ATI Radeon HD 4000 Series discussion

[Eastcoasthandle]

Here is a translation if anyone didn't understand the chart


Makes me wonder if the Pixel Fill rate is around 27.00 MT/s

[biohead]

Steamprocessor. Is that DX11 ?

[w0mbat]

No, still DX10(.1) - G80/R600 also had streamprocessors (no steam).

[whocaresbg]

Here is a translation if anyone didn't understand the chart


Makes me wonder if the Pixel Fill rate is around 27.00 MT/s

To has ~ 27.00 MP/s RV770XT needs 32 ROPs .

[Eastcoasthandle]

To has ~ 27.00 MP/s RV770XT needs 32 ROPs .

If the above chart is true why must RV770 have 32 ROPS
The arch design of this video card hasn't been release as of yet so I like to know how you come up with this?

[whocaresbg]

If the above chart is true why must RV770 have 32 ROPS
The arch design of this video card hasn't been release as of yet so I like to know how you come up with this?

16 * 850 = 13.6 MP/s
32 * 850 = 27.2 MP/s

[Mungri]

I'll layback and wait on one of the Dumped 3870 and run Crossfire on my next system.

Sorry for the late reply, but as people have already pointed out, it isnt worth crossfire'ing 3870's on a P35 due to how slow the second slot is.

I benchmarked my 3870 in both the 16x and 4x PCI-E slots for a comparison, and the results were really shocking, even I didnt expect such a big difference:

http://i118.photobucket.com/albums/o...vv/16xvs4x.jpg

A second card in the 4x slot would really be crippled. I am hoping that Intel / AMD can come up with a cheap chipset option that supports 2x 16x PCI-E lanes, just like Nvidias 650i and 750i.

If you have an X38 or X48, then crossfire is great. But on a P35 it is a waste of money.

[Eastcoasthandle]

16 * 850 = 13.6 MP/s
32 * 850 = 27.2 MP/s

What is this suppose to mean? Your post is to vague.

[AliG]

Not sure where he's getting 16 from, but I believe he means 32 TMUS * 850 mhz (which it should really be 1.05ghz going with his logic) = 27.2 MP/S.

[Eastcoasthandle]

I see, thanks

[whocaresbg]

Just if RV770 has 16 ROPs, this suppose a pixel fill rate of 13,6 MP/s, not around 27 MP/s ...

Texture Fill Rate = (# of TMUs) x (Core Clock)
Pixel Fill Rate = (# of ROPs) x (Core Clock)

[AliG]

highly doubt that could possibly happen, that mean the alu:text ratio would go even lower, which would never happen after the r600's failure to take the crown. If anything, it would go to 24 TMU if they didn't change anything besides upping the shader count 50%

[whocaresbg]

But the history shows that ATI is increasing ALU:TMU ratio in every new generation .

[Scimitar]

Just if RV770 has 16 ROPs, this suppose a pixel fill rate of 13,6 MP/s, not around 27 MP/s ...

Texture Fill Rate = (# of TMUs) x (Core Clock)
Pixel Fill Rate = (# of ROPs) x (Core Clock)

That is correct. Nice sumation. 16 ROPs is definately going to be a drawback. Texturing performance should be excellent though.

[whocaresbg]

Thanks . In Chiphell's rumours are writting that ATI will save the TMU-ROP ratio of 1 : 1 etc . 32 TMUs/32 ROPs .

[AliG]

But the history shows that ATI is increasing ALU:TMU ratio in every new generation .

And obviously they've learned that doing so isn't applicable from how badly the 8800gt spanked their offerings when the 8800gt had roughly 1/3 the alu power but something like 4 times the texture power. In order to make it competitive, they had to reduce the ratio, I don't think anyone would have bought the product if they didn't as gt200 would kill it without breaking a sweat (with af enabled). I personally want to see either a 1:2 ratio or if possible a 1:1 ratio again, that's proven to give the best results and it will continue to unless games are coded differently (which won't happen any time soon).


As for the rop power hampering performance, I doubt it, as aa is handled by the shaders with the r600 design, so it was actually the af that was killing performance, not aa. But I would like to see a 1:1 rop:tmu ratio, as I just said 1:1 ratios have proven to work with today's games and that's what matters for us. Servers can make more use of the shaders, so the r600 made a great firegly, but still a decent gamer

Besides, just to point something out, g92 has 16 ROPs but look at how it handles aa

[whocaresbg]

Therefore I think that the Eastern rumours sound more reliable cos ATI is fighting for the 200-300$ market and if the GT200-GT is 2 x faster than G92GT, AMD cannot able to catch up with NV .

