8800gts vs 2900 pro all tests made by me

[Tatty_One]

Originally Posted by DarkMatter

3DMark is not shader intensive? Cmon... How can then a less pixel fillrate, less texel fillrate, but much more shader powered HD 2900s be ahead of their respective competitors on this benchmark if they are just behind them on most games and benchies?

As for the other thing I was only trying to point out that on those conditions 14000 were not improbable at all, granted you believe the others at 15000. It's not that a big difference. You even mention on your post that only the CPU clock increase to 4.5Ghz can account for 800 points increase. In the same post you assume (based on core/memory) the shaders are running at 1860Mhz, but who knows? That is what I was trying to explain, I didn't say anywhere you didn't know nothing about shaders, but you oversaw the fact that they could be using a lower shader clock, and automatically you say it's imposible for a single GTS to achieve 14000 at those speeds. I was basically responding to that post on your part.
The "Since I can't reach..." was about other people, not you. That is why I said that first and then I talked to you. Should I have made 2 posts? Maybe. Still there are lots of post saying that they barely reach 14k with their GTX and thus 14k on the GTS is impossible, my first sentence was 4 them.

Sorry for the rant.

Okey dokey, we'll call it a day! but we do disagree on one point, 2006 is not particularily shader intensive and actually the 2900XT beats (since cat 7.8) even a GTX in a number of real world gaming scenario's....but thats another story! Just as a matter of interest, with no additional shader clocking, just leaving it in "sync" a core speed of 783Mhz auto clocks the shaders to 1761mhz which is pretty significant in itself!

[Lekamies]

Originally Posted by trt740

I do ,however, stand corrected on the 8800 gts it appears it can break 14000+ must have dry Ice strapped to it and a blessing or holy water in it's cooling system...

Warm water is enough to cool down for breaking that barrier.
Here is pic after 3d mark'03 run include rivatuners hardware monitor.

[DarkMatter]

I have heard and read lot of people claiming big performance boosts out of shader overclocking alone, so I will keep my thoughts.

The fact that new cat 7.8 makes 2900 perform better only corroborates my point. Shader power is what performance is asking for. And if there's something that Radeons have is shader power. Better said they have as much as 2x the theoretical peak shader power of Nvidia counterparts.

I didn't know how shaders overclock when overclocking the core, I assumed they did rise their clock but I doubt they do it in a linear fashion. Now I know it's not linear. Nevertheless you have to agree that 1860 (or 1836 for that matter) is higher than 1761, don't you?

Now if you want you can join me in the "more shader power, more performance" club, wich BTW is backed up by the 8800GT and many 3dmark records, for example lekamies case.

[trt740]

Originally Posted by Lekamies

Warm water is enough to cool down for breaking that barrier.
Here is pic after 3d mark'03 run include rivatuners hardware monitor.

I'm not impressed. THAT RESOLUTION NO WONDER!!!! and in a 4 1/2 year old bench. Lets see 3dmark06. As a matter of fact why don't you bench it to show me just how wrong I am.

[trt740]

Originally Posted by Lekamies

Warm water is enough to cool down for breaking that barrier.
Here is pic after 3d mark'03 run include rivatuners hardware monitor.

I'm gonna stop now because what your say is total garbage good luck. one more thing I still don't see a 3dmark06 19000 score with full screens cpuz, gpu clock etc. or for that matter a 8800 gts hitting 14000 with screen shot to back that up other than future marks. Future marks by it'self means zero. I want to see a completed test screen and all the other required screens I bet I won't see them.

[Tatty_One]

Originally Posted by DarkMatter

I have heard and read lot of people claiming big performance boosts out of shader overclocking alone, so I will keep my thoughts.

The fact that new cat 7.8 makes 2900 perform better only corroborates my point. Shader power is what performance is asking for. And if there's something that Radeons have is shader power. Better said they have as much as 2x the theoretical peak shader power of Nvidia counterparts.

I didn't know how shaders overclock when overclocking the core, I assumed they did rise their clock but I doubt they do it in a linear fashion. Now I know it's not linear. Nevertheless you have to agree that 1860 (or 1836 for that matter) is higher than 1761, don't you?

Now if you want you can join me in the "more shader power, more performance" club, wich BTW is backed up by the 8800GT and many 3dmark records, for example lekamies case.

