Do you mean NV35? We already had previews on sample silicon by August 2002 of the NV30. (Unless you mean final tapeout, and not initial)Uttar said:
NV36 at IBM, NV40 didn't tape-out yet.
- Thread starter Arun
- Start date
Previews on sample silicon by August 2002?
Well, there were two rumors of two tape-outs, yeah - one in March IIRC, one in August.
Wasn't the one in March using Low-K anyway? Don't know all those details frankly, so if you can explain them, it'd be a very good thing
Uttar
Well, there were two rumors of two tape-outs, yeah - one in March IIRC, one in August.
Wasn't the one in March using Low-K anyway? Don't know all those details frankly, so if you can explain them, it'd be a very good thing
Uttar
Uttar said:Sabastian said:
I doubt that, but who knows.
nVidia isn't behind schedule for tape-out yet anyway, don't know where that idea came from
Uttar
I was under the impression that the low k process allows for a higher core clock with less heat. (err, I could be way off base here.) But if that is the case then the NV40 running at the desired clock speed may be too hot running.
Uttar said:Previews on sample silicon by August 2002?
Well, there were two rumors of two tape-outs, yeah - one in March IIRC, one in August.
Wasn't the one in March using Low-K anyway? Don't know all those details frankly, so if you can explain them, it'd be a very good thing
Uttar
Now I am more confused.... :? The one in March I think he is refering to was in 2002 of the NV30 core, I think. That would not have been done using the .13um low k process. "Wasn't the one in March using Low-K anyway?" huh??
Well, a company like TSMC always offer Low-K, they even did in early 2002 I believe, and that even for very complex parts such as the NV30.
The question is whether the yields are sufficent.
Frankly, I must admit to not quite see what Russ is talking about though. Previews on sample silicon? In August 2002? Private, then?
Uttar
The question is whether the yields are sufficent.
Frankly, I must admit to not quite see what Russ is talking about though. Previews on sample silicon? In August 2002? Private, then?
Uttar
Uttar said:
Now that I think of it I do recall reading somewhere that TSMC did offer it back then that makes more sense now. As far as the previews in Aug... there were all sorts of rumors of it at that time. Anand was saying the card was coming soon in Aug. So maybe there were private showings, but they must have been miserable.
No, they probably weren't. If there were any private showings, remember that every impression about the NV30 before its launch was very, very positive. If it is true that a low-k NV30 was previewed by selected parties back in August, then it stands to reason that it was superior to the NV30 that we all know and love (or don't), if only due to the low-k process.Sabastian said:Now that I think of it I do recall reading somewhere that TSMC did offer it back then that makes more sense now. As far as the previews in Aug... there were all sorts of rumors of it at that time. Anand was saying the card was coming soon in Aug. So maybe there were private showings, but they must have been miserable.
I think nVidia just realized too late that it was going to be prohibitively expensive to produce the NV30 on the low-k process, due to low yields. This may have, in the end, also resulted in the cutting of certain features in the core (one would hope, for instance, that the FX12 debacle was a mistake rather than a design decision).
Chalnoth said:
Yes, it is most likely the single largest problem for the NV3X family high end parts, I think that the original design called for Low-k too. The NV40 may be suffering from the same sort of fate. But if they were looking for the low-k process on the .13um process for the NV30 and NV35 it stands to reason that the NV40 was designed with an even smaller process or absolutely needing the .13um low-k process. I know that is exploratory but within reason. Is it possible that because certain fabrication processes are not where nvidia would have anticipated or gambled it would be, that their designs are not as advantageous as they once thought? ATi’s R300 core design on the .15um process is becoming more of a brilliant move then anybody would have formerly thought. Also using the .13um process for the RV350 looks like a fortunate design selection for ATi as well. Seems ATi can do no wrong and that is the impression that I have gotten for sometime now. Heh, just when someone says something like that you’d anticipate something to go wrong.
What I find most troubling though is the fact that not even the 5900's are obtainable yet (not that gobs of people will be after it or anything.) and the NV40 tape out is looming. Will the 5900's amount to yet another case of vapourware like the ill fated NV30? Is it possible that they are simply trying to have ATi hold the high end against vapourware all the while focusing on an entirely new core design with the NV40? I know the NV30 isn’t really vapourware technically speaking, but it never did ship in any volume and despite all the marketing racket with regards to it very few were actually able to get one, same may hold true for the NV35. Nvidia is really in the doldrums it seems. It is July and we have not seen any high end NV3X parts available and bearing in mind the NV30 core was launched in Nov and the NV35 launched in March it does seem as though nvidia have thrown the NV3X blueprint out the window so to speak. I think it is feasible we will not see the NV35 ship in volume.
