GT 280/260 performance comparison to 3870X2

[Shintai]

Why should 4870 be only %50 (and even that "with a lot of wishful thinking") faster than 3870 when its processing power doubled, memory bandwidth doubled, memory amount doubled, texture addressing/filtering more than doubled, and z-fillrate at least doubled?

And if 4870 is %50 better than 3870, which makes 4850 only %10 or 15 better than 3870; how the hell can AMD/ATi justify 4850's price of $220, whereas HD3870's are as low as $150?

I would expect 4870 to be close to twice 3870, which would justify both part's prices and specifications.

Shaders aint double, ROPs aint changed at all and so on and on.

[gOJDO]

There are 50% shaders, but at 35% higher clock. So you might consider them doubled (1.5 * 1.35 = 2). There are same nubmer of ROPs as on 3870 but their frequency is being raised for 35%. The number of texturing units has been doubled and their frequency raised for 35% or x2.7 more texturing capabilities compared to 3870. Don't underestimate the 4870. IMO it will be at least 50% faster then 3870.

[Shintai]

There are 50% shaders, but at 35% higher clock. So you might consider them doubled (1.5 * 1.35 = 2). There are same nubmer of ROPs as on 3870 but their frequency is being raised for 35%. The number of texturing units has been doubled and their frequency raised for 35% or x2.7 more texturing capabilities compared to 3870. Don't underestimate the 4870. IMO it will be at least 50% faster then 3870.

Sure about all those clock raises and shaderspeeds?

When looking at the 4850 vs 3850 it doesnt seem to be so.



Memory is also a good 150Mhz lower than the rumour said.

Plus the core speed is 43Mhz lower than on HD3850 aint it?

And even on HD3870 vs HD4870 (If we trust the rumours) its abit udner 10% core speed increase. Not sure where you get those 35% from besides the shaderpart only. TMUs etc aint affected.

[annihilat0r]

Shaders haven't doubled and I didn't imply they did; but with increasing clock rates, math processing power has doubled. Z-fill per clock has either doubled or quintupled, I'm not sure; and as gojdo pointed out, the main bottleneck for R600-based cards, texture management rate, has nearly tripled (although I'm not sure on the clock rate, for the core except shaders is it really %35? I thought it was just 800 or 900MHz); and counting double the memory bandwidth and memory amount, nearly everything you can imagine has at least doubled.

The performance has to be somewhere close to double 3870. If it's only 1.5x, HD4850's price of $220 doesn't make any sense at all against the $150 HD3870. To me, everything points to the direction that 4870 will at least be 1.7 or 1.8 times HD3870, and 4850 will be at least 1.4-1.5 times HD3870, which fits in the 3dmark Vantage performance scores we have heard (4850 being slightly faster than 9800GTX)

[Hornet331]

Shaders haven't doubled and I didn't imply they did; but with increasing clock rates, math processing power has doubled. Z-fill per clock has either doubled or quintupled, I'm not sure; and as gojdo pointed out, the main bottleneck for R600-based cards, texture management rate, has nearly tripled (although I'm not sure on the clock rate, for the core except shaders is it really %35? I thought it was just 800 or 900MHz); and counting double the memory bandwidth and memory amount, nearly everything you can imagine has at least doubled.

The performance has to be somewhere close to double 3870. If it's only 1.5x, HD4850's price of $220 doesn't make any sense at all against the $150 HD3870. To me, everything points to the direction that 4870 will at least be 1.7 or 1.8 times HD3870, and 4850 will be at least 1.4-1.5 times HD3870, which fits in the 3dmark Vantage performance scores we have heard (4850 being slightly faster than 9800GTX)

with the score vr-zone reported its far from being faster then a 9800gtx (P6700 vs P9000)...

[annihilat0r]

Aha, I'm sorry? A 9800GTX gets a Performance score of 13.000 in Vantage??

[r1ch]

Shaders aint double, ROPs aint changed at all and so on and on.

Exactly. The 16 ROPs were a weakpoint of the 38x0 and 2900 cards, so the fact that they've not increased them yet again doesn't bode well in my eyes.

8800GT had 16
8800GTS had 20
8800GTX had 24
GTX 260 will have 28
GTX 280 will have 32

[Hornet331]

Aha, I'm sorry? A 9800GTX gets a Performance score of 13.000 in Vantage??

haha sorry, orb failed me... it showed sli results even i filterd for single cards...

Q6600 with 9800 gets ~5800.

[annihilat0r]

Okiez.

For the ROP thing, I've heard the z-fill per clock has doubled or something. That will probably remove the fillrate bottleneck

[Donster]

GX 280 still due out this month?

[largon]

How come ROPs (RBEs) were a weakpoint for R600/RV670?
And I don't think you can really compare nV's ROPs to ATi's RBEs...

