gtx 480 @ 857 core is really impressive!

[yh125d]

It's good news for everyone, surely, as it increases the probability that something genuinely impressive might come from Fermi a little further down the line, with a die shrink, no?

Quite. G80 begat G92, arguably the most successful GPU ever, and the debacle known as R600 (which has a lot of similarities to fermi) led to RV770 and RV870, arguably the most successful ATI GPUs

 

[Madcatatlas]

it may have to do with games not using the performance overhead that is already there. Also known as bad coding. Look at consoles, in particular the ps2, the games at the end of its lifecycle were better looking than the ones in the start, all with the same hardware.

 

[bunnyfubbles]

Yea If I was one of the 3% of the world gaming at 2500x1600 i'd worry.

Back on topic .....


How bout that scaling!

The purpose of my post was how well the gtx scaled when overvolted and overclocked.

I compared a overclocked gtx 480 to a 5970.
It's the only thing it compares to, right?

98% of us use 1900x1080 or lower, so stop blowing smoke.

I could show you bechies of overclocked sli gtx 480's beating crossfired 5970's, so what's your point?

This is not a ATI vs gtx thread. We all know the gtx is faster.

that's 101%

 

[jbh545]

I got my SLI pair of GTX 480 up to 870mhz each and they're not even all that hot with the fans I put in the case. Just keep fans blowing across the heatpipes and you can do this on stock cooling.

 

[Keysplayr]

I got my SLI pair of GTX 480 up to 870mhz each and they're not even all that hot with the fans I put in the case. Just keep fans blowing across the heatpipes and you can do this on stock cooling.

What is your voltage and GPU fan speeds?
And did you use Afterburner with Kombuster or something else?

 

[happy medium]

that's 101%

The actual % was 2.5% . I was just splitting the difference.

 

[jbh545]

What is your voltage and GPU fan speeds?
And did you use Afterburner with Kombuster or something else?

just the afterburner beta. they are volted to 1.1. The temps change a lot from game to game. Assassins Creed 2 at 8X AA 2560x1600 with vsynch on barely even stresses them so they're like 75c at 70% fan. The hottest I saw on on a wide selection of games is 87c at 88%. I don't find them substantially louder than my 5970+5870 at 875 core but they are about 10c hotter for a given noise level. A 50% 5970 fan is louder than a 80% GTX 480 fan.

they could probably take more voltage and clock higher but i've got the whole setup plugged into a 980watt UPS which also has a 140watt 30 inch monitor plugged in. With the cards volted to the max allowed for now which is 1.125 (and supposedly perfectly safe according to the manufacturers) the overload light would occasionally flicker on in games that used the cards at 99% and had a heavy CPU load (so far just GTAIV). Taking out the monitor and the 80% psu efficiency means that two of these cards overvolted and a 4.1 ghz i7 920 were pulling over 672w from the PSU while gaming. So, they get pretty power hungry, but a quality 1000w like my HX1000 can handle a pair even overvolted while staying in the good part of its efficiency curve. Furmark would probably max the PSU though...I'd have to plug the PC straight into the wall for that.

 

[citan x]

Yea If I was one of the 3% of the world gaming at 2500x1600 i'd worry.

Matches the 3% of the world gaming with an overclocked GTX480.

 

[tweakboy]

Very nice 857mhz what is it at stock ? What is the 3dmark 06 we will thx

 

[SlowSpyder]

It's good news for everyone, surely, as it increases the probability that something genuinely impressive might come from Fermi a little further down the line, with a die shrink, no?

Can anyone explain to this neophyte, though, why a gpu might _not_ 'scale' with increasing clock speed?

What are the factors that prevent you simply increasing a gpu's performance by x% by simply increasing the clock speed x% (assuming its not memory limited, or assuming you can increase the memory x% also)?

Are there parts of the GPU's pipeline that are fixed speed and don't simply increase with clock speed then?

Yea, I think a refined/tweaked Fermi card will be a beast. In it's current incarnation it leaves a bit to be desired in my opinion. No doubt future GPU's based on it will be great performers.

 

[tincart]

Certainly looks like good scaling. Looks to be a great enthusiast card for people to toy around with, install a water block, etc. I wonder how frequent an overclock like that is going to come up.

Hexus tops out at 815 with a waterblock.
Tweaktown managed 841 with the fan @ 100%

A quick survey gives various other results, some impressive with stock air and some not so impressive with after-market cooling. That's part of the fun, isn't it?

 

[Genx87]

Might be some nice head room in this arch once the process situation is fixed.

