Ah, got it. Thanks for explaining it.DaveBaumann said:
Pfffft, this is me you're talking too....I NEVER get hung up on the idea that it is on the up-and-up!
R520 Running
- Thread starter whql
- Start date
Diggy, I'm not trying to bash HB or anything. His info is just kinda nuts and is probably a lot of hyperbole (if there's anything there at all in the first place). Keep in mind that his theoretical R520 is ~50% more efficient than R480 per clock in terms of pixel fillrate--you could clock it at 360Mhz and get the same fillrate as X850XTPE. This is why I think 450-500Mhz is a LOT more likely for a 24-pipeline part than 600Mhz or some nutty thing like that. Remember, even when ATI went from .15 to .13 low-k, which they already had experience with from the RV360, clocks from 9800XT to X800XTPE only went from 412 to 520Mhz. Given the yields on those aggressively clocked R420s, I don't think ATI is going to be nearly so aggressive on clocks.
Ah, thanks Baron. I like your explanations, I can usually understand them.
I didn't think the clockspeeds would be that high because of the lack of low-K on the new .09 process, I think heat build-up will be more of an issue.
I didn't think the clockspeeds would be that high because of the lack of low-K on the new .09 process, I think heat build-up will be more of an issue.
But the 09 is much easier to make than the .13 was.
And it wasnt so much the poor yeilds as it was the lack of Fabs to get out more product. The 420 was a OEM design from the start and had to be a cool single slot solution.
The 420 core can get to 600 on air. If all they did was to make it be the baddest GPU on the block and clock so high as to need shopvacs to cool well they wouldnt sell many. And the 09 from TSMC IS low-k.
And it wasnt so much the poor yeilds as it was the lack of Fabs to get out more product. The 420 was a OEM design from the start and had to be a cool single slot solution.
The 420 core can get to 600 on air. If all they did was to make it be the baddest GPU on the block and clock so high as to need shopvacs to cool well they wouldnt sell many. And the 09 from TSMC IS low-k.
It is? I was under the impression that the 520 wasn't going to be low-k. :?karlotta said:
And TSMC is definately making the 520, right? :|
600-700MHz core is likely, IMHO *wink wink*. If it's appreciably slower than 650MHz I believe ATi will be disappointed. I doubt it's a 24 pipe part. 24 pipes@650MHz = 15600Mpixels/sec. Working backwards from that, I don't think available memory will be fast enough to constitute a "balanced" architecture. We generally don't see mismatches of this kind at the high-end. Besides, I think that is a rather improbable scenario when you consider what else they are packing into this ASIC and the implications in terms of die size/thermal characteristics.
ANova, sorry - cheap shot. I just saw that tagline on another forum and thought it was pretty funny (if a little harsh!).
ANova, sorry - cheap shot. I just saw that tagline on another forum and thought it was pretty funny (if a little harsh!).
The Baron said:
I realize that this is an oversimplification, but you're comparing the clock speeds of ATI's very last high-end, high-transistor 0.15u ASIC (after a long process maturation) to the very first high-end equivalent at 0.13u. If you instead compare R300 (275MHz) to R420, clock speed nearly doubled. And I may be overspeculating, but R4XX as a stop-gap solution might not have benefitted from as much development time as R300/R5XX to allow for the most efficient design. Even if it doesn't happen, I don't see how anyone can say that 600MHz is ludicrous.
So Moofy thinks it's a 650Mhz 16-pipe part, I'm thinking it's a 400-550 (nice big range, whee!) 24-pipe part. A year and a half passed from R300 to R420, during which time ATI gained plenty of experience with both .13u and low-k from RV3x0. From R300 to R360, clock speed increased from 325 to 412, so around 25% on the same process. I think the lack of any prior efforts with .09 is really going to hurt ATI's first effort in terms of efficiency and make them scale back the clocks a little bit to keep yields high (NV30, anyone?).kemosabe said:
But Moofy winked, so he probably knows things that I don't. Bah. You bastard, taking all the fun out of my speculation
(on a similar note, though, as a process matures, which would be harder to do--increase the clock speed in a more complex part with a lower clock or a less complex part with a higher clock to begin with? so let's say a 16-pipe GPU versus a 24-pipe GPU. would increasing a 16-pipe part at 600Mhz to 750Mhz be harder than increasing a 24-pipe part at 400Mhz to 500Mhz?)
I bet nV have a 0.09u, 16-pipe part now (NV46?). The news from their camp has been very confusing and I wouldn't be suprised if a lot of it has been smoke & mirrors.
I would imagine that NV40 is about the same speed as R520 on a per-clock basis, so they could produce something competetive in a relatively short timeframe with a shrink and some tweaks (assuming their relationship with IBM/TSMC might permit such a thing).
I would imagine that NV40 is about the same speed as R520 on a per-clock basis, so they could produce something competetive in a relatively short timeframe with a shrink and some tweaks (assuming their relationship with IBM/TSMC might permit such a thing).
Demirug said:
Only because of something cho said in a different thread and he's been on the money in the past. I haven't heard of it, otherwise.
I was wondering about those other listings too. What are the improvements? It is quite easy to see how they might have decided on NV4E and NV4F for new part names at this point in time.
I have no details about this improvements yet. The only thing I am know is that at the same time NV4E and NV4F appears at the chiplist a new profile for fragment programes appers, too. But this new profile have only a new minor number (1 in place of 0). That was the reason why I am add the "(small)". I am don't expect much. Maybe some new instructions for special accelerated functions like the NRMH that we allready have for NV40.
Hellbinder
Banned
The Baron said:
There is absolutely *nothing* nutty about what i posted. Either about the Specs, clock speeds or AA. Which i already said the Mentioned AA might not make an apprearance until the next round after this although it could have made it into this one.
Hellbinder
Banned
The Baron said:
I still dont understand why you think this is some kind of slam. I was basically dead on accurate about 12 months in advance. Perhaps you should take a closer look and rethink your critisisms of me.
The X800 is 4 "Quads" that are similar to the RV Quads and then beefed up to PS 2.0+ and some other things. This is the basic idea I put forth on Nvnews a long long time ago. Is it *exactly* accurate? No but it is pretty close.
The X800 Quads share several similarities to the RV360. Like for instance the AF method is identical, and not the same as the R300 or R360. Thats just one thing.
Hellbinder
Banned
aah.. Now i see...
between 500-600mhz is a more accurate statement regarding the Clock speed of the XTPE.
Remember this card is going to be a rare item sold in small Quantities and priced accordingly. Its for winning benchmarks and presiege. Not mass market. The Clock Requirements for this part are *PEAK* attainable for some thousands of parts.
you should keep that in mind.
between 500-600mhz is a more accurate statement regarding the Clock speed of the XTPE.
Remember this card is going to be a rare item sold in small Quantities and priced accordingly. Its for winning benchmarks and presiege. Not mass market. The Clock Requirements for this part are *PEAK* attainable for some thousands of parts.
you should keep that in mind.
Similar threads
