DarthShader
Regular
How come there is no next generation Nvidia speculation?
- Thread starter Mendel
- Start date
and 192 bit at that then? (6 mem chips.) How would Fermi's 192CC line-up against GT200's 192CC (GTX260) part? It would be interesting to get some shader performance comparisons (assuming texturing is underpowered on GF106)
dominikbehr
Newcomer
4. Alessandro Volta http://en.wikipedia.org/wiki/Alessandro_Volta
5. Anders Celsius http://en.wikipedia.org/wiki/Anders_Celsius
Celsius would just be too tempting!
Kelvin GTX580. That will actually be it's running temperature.
But seriously guys, if NV had originally launched Fermi in October last year, and we've already had the GF104 rework ~9 months after it's planned launch, we should be expecting a real refresh in the coming 4 months or so.
Last edited by a moderator:
And for a reason. Much more eneven frametime distribution males your average Fraps run or integrated benchmark basically worthless. Plus much more pronounced dependency on driver profiles and other stuff that just putting up some bars on your website doesn't cut it for comparison.
How is a fully enabled GF100 the next generation of a partially enabled GF100?
When it's launched as a GTX580
I wouldn't be surprised at all if Nvidia keeps the same 40nm chips or slight revisions of them and launches full versions as the 500 series. The only problem with that is it means those full versions problably won't show up till sometime mid next year.
I was talking about the HD 6000s, my point being that even if the benchmarks turn out to be faked, we should still expect a significant performance improvement, because that's typically what happens when moving from one generation to the next.
That would be incredibly disappointing, and probably not terribly competitive w/ATi either. At this point I'm hoping more for a new chip based on GF104's architecture. GF100 can only get incrementally faster, a scaled up GF104 at about the same die size as GF100 could bring a good deal more compute resources than GF100.
More computation resources how? I assume you mean the extra ALUs per "cuda core" (terrible marketing name), but from what people posted in the other thread, the actual worth of those units seem marginal at best since the chip seems bottlenecked internally elsewhere...
Scaling up the GF104 would mean dropping the ECC support and all the other high-end features to make room for the extra shader units, not sure how Nvidia would sit with that, seeing as they like to re-use the exact same chips both in high-end consumer and high-end professional markets; after all they've been doing that ever since the original Geforce, or maybe it was Geforce 2, I can't remember since it was so long ago now. Now that they've started on the ECC path with GF100, taking a step back would lose them face big-time.
Scaling up the GF104 would mean dropping the ECC support and all the other high-end features to make room for the extra shader units, not sure how Nvidia would sit with that, seeing as they like to re-use the exact same chips both in high-end consumer and high-end professional markets; after all they've been doing that ever since the original Geforce, or maybe it was Geforce 2, I can't remember since it was so long ago now. Now that they've started on the ECC path with GF100, taking a step back would lose them face big-time.
The latest forcewares show some 12 series mobile parts, the just released GTX460M was an 11 series:
N11E-GS = GTX460M
ForceWare 257.38:
[DEV_0A79&SUBSYS_14931028] NVIDIA N12M-NS-S
[DEV_0A79&SUBSYS_14941028] NVIDIA N12M-NS-S
I, Personally like real part numbers, codenames, code-code-codenumbernames. Whistler XT en Blackcomb XT suddenly make more sense to me.
N11E-GS = GTX460M
ForceWare 257.38:
[DEV_0A79&SUBSYS_14931028] NVIDIA N12M-NS-S
[DEV_0A79&SUBSYS_14941028] NVIDIA N12M-NS-S
I, Personally like real part numbers, codenames, code-code-codenumbernames. Whistler XT en Blackcomb XT suddenly make more sense to me.
GF119 on 40nm to test the architecture before 28nm? You'd expect GF119 to have taped-out by now if it's on 40nm but not to be shipping for some time - we did see some G9x in drivers a *lot* before release. I don't think we've seen GF108 in any drivers yet, but surely that can't be N12.
The thing is, N11E-GS was already announced by IBM back in march in a design document and it might even already have had driver support before that. but the N indicates that it's a Notebook part.
Well you'd expect NV to focus on the notebook SKU first in a low-end part as the lead times are longer. Strangely enough I'd expect it to be too late for Q1 notebook refreshes, but who knows. Is there something else you were implying?
Actually that they might be on time with this.
If they have it sampling now it might get some wins for Sandy Bridge Mobile, who knows.
Hah. Well that would definitely put into question how it will co-exist, if at all, with GF108... Or are we looking at a 128-bit DDR3 part here?Actually that they might be on time with this.
If they have it sampling now it might get some wins for Sandy Bridge Mobile, who knows.
What do you want to speculate about it if pretty much everything is known about it anyway, except the release date?
Assuming full-chip 750/1500/1000Mhz (or so) configuration, should be slightly faster than HD5850, very close to GTX470 (to be named GTX475 imho it would need a bit higher core clock like at least 800Mhz). Power draw above HD5850 but still below HD5870 and GTX470.
The only interesting part about that card is WHY it hasn't been released, and there are still only 2 options which were discussed a lot already:
a) nvidia can't produce full GF104 chips in large enough quantities. Seems somewhat unlikely to me (unlike GF100, so far all parts only have one simd disabled (plus some have rops disabled, but I think that's got more to do with cost savings for ram chips), and they have no issues with clocks whatsoever neither).
b) nvidia is unwilling to remove GTX470 from its lineup. Possibly because this would only leave GTX480 chips in its GF100 consumer lineups, which supposedly not many chips qualify for. I have no idea though how many GTX480 cards are sold vs. quadro (or tesla) cards, since any chip not meeting GTX480 qualification could still be used there (some of those workstation cards there pretty much use half the chip only - at half the clock...). Might be possible though there are more GF100 chips sold as GTX470 currently than all other combined.
There could be other reasons (like holding it back until AMD releases HD67xx so they have at least something new) or not really wanting to compete with HD5850 but I'm not sure these make a lot of sense.