 

[yh125d]

I predict Fermi successor at 28nm will be awesome

 

[dust]

I predict Fermi successor at 28nm will be awesome

As long as it brings the price on the 5870 down.....

 

[StrangerGuy]

I predict Fermi successor at 28nm will be awesome

If Fermi 40nm was already such a PITA to manufacture, you really think an even smaller 28nm node is going to be easier? And this is considering TSMC is skipping the 32nm altogether, so it's an even larger jump compared to before.

 

[yh125d]

If Fermi 40nm was already such a PITA to manufacture, you really think an even smaller 28nm node is going to be easier? And this is considering TSMC is skipping the 32nm altogether, so it's an even larger jump compared to before.

I didn't say anything about how easy it's be to manufacture, but the 40nm problems aren't necessarily indiciative of upcoming problems at 28nm, NV just wasn't prepared for 40nm problems the way ATI was. NV won't likely make the same mistake of huge die + considerable GPU overhaul + not doubling vias etc twice in a row

 

[Lonyo]

If Fermi 40nm was already such a PITA to manufacture, you really think an even smaller 28nm node is going to be easier? And this is considering TSMC is skipping the 32nm altogether, so it's an even larger jump compared to before.

1) It's a big die, that makes it difficult to manufacture.
2) The process itself had problems, which was one of the major issues with the whole thing.
3) The design from NV didn't try and alleviate those problems. There was an article on AT which talks about what ATI had to do with their architecture (for the HD5870/etc) in order to make sure yields were good.

On a GOOD 28nm process, many of those issues will go away. On a bad one they might try planning to alleviate the negative impact of them more than they did for 40nm.
And die size might change depending on what approach they take, for instance if they decide to only increase functional units by 50% and then ramp up clock speed, meaning the die would be smaller.

Also TSMC skipped '45nm' and went to 40nm, so it was 55nm -> 40nm, and now it's 40nm -> 28nm, skipping 32nm. Exactly the same step as the last one.

 

[yh125d]

Also TSMC skipped '45nm' and went to 40nm, so it was 55nm -> 40nm, and now it's 40nm -> 28nm, skipping 32nm. Exactly the same step as the last one.

TSMC 40nm GPU node is viewed as the same node level as Intel/TSMC 45nm CPU process. 55nm to 40nm didn't skip anything inbetween

 

[MrK6]

Fermi is a good design, however the execution and realization of the design was just bad, and shows how much each step in the whole process of delivering a product to market counts. I usually use one of the Crysis games to compare raw horse power of graphics cards, and in that review while the GTX 480 is only a bit faster than the 5870, overclocked it's closer to 30%. I imagine this is the performance level NVIDIA was shooting for and would have liked to release the GTX 480 at if it was at all feasible.

As others have mentioned, the power consumption/heat and yields need to be improved. What future revisions of the chip hold, who knows.

 

[Genx87]

If Fermi 40nm was already such a PITA to manufacture, you really think an even smaller 28nm node is going to be easier? And this is considering TSMC is skipping the 32nm altogether, so it's an even larger jump compared to before.

They are having issues with yield. Others are power and heat. Power and heat can be cut with a smaller process. Yield can be helped due to more die per wafer. Even if they manage to get the same yield % they will get more die per wafer due to the smaller size.

 

[Madcatatlas]

They are having issues with yield. Others are power and heat. Power and heat can be cut with a smaller process. Yield can be helped due to more die per wafer. Even if they manage to get the same yield % they will get more die per wafer due to the smaller size.

lol.. for the sake of nVidia i hope that is not the scenario they will see. I do hope they do something to have better yields regardless of process size and it would be only natural with more hands on time with the new architechture

 

[Genx87]

lol.. for the sake of nVidia i hope that is not the scenario they will see. I do hope they do something to have better yields regardless of process size and it would be only natural with more hands on time with the new architechture

I am sure they will get better yields. But I was stating an example where even if they didnt there is still benefit in the smaller process.

 

[SolMiester]

Fermi is a good design, however the execution and realization of the design was just bad, and shows how much each step in the whole process of delivering a product to market counts. I usually use one of the Crysis games to compare raw horse power of graphics cards, and in that review while the GTX 480 is only a bit faster than the 5870, overclocked it's closer to 30%. I imagine this is the performance level NVIDIA was shooting for and would have liked to release the GTX 480 at if it was at all feasible.

As others have mentioned, the power consumption/heat and yields need to be improved. What future revisions of the chip hold, who knows.

And if they could of release with all 512 shader units too!