Next NV High-end

[Ho_Ho]

Since when exactly is a P4@3.8GHz exactly a slouch?

If it is fast then why are there almost no P4's in futuremark top50?

 

[Ailuros]

If it is fast then why are there almost no P4's in futuremark top50?

That's not really answering my question now is it? If you want to side-flop come up at least with a more serious argument than that. Is a high end gaming system at a serious disadvantage with a P4@3.8GHz yes or no?

And before anyone says anything I've myself an Athlon FX but that's besides the point. Even more to the bullshit above I get with 7800GTX@500/685MHz and an FX-55 something over 8700 points in 3dmark05. Now recheck the results above and tell me what exactly you guys are refering to.

 

[KimB]

I'd say that the performance of the P4 3.8GHz is quite good. You really can't do that much better with an Athlon 64, and especially not for games when you're typically going to be GPU-limited anyway.

The disadvantages with the P4 are in price and power consumption, and neither is going to be correllated with benchmark results.

 

[ANova]

The difference between a high end P4 and a high end A64 in gaming is typically only a couple frames per second with the maximum being around 10 fps.

I would imagine the G70 is designed to operate at less then 500 MHz, anything higher greatly reduces its MTBF.

 

[Ailuros]

I would imagine the G70 is designed to operate at less then 500 MHz, anything higher greatly reduces its MTBF.

As I said even if you use some exotic cooling measures to get it way beyond that, if you don't manage to get an at least equivalent increase in memory frequency it won't help that much.

 

[KimB]

Well, there are benchmarks that suggest that merely increasing the memory frequency of the GTX will allow it to at least match the X1800 XT. But given the large number of "OC" versions of the GTX, we know that nVidia could possibly release a card with default higher clocks than the stock GTX.

 

[Ailuros]

Well, there are benchmarks that suggest that merely increasing the memory frequency of the GTX will allow it to at least match the X1800 XT. But given the large number of "OC" versions of the GTX, we know that nVidia could possibly release a card with default higher clocks than the stock GTX.

Since I've ran extensive tests with higher frequencies on the G70 merely increasing the ram by 15% gives you in most occassions about 5% more performance. Overclock both core and ram by 15% f.e. and you get damn close to +15%.

 

[Geo]

Since I've ran extensive tests with higher frequencies on the G70 merely increasing the ram by 15% gives you in most occassions about 5% more performance. Overclock both core and ram by 15% f.e. and you get damn close to +15%.

And here we thot this gen would be grossly bandwidth limited. That doesn't suggest it re G70.

 

[Ailuros]

And here we thot this gen would be grossly bandwidth limited. That doesn't suggest it re G70.

I guess that part of the reason why it doesn't show itself as much could be that it handles it's bandwidth quite different than NV40. Even more so since the NV40 wasn't "fine-tuned" as much for ultra-high resolutions.

What would be interesting to see on a purely theoretical level, is how much in performance a R520 loses when you downclock it's memory frequency; I wouldn't be surprised if the difference to default clocks is marginal too.

 

[KimB]

What would be interesting to see on a purely theoretical level, is how much in performance a R520 loses when you downclock it's memory frequency; I wouldn't be surprised if the difference to default clocks is marginal too.

This has already been shown, hasn't it? Where a R520 was clocked to the same level as a G70 with 2 quads disabled, and the result was that the R520 lost (typically).

But it has been stated that it is quite possible that this analysis is not entirely valid because the R520's memory would be running at higher CAS latencies.

 

[Ailuros]

This has already been shown, hasn't it? Where a R520 was clocked to the same level as a G70 with 2 quads disabled, and the result was that the R520 lost (typically).

But it has been stated that it is quite possible that this analysis is not entirely valid because the R520's memory would be running at higher CAS latencies.

Wasn't that a theoretical case where a X1800XL has been used? I'm not talking about such a case at all. I'm talking about a theoretical case where you'd take a X1800XT@625 and downclock it's ram to 650MHz or lower if possible.

 

[trinibwoy]

Wasn't that a theoretical case where a X1800XL has been used? I'm not talking about such a case at all. I'm talking about a theoretical case where you'd take a X1800XT@625 and downclock it's ram to 650MHz or lower if possible.

Isn't that exactly what they did? (in addition to downlclocking the core to 450)

 

[Ailuros]

Did they leave the core on a X1800XT at 625 and downclocked only the ram? (honest question)

 

[Pete]

No, I think DH used an X1800XL, and so the memory was 500MHz to begin with. At least, that's what I get from their before and after underclocking screenshot. IIRC, they used an X1800XL and downclocked the core, and a 7800GT and disabled a quad and upclocked the core (a bone of contention with me, as I don't recall 450MHz being a valid clock stepping for G70, and the VS remain clocked higher).

 

[Ailuros]

No, I think DH used an X1800XL, and so the memory was 500MHz to begin with. At least, that's what I get from their before and after underclocking screenshot. IIRC, they used an X1800XL and downclocked the core, and a 7800GT and disabled a quad and upclocked the core (a bone of contention with me, as I don't recall 450MHz being a valid clock stepping for G70, and the VS remain clocked higher).

My memory didn't betray me after all then.

What I'd like to know is how much of performance a X1800XT loses when you downclock only it's memory, while leaving the core at 625MHz.

 

[IbaneZ]

What I'd like to know is how much of performance a X1800XT loses when you downclock only it's memory, while leaving the core at 625MHz.

It'll lose a lot of performance. The XL feels pretty weak at 575/550.

 

[Ailuros]

It'll lose a lot of performance. The XL feels pretty weak at 575/550.

Hmmm ok that's still 50MHz below XT core speeds, but pretty close to what I'm asking about. How high would you rate the performance difference on average to a X1800XT@default frequencies?

 

[IbaneZ]

How high would you rate the performance difference on average to a X1800XT@default frequencies?

Hard so say since i don't have an XT.

But i'm sure my XL would be a lot happier with more memory bandwidth. Unfortunately i can't OC the mem more than 550 atm.

 

[Moloch]

Hard so say since i don't have an XT.

But i'm sure my XL would be a lot happier with more memory bandwidth. Unfortunately i can't OC the mem more than 550 atm.

When the XT comes out, buy one and swap the memory chips and return it

 

[Pete]

My memory didn't betray me after all then.

Mine did, a little. It was a GTX and an XL. (I don't recall DH clearing up the GTX's actual clocks, though.)