[JohnZS]

I'm looking forward to the new GTX 280
Will be a worthy replacement for my 8800GTS
I just hope my Enermax Infinity 720W PSU is man enough for the task.
John

[JameCa]

im 100% sure this is bull. u wanna know why?.. how come the card preforms exactly the same in all games.. and why dosent the gt 280/260 preform the same in all.. a lil bit strange in my eyes and one moore thing.. inpossible that the bouth nvidia cards can preform the same in one game..

advertising is what i call it.. no test

[Shintai]

im 100% sure this is bull. u wanna know why?.. how come the card preforms exactly the same in all games.. and why dosent the gt 280/260 preform the same in all.. a lil bit strange in my eyes and one moore thing.. inpossible that the bouth nvidia cards can preform the same in one game..

advertising is what i call it.. no test

CPU limited? Even tho NVidia says no such thing!

[Psykocyber]

->JameCa, maybe you should think this through one more time?
The refence card will allways be at score 1, since it's the refence. The FPS doesn't matter, it's the difference that matters, and what Nvidia showed.

[jaredpace]

looks like COD4, UT3, and Fear (all 3 highly optimized & well programmed games) show raw graphics computational power increase in the GTX 200's. The other games show this as well as Forceware & DX communication improvements.

Edit: And, whats up with COJ and WIC? Are they still cpu limited? LOL

[Monkeywoman]

what does a 3870x2 score in vantage with performance level? the 4850 hits 6466. that doesnt sound bad. source for score http://www.vr-zone.com/articles/Rade...sted/5829.html

remeber the test was done on premature drivers

[zerazax]

ROPS were NOT the weak point of the RV670 series

It was texture filling. Look at these tests hoom at B3d did on a RV670 w/ 3dMark06: http://forum.beyond3d.com/showpost.p...postcount=2376

Just for kicks & since I'm currently reading a good book, I have just done a quick bunch of 3Dmark06 runs on my 3870 with various AA/AF configs.
E6600 @stock, 3870 @stock, Cat 8.5

no AF/AA 9883
16* AF/8*AA 8098
8* AF/no AA 8552
16* AF/no AA 8164
no AF/4*AA 9456
no AF/8*AA 9449
no AF/24*AA 8903

This was just one run of each config & I have stuff open in the background so not by any means scientific but there seems to be a pretty clear pattern.

I don't know about you, but it's pretty clear that it's AF that hurts the 3870 performance. Look at those numbers - AA is actually incredibly efficient. 8 x AF scores worse than 24 x AA!

If they doubled the TMU's, the AF issue should be fixed.

This makes sense anyways since no reviewers ever turn on AA w/o AF - however, the conclusion that it was AA killing performance was wrong.

[Extelleron]

with the score vr-zone reported its far from being faster then a 9800gtx (P6700 vs P9000)...

No, it is far faster than a 9800GTX.

With a quad-core CPU + 8800GTS 512MB (same as 9800GTX) @ 785/1891/2160, I get 6131, and my graphics score is 5609. I'm not sure whether or not they were talking about the graphics score or the overall, but either way the HD 4850 (6466) scores much higher than a heavily oc'd 8800GTS 512MB.

ROP's are not a bottleneck for the architecture right now. RV670 has much more pixel performance than G92 with its higher clock rates, yet it does not perform better. AMD already has plenty of ROP performance, it doesn't need more.

In comparing the 16 ROPs of RV770 to the 32 ROPs of GT200, you must remember (1) that the two products are not competitors, and (2) that AMD runs at a much higher clock speed. A HD 4870 with 850MHz core has only 18% less ROP performance than a GTX 260 @ 575MHz, if the two architectures are equal clock for clock.

[GoThr3k]

CPU limited? Even tho NVidia says no such thing!

there is no such thing if you play on normal resulotions, and only use a single VGA

[JameCa]

CPU limited? Even tho NVidia says no such thing!

i dont think the cpu is the bottleneck

[annihilat0r]

ROPS were NOT the weak point of the RV670 series

It was texture filling. Look at these tests hoom at B3d did on a RV670 w/ 3dMark06: http://forum.beyond3d.com/showpost.p...postcount=2376



I don't know about you, but it's pretty clear that it's AF that hurts the 3870 performance. Look at those numbers - AA is actually incredibly efficient. 8 x AF scores worse than 24 x AA!

If they doubled the TMU's, the AF issue should be fixed.

This makes sense anyways since no reviewers ever turn on AA w/o AF - however, the conclusion that it was AA killing performance was wrong.

I saw that as well and agree that most of the "AA kills ATI" stuff is because reviewers always enable AA with AF; and it's in most cases AF killing ATI, not AA. But still, ATI's AA implementation isn't efficient in a lot of games (for example Crysis and DX10 Bioshock - AFAIK any form of AA nearly halves the performance).

Anyway, that post proves that ATI's bottleneck is really the texture units, which should have been overcome by now.

Now to think about it, I doubt NVidia has any answer to give to AMD in the $200-350 range. G92b? I doubt it'll hit 900MHz or something which will put it in a competition with the 4870.

If AMD dominates the 200-350 range, they dominate half of the market. And NVidia will sit around like ducks with its $700 cards.