Did R480/430 require a new tapeout?
- Thread starter kemosabe
- Start date
I thought the very fact that both GPUs have different names than R420/423 is proof of a new tapeout? Josh explicitly states that R30/480 are new tapeouts near the end of his article.
So, yes, I think you should enjoy that BLT.
The article is interesting in noting the tapeout disparity. Perhaps ATi is now slogging through some of nV's FX learning curve in working with 130 and 110nm? Or perhaps they're just more public/twitchy with their respins? Maybe nV spend more time finessing NV40 before its "launch" than ATi did R420, or perhaps they don't give a new product name to each slight revision/respin?
So, yes, I think you should enjoy that BLT.
The article is interesting in noting the tapeout disparity. Perhaps ATi is now slogging through some of nV's FX learning curve in working with 130 and 110nm? Or perhaps they're just more public/twitchy with their respins? Maybe nV spend more time finessing NV40 before its "launch" than ATi did R420, or perhaps they don't give a new product name to each slight revision/respin?
Pete said:
Err, that's why I thought it relevant to raise the question in the first place since I didn't recall it being explicitly reported before.
Ok, so I'll have the onions, tomatoes, and heavy on the hot pickles.
It may have been worth the investment as the X800XL and X800 should make it a more interesting battle in the $250-$350 segments, at least for PCI-E.
DegustatoR
Legend
R430 is the new chip (remade on 110nm). R480 is just a new revision of R423 AFAIK. R480 is more of a marketing name than something really new.
R480 is on an updated 130nm low-k process and does have a few silicon tweaks in there. The talk of power management updates is the case as it appears that M28 / M28 PRO will actually be using R480 not R423 (which is why they had not released them immediately).
Heh, sorry, for some reason I thought you said he noted it *didn't* require a new tape-out. Save some of that BLT for me!kemosabe said:Pete said:
Err, that's why I thought it relevant to raise the question in the first place since I didn't recall it being explicitly reported before.
Ok, so I'll have the onions, tomatoes, and heavy on the hot pickles.
Haha, no offense taken.
Well, nobody from NVIDIA has told me anything about NV48 and NV50. Someone outside of NVIDIA gave me hints about NV48... basically said that the advantages of this chip were outweighed by the cost of producing it. The 20 to 30 MHz and a couple of fixes didn't seem economically wise to NV (or so I hear...)
As for NV50, I think it is far too early to say that something like that has been cancelled. Probably one of the future chips was scrapped, but who knows really which one? Anyway, we will still probably see a NV50, but we just won't know if it was the original design, or something that NVIDIA shoehorned into the lineup.
Faud seems like an interesting guy with his ear to the ground, but I would say he is about 50% right 80% of the time (the other 20% he is so far off base...) So if you can figure out that magic math, you are smarter than I am.
The one thing I am very curious about will be ATI's FP 32 performance and how well they implement SM 3.0 in the R520. Also, will they utilize FP 16 (partial precision)? I personally think they will, as it is a very useful concept (save space and resources when going to FP32 won't make a difference in output quality). If this happens, then I am anxious to see the kind of spin ATI does after berating NVIDIA for evangalizing the use of partial precision. I think the next 5 months are going to be a lot of fun!
Well, nobody from NVIDIA has told me anything about NV48 and NV50. Someone outside of NVIDIA gave me hints about NV48... basically said that the advantages of this chip were outweighed by the cost of producing it. The 20 to 30 MHz and a couple of fixes didn't seem economically wise to NV (or so I hear...)
As for NV50, I think it is far too early to say that something like that has been cancelled. Probably one of the future chips was scrapped, but who knows really which one? Anyway, we will still probably see a NV50, but we just won't know if it was the original design, or something that NVIDIA shoehorned into the lineup.
Faud seems like an interesting guy with his ear to the ground, but I would say he is about 50% right 80% of the time (the other 20% he is so far off base...) So if you can figure out that magic math, you are smarter than I am.
The one thing I am very curious about will be ATI's FP 32 performance and how well they implement SM 3.0 in the R520. Also, will they utilize FP 16 (partial precision)? I personally think they will, as it is a very useful concept (save space and resources when going to FP32 won't make a difference in output quality). If this happens, then I am anxious to see the kind of spin ATI does after berating NVIDIA for evangalizing the use of partial precision. I think the next 5 months are going to be a lot of fun!
With an ATI-like design, the performance improvement of FP16 would be pratically nil. The only reason it's faster on NV3x/NV4x is registers, and to a lesser extend instructions like NRM_PP. I doubt ATI is going to bother about FP16, thus, except perhaps for specific instructions. And even then, they might consider it's not worth the trouble...JoshMST said:
Uttar
DegustatoR
Legend
Well, yes, but R520 is going to be a big change in the R300 design since it'll have to support SM3. NV's design approach for NV30 was made for something like SM3 in mind from the beginning. ATI's design for R300 was pure SM2 and almost nothing more (almost because of centroid sampling and geometry instancing which didn't make it to SM2 specs).Uttar said:
Where I think the article is out to lunch a little is the idea that validiation of chips will eat into next generation engineering - generally speaking they are completely separate groups.
Unknown Soldier
Veteran
Just a question.
Tapeout does mean they've went silicon. i.e A01 etc.
Or does tapeout mean something else? i.e. Design phase
Another thing .. is it tapeout or tape out
Thx
US
Tapeout does mean they've went silicon. i.e A01 etc.
Or does tapeout mean something else? i.e. Design phase
Another thing .. is it tapeout or tape out
Thx
US
Good links Dave, I hadn't read any of those in the past, so they are new to me. I guess from that standpoint it seems pretty obvious that they will be FP32 all the way through, and won't find a need to support partial precision.
As for tapeout, it is the final design stage before the plans are sent out to have masks created which are then sent to the foundry to fabricate the chip. Once tapeout occurs, it generally means the fabless semi is done with the design of the chip, and it is ready to be produced. After this there may be other revisions and respins, but tapeout is generally when the company says, "we finished our design".
As for tapeout, it is the final design stage before the plans are sent out to have masks created which are then sent to the foundry to fabricate the chip. Once tapeout occurs, it generally means the fabless semi is done with the design of the chip, and it is ready to be produced. After this there may be other revisions and respins, but tapeout is generally when the company says, "we finished our design".