[AliG]

perhaps, but its not meant to, the r700 wil go against gt200. Not to mention, because of its design, the r700 actually be more powerful than simply 2 xfired rv770s.

[hlopek]

Yea, I'm one of the guys who's been posting quite a lot of stories about the upcoming chip

There are talks about up to 800 SPs

I'd like to see a logical chart explaining how 64SP/16ROP/16TMU becomes, 96SP/16?ROP/32TMU.

That are just mouth full of lies .... 800SP is just kiddy talk. There are some "shoots of the chip" and they talk about 220mm2 so that could keep themselves (ATi) from market failure they done last year with enormously extra large R600 core. In the end its only 480SPs more TMUs (32) and only 16ROPs again so that the only radical performer will be gang member RV770x2 (R700)

btw. in most games old R600 underclocked @500M is pretty monsterous, so if you kept yourself away from crappy Crysis and Clones stuff you don't really need even that mainstream monster of RV770.

ntm. that only RV670 core will be inside Fusion an that's going to see light of the sun late next year.

[adamsleath]

i have faith that the producers know exactly what they are doing.

my early attempts at connecting an abacus to an exosketch were a dismal failure...

[biohead]

No, still DX10(.1) - G80/R600 also had streamprocessors (no steam).

the translation said "steam processor" but he corrected it =)

[NH|Delph1]

That are just mouth full of lies .... 800SP is just kiddy talk. There are some "shoots of the chip" and they talk about 220mm2 so that could keep themselves (ATi) from market failure they done last year with enormously extra large R600 core. In the end its only 480SPs more TMUs (32) and only 16ROPs again so that the only radical performer will be gang member RV770x2 (R700)

btw. in most games old R600 underclocked @500M is pretty monsterous, so if you kept yourself away from crappy Crysis and Clones stuff you don't really need even that mainstream monster of RV770.

ntm. that only RV670 core will be inside Fusion an that's going to see light of the sun late next year.

No one has presented any actual truth for either of the scenarios. If you go back and read some of me other posts you will see that I'm very doubtful of 800SPs. I'm only pointing out that Eastern sources are very persistent in their claims of 800SPs. They think we are just as stupid clinging to 480SPs.

And the chart posted here (http://www.hardware-infos.com/news.php?news=2008) looks like it was based on the stuff Theo posted.

//Andreas

[perkam]

That are just mouth full of lies .... 800SP is just kiddy talk. There are some "shoots of the chip" and they talk about 220mm2 so that could keep themselves (ATi) from market failure they done last year with enormously extra large R600 core. In the end its only 480SPs more TMUs (32) and only 16ROPs again so that the only radical performer will be gang member RV770x2 (R700)

btw. in most games old R600 underclocked @500M is pretty monsterous, so if you kept yourself away from crappy Crysis and Clones stuff you don't really need even that mainstream monster of RV770.

I imagine the 800SP talk came form the X2 itself, which will ultimately comprise of 960 SPs if the 480 number is true. When you think about it though, the increase in shaders alone is enough to bring out a theoretical 50% (closer to only 20% in reality due to diminishing returns) increase in performance in games, and that is if they left it at 480/16/16. The doubling of the TMUs only adds to performance and to my belief that ATI doesn't need 32 ROPs at the moment, even with the GT-200, as it can mass produce the 4870s on 16 ROPs with similar architecture to the 3870s, keep its costs low and under cut the GT-200.

Remember that your argument of prior generation sufficiency works both ways. If most people will never need more than a 500mhz core RV670, then they will never need a GT-200 if the RV770 or RV770 X2 ends up cheaper, which in my experience is very likely as, if it ends up on the same architecture as the HD 3800 series, their costs, even for the X2, will be lower than Nvidia's brand new architecture.

With that said, I am a firm believer that people at XS will buy the card with the best performance, irrespective of the price, so there's little doubt that ATI will need a more powerful high end, considering their past record against new Nvidia cards (7800 vs R500, 8800 vs R600).

Perkam

[flopper]

With that said, I am a firm believer that people at XS will buy the card with the best performance, irrespective of the price, so there's little doubt that ATI will need a more powerful high end, considering their past record against new Nvidia cards (7800 vs R500, 8800 vs R600).

Perkam

if they get a 50% up and get a good price, I aim to get 2 cards for crossfire.
I have enough of power atm and overkill is plenty with ati cards for gaming.

dont use 24 or 32 inch screens so I am good.

[whocaresbg]

FUD! RV770XT supports 512bit memory :

http://www.fudzilla.com/index.php?op...=6994&Itemid=1

According to this scenario the specs should look thereby : 480/32/32 or 800/32/32 .