I agree completely with the shader power, I think you are misunderstanding what I am trying to say regarding the 1761. The 1761 is not independantly overclocked, unlike the R600 series (where the shader clock is locked and fixed at a lower speed than Nvidia) on the G80, as you increase the core clock the shader clock automatically increases with it so, when a user sets his 8800GTS core speed to 863 Mhz he already has a shader clock speed of 1761, then he can go into rivatuner and overclock the shader clock seperatly (he has probably hit his core max already at 863mhz) without touching the core, so my point was the chances are his shaders are well beyond 1761 if he knows what he is doing, and to acheive that score on an 8800GTS suggest he does.....make sense? So I am saying the minimum it can be is 1761 and is likely to be much higher.....but of course we dont know, what we do know is though that you would have to expect it to be a lot LOWER because the difference in CPU speed and Core clocks is a lot and the faster (perhaps not the best terminology ) is actually the slower.

[ccleorina]

I just want to ask... Is ATI HD2900 have shader clock??? I cant see in GPUZ?

If the shader clock is locked? So what is the real shader clock?

Thanks.....

[yogurt_21]

I think it's tied to the stock core clock, and from what I've heard it doesnt change. even when you oc the core.

[Wile E]

Originally Posted by yogurt_21

I think it's tied to the stock core clock, and from what I've heard it doesnt change. even when you oc the core.

I wish somebody would make a good BIOS editor for R600.

[cefurkan]

Originally Posted by ccleorina

I just want to ask... Is ATI HD2900 have shader clock??? I cant see in GPUZ?

If the shader clock is locked? So what is the real shader clock?

Thanks.....

shader and gpu clock is same at ati

[Tatty_One]

Originally Posted by ccleorina

I just want to ask... Is ATI HD2900 have shader clock??? I cant see in GPUZ?

If the shader clock is locked? So what is the real shader clock?

Thanks.....

Yes it is locked and does not increase with core overclocking, the 8800GTS stock shader clock is at 1185mhz and will increase with core overclocks, it can also now be increase independantly in rivetuner the 2900XT shader clock runs at 800Mhz and stays there, one of the reasons why it has so many SP's, to compensate (although thats not the specific reason).

Just to bore you though Technically tho for practical purposes, although the 2900XT has 320SP's it actually has 64 groups of 5 shaders. Each group of 5 shaders can only run 1 thread each while each single shader in the 8800 can run 1. This means that the 8800GTX runs 128 shaders per clock and the 2900 runs 64. BUT! Each thread is worked on by 5 shaders and hypothetically can have 5 instructions ran per thread per clock. This equates to 320 intructions per clock versus the GTX's 128. While it is easy to divide the threads to all 64 groups of shaders, it is very difficult to keep all 5 shaders in each group working. This means that on a best case senario (all 5 shaders per group working to max) the 2900 does 2.5x's the work per clock than the 8800GTX and on a worse case senario it does half the work per clock than the 8800GTX. This means that the perfromance of the 2900XT GREATLY relys on the ability of the driver to distribute the instructions of the game being played. This is why we see poorish performance in some games and spectacular performance in others with the 2900XT. Future drivers can help this greatly.

[DarkMatter]

I understood your point of shader clocks been locked. And I understand why you assume it is even higher, it makes sense after all. Happens that I don't, but maybe I'm biased in this respect. I used to frequent a local forum of overclockers who's only goal was to achieve the higher clocks possible, or better said the higher conbined clocks they could reach, of course being it stable 100%, and they used to say it was common in their circles. The problem is that higher clocks not always translate into better performance. At some point there are current leakages, sinc problems between modules, etc. When this happens nobody knows what could happen and one posibility is that higher speeds give worse results, even if it's (or seems to be) totally stable. Ok, we all know this and I am talking about CPU overcloking for the most part, but the same applies to GPU, more in this case where core and shaders are asinc'ed. And although you have 3DMark to try the performance as you increase clocks, not everybody benches all speeds, they use only stability tests instead.

