX 4080 16GB and RTX 4090. The clocks remain similar to the 4090, but power consumption is substantially reduced to just 262W. The card was not power throttling due to a power limit, as demonstrated by GPU-Z's peak TDP of just 91.8% of the total power target.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
But the 4080 12GB's performance in 3DMark says it all. According to the Chiphell post, the RTX 4080 12GB tested managed a GPU score of just 13,472 points in 3DMark Fire Strike Ultra and 10,794 points in TimeSpy Extreme.
For comparison, these numbers translate into equal to slightly better performance than an RTX 3080 12GB in the exact same benchmarks, according to 3DMark's benchmark browser. It's worth noting that the 30 series GPUs in the browser's results were manually overclocked, but still, that means the RTX 4080 12GB sits right in between a stock RTX 3080 12GB and RTX 3080 Ti.
It's easy to see why Nvidia cancelled the RTX 4080 12GB when it did, the RTX 4080 12GB originally had an MSRP of $900 when announced. But with the performance numbers we see here, the GPU wouldn't even be an improvement over Nvidia's older RTX 3080 12GB or RTX 3080 TI which can be found for well under $900 right now.
Aaron Klotz is a contributing writer for Tom’s Hardware, covering news related to computer hardware such as CPUs, and graphics cards.
-
TCA_ChinChin Don't know how reputable that source is but if it is true, then the RTX-4080 12GB (really the RTX-4070) is sorely disappointing.Reply -
helper800
Die size comparison really makes the 4080 12GB a 4060 tier card IMO.TCA_ChinChin said:Don't know how reputable that source is but if it is true, then the RTX-4080 12GB (really the RTX-4070) is sorely disappointing. -
trance77 I would be surprised if that is true considering the performance of the 4090 being 50-77% higher than the 3090 TI.Reply -
pclaughton
But this is akin to saying that you a doubt a Nissan Frontier review because the GT-R goes much faster. Nvidia crippled the "4080 12GB" to the point where these results are completely believable if still worth a grain of salt.trance77 said:I would be surprised if that is true considering the performance of the 4090 being 50-77% higher than the 3090 TI. -
AgentBirdnest If you look at the average scores, instead of the high scores, this "4080 12GB" result is 15% faster than an average 3080 12GB, and 5-7% faster than the average 3090. 4-5% worse than the average 3090 Ti.Reply -
btmedic04 Sigh, at least get the specs right Aaron. The 4080 12gb used AD104 and had a 192 bit memory bus whereas the 4080 16gb uses AD103 and a 256 bit memory busReply -
trance77 pclaughton said:But this is akin to saying that you a doubt a Nissan Frontier review because the GT-R goes much faster. Nvidia crippled the "4080 12GB" to the point where these results are completely believable if still worth a grain of salt.
Not really, if the 4090 has such a big increase in performance from the last generation you would expect the other cards to also have a big increase, not necessarily the same but still significant. Previous example would be a 3060 TI being equal to a 2080 Super. I'm not saying its not true but would be surprising and also they will be in big trouble. -
cfbcfb trance77 said:I would be surprised if that is true considering the performance of the 4090 being 50-77% higher than the 3090 TI.
Pretty sure that the real world #'s have turned out to show that Nvidia may have been leaning on the scale with DLSS for a lot of the 4090 performance claims?
I think the real numbers are around 25% to 50%, with the wind blowing in the right direction?
In any case, the author is correct. 3080ti deals are to be had. So I bought one yesterday for a bit over $700 new. Might be lacking in some ray tracing performance vs the 4xxx series at the same performance/$ level, but I'd have to wait on the 4080 to come down or the 4070 to be a better performer than the 4080 12GB? Which it won't?
I'll check back in with the 5000 series at launch. See if they've figured out by then to not try and save $10 on a cable for a $1600 video card? -
trance77 cfbcfb said:Pretty sure that the real world #'s have turned out to show that Nvidia may have been leaning on the scale with DLSS for a lot of the 4090 performance claims?
I think the real numbers are around 25% to 50%, with the wind blowing in the right direction?
In any case, the author is correct. 3080ti deals are to be had. So I bought one yesterday for a bit over $700 new. Might be lacking in some ray tracing performance vs the 4xxx series at the same performance/$ level, but I'd have to wait on the 4080 to come down or the 4070 to be a better performer than the 4080 12GB? Which it won't?
I'll check back in with the 5000 series at launch. See if they've figured out by then to not try and save $10 on a cable for a $1600 video card?
The 50-77% I quoted was from the review Jayztwocents did and DLSS was turned off. Also similar from Tomshardware review -
'The RTX 4090 and Ada Lovelace are, frankly, impressive as hell. From a performance and technology perspective, Nvidia has pushed things further than we've likely ever seen between GPU architectures. In our testing, we saw performance improvements of over 50% at 4K ultra, and a 78% increase in ray tracing heavy games. Toss in DLSS and DLSS 3 Frame Generation and the potential gains are even more impressive.'
That's why I am doubting this report but of course it could be true.
Anyway that looks like a good deal for a 3080TI and you can get on with gaming. The price increases are also extreme, in the UK the 3080 FE was £649 retail at launch and the 4080 FE is £1269! -
The Historical Fidelity To the Author: your rtx 4080 12GB core specs are incorrect, you are using the 16 GB core count.Reply