AMD: R7xx Speculation

Status
Not open for further replies.
That things like the performance don't matter because it's all about the mhz. Until it's not, then it's all about the precision, until it's about the cores. Until someone else has more of them, unless it's useful to claim that Nvidia have more cores than an Intel CPU.

I don't think that's accurate. Where has Nvidia ever claimed that "performance doesn't matter"? I think you're guilty of some of the things that you accuse Nvidia of. They never claimed that the number of cores doesn't matter. What they claimed is that their "128 core" part had equal performance to a competing "800 core" product. I don't see how that's misleading if it's true!

It's a general level of dishonesty that goes beyond putting your best in a good light, and into pretending that black is white and that what was good yesterday (because you had it) is bad today (because you don't).

That's the definition of marketing.

Cheats in drivers, exaggerated scores in benchmarks that are not reflected in gameplay, "leaked" marketing slides about your competitors with incorrect info, etc.

All things that AMD does as well.

It's a trend for dishonesty that should not be encouraged in any industry, no matter who does it. And people like you who are jaded by it and don't want to hear it any more are helping companies get away with conning the majority who are not as savvy as you are.

I may be jaded but that's not the reason why I don't get all upset. These companies aren't playing hopscotch in the park. I just find it silly that Nvidia pointing out what is essentially the truth would offend anyone. They have a history of being underhanded but in this particular case I don't see anything wrong with what they're saying as a practical matter.
 
I may be jaded but that's not the reason why I don't get all upset. These companies aren't playing hopscotch in the park. I just find it silly that Nvidia pointing out what is essentially the truth would offend anyone. They have a history of being underhanded but in this particular case I don't see anything wrong with what they're saying as a practical matter.

How can it be the truth when they change their minds all the time?

More cores = important when Nvidia has them and ATI doesn't (G80).

More cores = irrelevant when ATI have them against Nvidia (R770)

More cores = important when Nvidia have them against Intel. (CPU)

Sure, you can look at the power of the cores, but that's not what Nvidia is doing unless it suits their message because they would then have to consider just how ridiculously over the top the number of cores in RV770 is, to the extent that it's making up the difference. Nvidia's certainly happy not to look at the big picture when it comes to bragging about all those cores it has more than an Intel CPU, because then they would have to consider the lack of X86.


Anyway, I'm drifting off-topic so I'd better stop.
 
Last edited by a moderator:
Little/big question:

With the new PLX (gen2 with 32 lanes)...

...and with a 4 pcie 32x motherboard...


...is it possible to have a beast with 4 x R700???
 
Little/big question:

With the new PLX (gen2 with 32 lanes)...

...and with a 4 pcie 32x motherboard...


...is it possible to have a beast with 4 x R700???

If R700 does not work on AFR-base, it should be possible. Else you would have a 7 FPS input-lag. ;)
 
More cores = important when Nvidia has them and ATI doesn't (G80).

More cores = irrelevant when ATI have them against Nvidia (R770)

More cores = important when Nvidia have them against Intel. (CPU)

You're simply stating the obvious which is that marketing is focused on highlighting a product's strengths vs the competition.

What you forgot to mention is that in all three scenarios the measured performance backs up the claims so I see nothing to get upset over. In the comparisons where Nvidia had more "cores" they also had more performance, this isn't the case when AMD is winning the core count comparison.
 
Best die shots yet

Supposed 4870 performance provided by Sapphire:

http://forum.donanimhaber.com/m_24110474/tm.htm

Looks like the card will be 20-25% faster than the 4850 after all. Looks like 20% higher core clocks + better memory bandwidth should help, and I bet the gap gets larger as higher AA and res's are used.

Those benchmarks make no sense. The 4850 vs 9800GTX scores in Crysis don't tally up to other reviews.

Although I suppose as a 4850 vs 4870 comparison point it could be usefull.
 
Best die shots yet

Supposed 4870 performance provided by Sapphire:

http://forum.donanimhaber.com/m_24110474/tm.htm

Looks like the card will be 20-25% faster than the 4850 after all. Looks like 20% higher core clocks + better memory bandwidth should help, and I bet the gap gets larger as higher AA and res's are used.
In those benchmarks the 4870 loses only 13%, going from no AA to 8xAA in Crysis at 1920x1200. Pretty amazing.
 
Best die shots yet

Supposed 4870 performance provided by Sapphire:

http://forum.donanimhaber.com/m_24110474/tm.htm

Looks like the card will be 20-25% faster than the 4850 after all. Looks like 20% higher core clocks + better memory bandwidth should help, and I bet the gap gets larger as higher AA and res's are used.

Nice looking card. I really like what AMD has done with the black/red contrast this time around.

Sapphires Vantage scores for the 8800GT and 4850 seem to be lower and higher respectively compared to http://www.pcper.com/article.php?aid=579&type=expert&pid=6 though.
 
unbenanntve5.jpg

http://www.computerbase.de/artikel/hardware/grafikkarten/2008/kurztest_ati_radeon_hd_4850_rv770/

:D

RV770 = new R300
 
Assuming that RV770's die is square, and assuming that each of those 10 rows is a SIMD:
  • 4 MADs including register file
  • 1 MAD/transcendental with no register file
  • sequencer including constant/instruction cache
then 6 SIMDs (i.e. what's been added in comparison with RV670) amount to 24.4% of the entire die. All 10 SIMDs amount to 40.8% of the die.

So, one SIMD is about 4.1% of the die, which I guess is about 10.4mm2

How big is one multiprocessor in GT200? About 5mm2 on 65nm? (According to Arun 30 SMs are 26.5% of the die - then assuming 576mm2 die.)

Jawed
 
How much faster are the clocks on 4870? Is it 750 core or 850?

If it's 750, that's 20% faster than 4850, assuming all else equal between them, which of course isnt right but for simplicity, then going back to the French review, the 4870 will give the 260 all it wants. Basically being just as fast overall, sometimes faster, sometimes slower. If the 329 price on 4870 is true, that will present a problem. I hate to keep relying on that french review, it might be overly ATI friendly for some reason, but it does have 4850/GT200 benches all in one place which makes it easy.

Edit: Now I see the 4870 clock is indeed 750, but the price is an even better 299.
 
That's the definition of marketing.

Hardly. That's the definition of dishonesty.

All things that AMD does as well.

Not to nearly the same degree.

NV's marketing could be a lot better, they've got a down right horrible track record when it comes to this stuff. The stuff they did to promote NV30 vs R300 was just ridiculous and the fact that many consumers were still mislead by their BS campaign just proves how unethical it is. It's on the same level as companies selling $500 power cables but arguably worse because in this industry the consumer requires far more knowledge to make an informed purchase.

This time around it will not be nearly as bad as the whole NV30 dealy but NVs PR tactics still annoy the hell out of me.
 
So, RV770's 1200GFLOPs take 104mm2, while GT200's 1000GFLOPs take 153mm2:
  • RV770 is 11.5 GFLOPs/mm2 on 55nm
  • GT200 is 6.6 GFLOPs/mm2 on 65nm
For double-precision:
  • RV770 is 2.3 GFLOPs/mm2 on 55nm
  • GT200 is 0.5 GFLOPs/mm2 on 65nm
Jawed
 
Status
Not open for further replies.
Back
Top