Another thing that I take into acount when doing my assumptions is this: the whole 8800 line is the same chip when manufacturing it, and then they choose wich one is wich model based on yields or demand. This is nothing new, we all know, but I don't see anyone paying attention to this as much as I think we should. There's a big difference between the models, as much that it could have been called a different chip. For example a chip that has been chosen as GTS because a defective ROP alone, is going to overclock a lot better than one that has been chosen because it couldn't reach GTX speeds as well as Nvidia wanted. This isn't new to anyone either, but I don't see anyone giving it the importance it deserves.

So my point about the thing was:
first, do we know for sure they are overclocking in a proper manner, so they get better performance? or on the contrary they are aiming for higher clocks only? (ok I didn't read the whole link, I just looked at screenies and little more so...) There are so many records or record claims out there with higher scores, but still lower clocks than those on your link, that I know where my two cents are going to stay by now.
second, can a specific 8800gts perform better at 720/1850 (core/shader) than other(s) at 860/1750? From my point of view it can.

Wow! That was a long post! I hope it explains my point of view. I will admit that I don't have personal experience with new cards, both Nividia 8 or HDs, since I'm out of this bussiness right now, but I do have some knowledge and I read a lot. So yeah I speak out of theory for the most part, and I have to rely on others experiences, and so I have to believe them. But I have learnt something on technology: Impossible is nothing (or was it adidas?)

[DarkMatter]

About the HD 2900 5 instructions per clock... I remember that when I saw all the specs and diagrams, my guess was that despite AMD's claims it was more likely that the chip wouldn't be able to effectively use both scalar and vectorial units at the same time. Can't remember why though, and from that day until now I look at the HD SPs as 4 ops/cycle units.

I'm too lazy to go back on reading all the stuff, so I would love your opinion. So please everybody share your thoughts on this.

[bigboi86]

Originally Posted by trt740

In this bench it does because part of the bench is cpu performance . Regardless your wrong a slow cpu can hold back a gpu it's called bottle necking.

That doesn't mean that the CPU will cause the graphics card not to overclock as much. They are completely different subsystems.

I am not wrong.

This has nothing to do with benchmarking.

[Lekamies]

Originally Posted by trt740

...or for that matter a 8800 gts hitting 14000 with screen shot to back that up other than future marks. Future marks by it'self means zero. I want to see a completed test screen and all the other required screens I bet I won't see them.

I had only this orb link to proof my 3d mark'06 score.

So i run it again at same clocks and take screenshot it for you.
Here

[trt740]

Originally Posted by Lekamies

I had only this orb link to proof my 3d mark'06 score.

So i run it again at same clocks and take screenshot it for you.
Here

Oh I already said i believe this I want to see the GTX doing 19000. Nice bench

[ccleorina]

Originally Posted by Tatty_One

Yes it is locked and does not increase with core overclocking, the 8800GTS stock shader clock is at 1185mhz and will increase with core overclocks, it can also now be increase independantly in rivetuner the 2900XT shader clock runs at 800Mhz and stays there, one of the reasons why it has so many SP's, to compensate (although thats not the specific reason).

Just to bore you though Technically tho for practical purposes, although the 2900XT has 320SP's it actually has 64 groups of 5 shaders. Each group of 5 shaders can only run 1 thread each while each single shader in the 8800 can run 1. This means that the 8800GTX runs 128 shaders per clock and the 2900 runs 64. BUT! Each thread is worked on by 5 shaders and hypothetically can have 5 instructions ran per thread per clock. This equates to 320 intructions per clock versus the GTX's 128. While it is easy to divide the threads to all 64 groups of shaders, it is very difficult to keep all 5 shaders in each group working. This means that on a best case senario (all 5 shaders per group working to max) the 2900 does 2.5x's the work per clock than the 8800GTX and on a worse case senario it does half the work per clock than the 8800GTX. This means that the perfromance of the 2900XT GREATLY relys on the ability of the driver to distribute the instructions of the game being played. This is why we see poorish performance in some games and spectacular performance in others with the 2900XT. Future drivers can help this greatly.

Thanks for nice info....

[Tatty_One]

Originally Posted by ccleorina

Thanks for nice info....

Always glad to help!

[cefurkan]

well i am upping this topic for

IQ compare

already both card are meaningless since

8800gt and 3870 on market

[Lionheart]

nvidia fanboyism to the max

[marsey99]

dead thread res to the max