We’ve been handled to a contemporary replace on higher-end Lovelace graphics playing cards from a typical supply of Nvidia leaks on Twitter, who equipped a couple of extra specifics, together with some eye-opening information at the RTX 4090.
This comes from Kopite7kimi, who gave us a complete bunch of main points on those incoming GPUs just a week and a part in the past.
Continue to replace. RTX 4090, AD102-300-A1, 16384FP32, 384bit 21Gbps 24G GDDR6X, 450W, base 2235 spice up 2520 precise max >2750; RTX 4080, AD103-300-A1, 10240FP32, 256bit 21Gbps 16G GDDR6X, 420W, RTX 4070, AD104-275-Kx(x is a bunch)-A1, 7168FP32, 160bit 18Gbps 10G GDDR6, 300W.July 4, 2022
As chances are you’ll believe, in this kind of quick house of time, numerous the purported specifications spilled have stayed the similar as within the earlier leak from Kopite7kimi. But there are a couple of key variations right here, significantly some nippy-sounding clock speeds for the flagship Lovelace graphics card (construction on what the leaker not too long ago hinted at).
The velocity that the RTX 4090 runs at is about to be a base clock of 2235MHz with a spice up to 2520MHz, even supposing the true most that overclocking may achieve is in far more than 2750MHz, the leaker believes (and it is only that – a trust in what they’ve been instructed by way of resources – so let’s now not learn an excessive amount of into any of this simply but).
Kopite7kimi additionally pins an influence utilization of 450W at the Lovelace flagship, which goes what we’ve heard up to now. The RTX 4090 will likely be a cut-down AD102 chip, in fact, and tool utilization might be driven so much additional (love to 600W) with the full-fat model (which may be a 4090 Ti, or perhaps even a brand new RTX Titan).
The information at the RTX 4080 is once more just about the similar tale, with the exception of now Kopite7kimi thinks that the 16GB of VRAM will likely be GDDR6X – the leaker wasn’t positive in this remaining time – and the prediction of that RAM being 18Gbps has been revised upwards to 21Gbps, in order that’s some excellent (theoretical) information.
Power utilization is pegged at 420W, which is once more repeating what the leaker mentioned a month in the past, albeit they added a query mark to that wattage again then – so this turns out to signify that Kopite7kimi is changing into extra assured that that is the extent the RTX 4080 might be pitched at.
Finally, in regards to the RTX 4070, there’s not anything new on the subject of the spec since Kopite7kimi’s remaining main replace. A 300W energy utilization is once more discussed, which has been the leaker’s best possible wager for a while now. Indeed, Kopite7kimi first floated that determine again in April, ahead of even having a lot of a clue concerning the RTX 4080’s energy calls for.
Analysis: A Lovelace flagship that’s two times as robust because the RTX 3090?
What’s relatively telling this is that one of the most energy utilization figures given seem to have been maintained as predictions for a while now, which implies that those would possibly simply be set in stone, because it had been. Or a minimum of they might now constitute a much more likely proper estimate of the place Nvidia will finally end up with TGPs for RTX 4000 graphics playing cards.
None of the information equipped right here is far of a wonder, then, and it’s extra about consistency than contemporary revelations – apart from the ones particular (purported) clock speeds for the RTX 4090. With 2750MHz plus being discussed, Nvidia appears to be having a look in opposition to the 3GHz mark for the flagship, versus the way more conservative clocks observed on current-gen Ampere (the 3090 has a spice up velocity of one.7GHz, as formally rated by way of Nvidia – even though it may, in fact, be driven sooner – or even the Ti model simplest hits 1.86GHz).
And with the RTX 4090 loaded up with CUDA Cores as it’s – 16,384 of them – coupled with that theoretical 50% building up in spice up velocity, this issues in opposition to a beautiful robust next-gen flagship. In reality, it might be two times as robust because the RTX 3090, which is a efficiency bounce we’ve observed predicted ahead of, as chances are you’ll recall.
While not anything is discussed about different clock speeds right here, having a look on the relative CUDA Core counts, and the RTX 4080 coming lovely on the subject of the RTX 4090 for TGP (420W as opposed to 450W), it’s simple to believe that the 4080 will push tougher nonetheless with its clock speeds.
It is sensible that Nvidia would wish to be riding as onerous as it may to get the most efficient good points imaginable from the nice drop to 5nm structure with Lovelace, and certainly with the flagship, there’ll be a variety of room to construct one thing even beefier than the preliminary 4090 (bear in mind, we’re having a look extra at 600W for the 4090 Ti, as in step with the rumor mill).
Finally, it’s value noting that even if it’s now not new information, there’s nonetheless sadness being aired on-line across the purported reminiscence loadout of the RTX 4070, and that it would simplest be 10GB of VRAM. Previous rumors urged a extra palatable 12GB, however it kind of feels that this configuration could be reserved for a 4070 Ti style – take all of this with the right serving to of salt, naturally.