Best guess for NVidia's next architecture lovelace is September 2022...?
I don't even know at this point. The rumors for AMD are straightforward and from a previously reliable source. We already know they're doing two chiplet... I swear we need a new name for AI accelerators or something, anyway this year with CDNA. That they'd do the same for consumer GPUs just seems blindingly obvious; that they're sticking TSMC and going to 5nm is public; that they're improving hardware RT is an easy assumption backed up by patents.
But there seems to be no concrete details whatsoever for Nvidia, nor any sources that are nearly as reliable. What process are they going to use? How are they going to solve their bandwidth problem? What other improvements are there going to be? Until there's any actual details that make sense I'm not ready to believe any of these rumors; including on 30XX Super cards. So I guess any guess is as good as any other guess right now.
End of 22 is the expected timeframe for both though. And both should tape out about a year prior, so not sure what's there to pull forward or even how at this point.I think that article misses an important aspect of MLID's video: NVidia will pull Lovelace forward to match the release timeframe of RDNA 3. In other words 2022Q4 or 2023Q1 is looking unlikely for Lovelace.
Unless RDNA 3 slips. Which, honestly, I'm expecting. But there's no reason for NVidia to slip - Lovelace not being chiplet based (guess) would make it less likely to slip.
I think that article misses an important aspect of MLID's video: NVidia will pull Lovelace forward to match the release timeframe of RDNA 3. In other words 2022Q4 or 2023Q1 is looking unlikely for Lovelace.
Unless RDNA 3 slips. Which, honestly, I'm expecting. But there's no reason for NVidia to slip - Lovelace not being chiplet based (guess) would make it less likely to slip.
Nope.End of 22 is the expected timeframe for both though
Literally nothing is allowed to slip at AMD.Unless RDNA 3 slips. Which, honestly, I'm expecting
Nah.Pre-pandemic rdna3 was scheduled to be released in November of 2021
They don't.Also, how does NVidia pull Lovelace forward...?
RDNA 3 has never been rumoured for 2022Q4 or later.End of 22 is the expected timeframe for both though. And both should tape out about a year prior, so not sure what's there to pull forward or even how at this point.
Nope.RDNA 3 has never been rumoured for 2022Q4 or later.
Jeez why even try.MLID
Hooooooly shit not even H100 is GA by then.implying a shit-ton of RTX 4050/4060 (which would offer 3070/3080 performance) in 2022Q1 or 2022Q2.
Product schedules are fixed.may pull Lovelace forward by releasing lower-tier cards first
It doesn't actually say that's Q4 anywhere, just that it's between start of 2019 and end of 2022 + after RDNA2. Pretty sure at least most rumors are talking mid 22's, meaning Q2-Q3Yes:
It doesn't actually say that's Q4 anywhere, just that it's between start of 2019 and end of 2022 + after RDNA2. Pretty sure at least most rumors are talking mid 22's, meaning Q2-Q3
These twittos are late. I already mentioned weeks ago that Nvidia wont let AMD take the perf crown, whatever the cost. In other words, Hopper MCM will come to obliterate RDNA3 in gaming space if Lovelace is not enough...https://videocardz.com/newz/nvidia-...ies-design-allegedly-finalized-using-tsmc-5nm
The idea that there may not be AD102 and be GH202 in it's place don't make much sense to me tbh
How is this relevant to the die actual shipping?Navi 23 desktop missed 2021H1.
Nah, those are intentionally vague as fuck.Yes:
Product schedules are FIXED.ed: and I agree that now, AMD will push it back because they are going to continue to milk rdna2 and refine it
Cute LARP but that's not what is actually happening.Hopper MCM will come to obliterate RDNA3 in gaming space if Lovelace is not enough...
Nah, those are intentionally vague as fuck.
Product == die, yes it was "on time" (was it actually, were any laptops seen?)How is this relevant to the die actual shipping?
Yes they are.NO, they are not vague
That's what IHVs think about when they quote roadmaps.Product == die