DegustatoR
Legend
Kinda expected so not too sure if it's legit or someone is just making an obvious guess.
isn´t it too (s)low for mighty Nvidia ?
DegustatoR
Legend
It's a 30% bump from 4090 which would be impressive on the same process and a more complex chip.
D
Deleted member 2197
Guest
People noted if the rumor is true the chip sounds somewhat special and would get pretty insane boost clocks. (compared to the pretty high stock and OC boost clocks in Lovelace)
DegustatoR
Legend
Not necessarily. We don't know if Blackwell will have the same boost behavior.
If it will and this rumor is true though then yeah, sustained boost clocks should be around 3.6Ghz.
If it will and this rumor is true though then yeah, sustained boost clocks should be around 3.6Ghz.
How did you get to 3.6? The 4090 tops out at around 2.7.
And its base is 2.235 GHz. Applying a similar ratio for a base of 2.9 GHz I get 3.5 rather than 3.6, but still much higher than Ada.
With the obvious caveat that two huge assumptions are being made there.
DegustatoR
Legend
Yeah let’s try to avoid using souped up AIB numbers when speculating on upcoming hardware.
Ah that makes sense. I mistakenly used 2.32 and not 2.23 for base. Either way there’s no guarantee the boost algo works off percentages that way and certainly no guarantee Blackwell uses the same ratios. Anything around 3.5 will be pretty impressive.
Ah that makes sense. I mistakenly used 2.32 and not 2.23 for base. Either way there’s no guarantee the boost algo works off percentages that way and certainly no guarantee Blackwell uses the same ratios. Anything around 3.5 will be pretty impressive.
DegustatoR
Legend
These two cards have the exact same base clock so there's nothing "souped up" there aside from power limit.
Lovelace seem to top out at ~2.9 across all chips really. You need to do w/c or LHC to push it higher.
Point is that any leaks are likely based on FE specs and power.
Frenetic Pony
Veteran
Edit - I can see boost clock for the highest clocking GPUs of this arch, that's a reasonable target. That the highest tier consumer part would also target the highest tier boost clock is questionable, those parts are usually power limited and I see every reason to assume that's the case here.
Last edited:
DegustatoR
Legend
Which if the earlier leaked unit numbers are true seem like the most likely option of achieving this target aside from some serious internal re-architecture.
yeah, they just add second 16 pin power conector to prevent melting
DegustatoR
Legend
The cooler will be smaller than on 4090 according to rumors ¯\_(ツ)_/¯
D
Deleted member 2197
Guest
Nvidia RTX 50 graphics card family TDPs 'leaked' by Seasonic
RTX 5090 buyers may need 500W just to cover GPU power consumption.
DegustatoR
Legend
I really really doubt that even Nvidia knows what TDP a 5060 will have so these are likely just placeholders on part of Seasonic.
TopSpoiler
Regular
Without a proper node jump, I'd say reasonable expectations should be more in the 35-40% increase range.
Looking back to the last two times Nvidia stuck to the same (general) process between generations, we have:
780Ti -> 980Ti being about a 40% increase in performance at 1440p, increasing to 45% if we want to talk about the Titan X. This was overall considered a really excellent result. Die size increased from 561mm² to 601mm²(7% increase).
1080Ti -> 2080Ti was a 33% increase in performance, actually decreasing slightly if we compare against Pascal Titan X/XP. And given the incredible increase in die size from 471mm² to 754mm²(60% bigger), this was widely considered a very disappointing result. Granted, Turing used a chunk of the new silicon for all new features that wouldn't be widely used yet, but this in no way felt like a worthy generational leap regardless.
So pretty different situations, but it shows that even Nvidia at its best was not capable of providing a Pascal/Lovelace-esque improvement without a proper node leap. I would hope that we'll get better than what we got with Turing, but we really dont know. I just dont think the ceiling for performance improvement is really as great as such a rumor is suggesting. That would be almost miraculous, even going with some 700mm²+ behemoth.
Looking back to the last two times Nvidia stuck to the same (general) process between generations, we have:
780Ti -> 980Ti being about a 40% increase in performance at 1440p, increasing to 45% if we want to talk about the Titan X. This was overall considered a really excellent result. Die size increased from 561mm² to 601mm²(7% increase).
1080Ti -> 2080Ti was a 33% increase in performance, actually decreasing slightly if we compare against Pascal Titan X/XP. And given the incredible increase in die size from 471mm² to 754mm²(60% bigger), this was widely considered a very disappointing result. Granted, Turing used a chunk of the new silicon for all new features that wouldn't be widely used yet, but this in no way felt like a worthy generational leap regardless.
So pretty different situations, but it shows that even Nvidia at its best was not capable of providing a Pascal/Lovelace-esque improvement without a proper node leap. I would hope that we'll get better than what we got with Turing, but we really dont know. I just dont think the ceiling for performance improvement is really as great as such a rumor is suggesting. That would be almost miraculous, even going with some 700mm²+ behemoth.