Ahh, so many questions. Seems just like yesterday the Radeon 9700 Pro was launched, my how the big picture has changed for both ATi and Nvidia. We are already looking for the next gen cores.
http://www.newegg.com/app/viewproduct.asp?DEPA=&submit=Go&description=GeForceFX+5900+Ultra
Chalnoth, was it possible for nVidia to retool the NV30 into its current dual FX12 config at the late date by which they realized 130nm with low-K wasn't going to be feasible?
Sabastian, I don't think you can reasonably expect that any $500 video card will ship in volume. Your link showed no cards found, which may be your point, but NewEgg appears to have nine Gainward 5900U's in stock ATM, with the eVGA 5900U due in two days.
Sabastian, I don't think you can reasonably expect that any $500 video card will ship in volume. Your link showed no cards found, which may be your point, but NewEgg appears to have nine Gainward 5900U's in stock ATM, with the eVGA 5900U due in two days.
horvendile
Regular
Sabastian said:
I can find all sorts of 5900's in stock here (provided there are only three kinds), even the 256 Mb Ultra. Admittedly the quantities are not huge, but my experience is that when graphic cards finally appear in Sweden they are already used as doorstops in Ougadougou.
(Note: This contradicts the letter but not the spirit of your post. Although I could lay my hands on a 5900 should I be inclined to, it can easily be argued that the large volumes are not here. (On the other hand, supplies of the R350 are not immensely better.))
Chalnoth said:
I doubt that. I don't even think Low-K allows higher transistor counts.
What I wouldn't be surprised by, however, is that the register usage problems were either created or exagerated by the impossibility to use Low-K. I really hope it's all gone in the NV40...
What might be interesting to watch there is the NV36 - if it uses SOI. Perhaps that with SOI, the design wouldn't have that problem anymore - although I'd bet it'd still have it, at least for the most part, considering Low-K and SOI are fundamentally different.
Uttar
horvendile said:Sabastian said:
I can find all sorts of 5900's in stock here (provided there are only three kinds), even the 256 Mb Ultra. Admittedly the quantities are not huge, but my experience is that when graphic cards finally appear in Sweden they are already used as doorstops in Ougadougou.
(Note: This contradicts the letter but not the spirit of your post. Although I could lay my hands on a 5900 should I be inclined to, it can easily be argued that the large volumes are not here. (On the other hand, supplies of the R350 are not immensely better.))
Well I guess I should have looked a little harder.
AFAIK you can buy a Radeon 9800 pretty easy and OEMs widely carry it. I wonder how many they have sold... my guess with all the unenthusiastic press it will end up like another 5800 kind of release. I think over all the Geforce Fx series has been quite a humiliation for nvidia.( Not to mention their driver cheating.) Anyhow thanks for pointing out that there are some available.Pete said:
Well it didn't stop the Geforce 4 4600 from shipping heavily. In volume, I would suggest that, it means it is widely offered from a large array of vendors, of which the 5900 is definitely not nor do I think it will. The 5900 is an expensive card to produce with weaker margins then normal and performance is alright but not absolutely better at all.( http://www.ocaddiction.com/articles/video/fx_5900ultra_vs_radeon_9800pro/ )
ATi has the Radeon 9800 in both the OEM and retail channels and while that may not be mid stream volume shipments I'd think that it would be significant shipments for the high end portion of the market. In other words the Radeon 9800 is relatively shipping in volume, AFAIK ATi is having a time with supplying the demand.
NVIDIA NV33 and NV36 to Come Soon. NV40 – Not so Soon
So we're looking at Spring for both NV40 and Loki? I wonder how much more ATi can wring out of the R300 core....
So we're looking at Spring for both NV40 and Loki? I wonder how much more ATi can wring out of the R300 core....
Luminescent
Veteran
I surmise Ati can wring out quite a bit from the R300 core.
Luminescent
Veteran
I surmise Ati can wring out quite a bit from the R300 core.
Christmas is in spring now?Pete said:
Similar threads
