My performance simulation of GeForce 9800GX2

[shogo_ca]

Wait, isn't the 9800 GTX nothing more then a 8800 GTX die shrink?

i think its more like a 8800GS 512mb die shrink (g92) instead of (G80) die shrink. But anyways, this thing does not deserve the *98XX name series, its not new architecture, we want new chip, we want G100 or g98 whatever the hell its called..

But still, nice little "preview" but this is based on what we know, im sure nvidia has something more than a die shrink on its hands to at least have a product that deserves its place in the high end market.

[kemo]

Some one should compare 7950GX2 to underclocked 7900GTX SLI to see if they will follow the same pattern

[thephenom]

Nice simulation, but I would imagine that 9800GX2 would only shine and be "30%" faster than Ultra only in AA/AF performance.

And with BioShock, I don't think it's a quality benchmark to show the performance difference since you have to test manually with FRAPs. Any possible chance to show CoH, WIC and Lost Planet with built-in benchmarks.

[Soulburner]

You can show framerate with Rivatuner.

[shogo_ca]

benchmark should be calculated in timedemos so the exact same frames are being rendered by different hardware, but still overall it gives a good overview

[thephenom]

You can show framerate with Rivatuner.

But BioShock still requires manually run of the app, it doesn't have timedemo function built in. So the results aren't exactly perfect since you have now introduced human errors in the equation.

[Soulburner]

But BioShock still requires manually run of the app, it doesn't have timedemo function built in. So the results aren't exactly perfect since you have now introduced human errors in the equation.

All you have to do is save the game at the point you which to measure performance.

Hit "Continue" and load up the game each time. Don't move, just take down the numbers. It is a very repeatable procedure, with no variables to account for unless a Splicer happens to walk on screen, which is unlikely. If that happened though, the test would be repeated.

[RedBull78]

Remember, this was just a simulation, not real results.

Thanks for reminding me. But just like with the last gx2 the performance was less then what 2x same cards did in pure SLI mode. So this simulation is probably withing %3 of what to expect when it finally comes out. My guess it'll cost around 599 dollars and no smart buyer would even consider buying this card considering the alternatives.

[Soulburner]

Thanks for reminding me. But just like with the last gx2 the performance was less then what 2x same cards did in pure SLI mode. So this simulation is probably withing %3 of what to expect when it finally comes out. My guess it'll cost around 599 dollars and no smart buyer would even consider buying this card considering the alternatives.

Which is why he underclocked both cards.

[thephenom]

All you have to do is save the game at the point you which to measure performance.

Hit "Continue" and load up the game each time. Don't move, just take down the numbers. It is a very repeatable procedure, with no variables to account for unless a Splicer happens to walk on screen, which is unlikely. If that happened though, the test would be repeated.

I guess that would be a test with little variant, but I personally still prefer timedemo that gives you a good average performance of the game being tested.

[DilTech]

9800gtx is a 8800gts g92 but with higher clocks afaik

Only 3 things are known about the D9E(which is assumed to be the 9800GTX).

1.) 512bit bus(this alone should make quite the difference)
2.) 32 rops(see 1). AA will be a breeze for this thing.
3.) 65nm

It's assumed to also be the 1-tflop core NVidia talked about last year.

I don't think that's a 8800GTS G92 with higher clocks. It may be based on that design, but the memory bus alone tells us it cannot be the same core.

As for the 9800GX2, I'm not thinking it's going to be the flagship from nvidia. Either that, or because NVidia doesn't fear the 3870x2 they're just sitting on the D9E for now and will release it as a 9900. Why play the ace of spades when the queen will still win the book?

[GAR]

nvidia loosing its touch? you got to be kidding.

[DilTech]

nvidia loosing its touch? you got to be kidding.

They aren't, this is just the effect of a lack of competition.

[KoHaN69]

you forgot one thing, the 9800gx2 will have 16lanes for 2 cards, so each gpu will have 8 lanes only, so itll be even slower than that

thats what i was thinking too..

i guess 2x 8800 in 8x PCIe could simulate that

[Nanometer]

cool simulation

[OBR]

you forgot one thing, the 9800gx2 will have 16lanes for 2 cards, so each gpu will have 8 lanes only, so itll be even slower than that

16x lanes PCIe 2.0 has bandwidth like t16+16 PCIe 1.1! Dont forget on PCIe 2.0! On this chipset and cards!

[Tim]

How Is This News?!!!!

[saaya]

Only 3 things are known about the D9E(which is assumed to be the 9800GTX).

1.) 512bit bus(this alone should make quite the difference)
2.) 32 rops(see 1). AA will be a breeze for this thing.
3.) 65nm

It's assumed to also be the 1-tflop core NVidia talked about last year.

I don't think that's a 8800GTS G92 with higher clocks. It may be based on that design, but the memory bus alone tells us it cannot be the same core.

As for the 9800GX2, I'm not thinking it's going to be the flagship from nvidia. Either that, or because NVidia doesn't fear the 3870x2 they're just sitting on the D9E for now and will release it as a 9900. Why play the ace of spades when the queen will still win the book?

well who says g92 doesnt have more rops and a 512bit bus?
the yields are probabaly just not there yet, hence they started with 8800gt, then 8800gts and soon the top parts will be 8800gtx g92 and the rest will be 8800gts and gt

[saaya]

16x lanes PCIe 2.0 has bandwidth like t16+16 PCIe 1.1! Dont forget on PCIe 2.0! On this chipset and cards!

but 8800 is already pciE 2.0 isnt it?
so its still the same... the gx2 will have half the chipset/ram/cpu bandwidth of a real sli solution

[generics_user]

but 8800 is already pciE 2.0 isnt it?
so its still the same... the gx2 will have half the chipset/ram/cpu bandwidth of a real sli solution

but no effect on performance, just look at the 8800gtx, pcie1.0 and similard scaling as 8800gt in SLI

pcie x8 2.0 = pciex16 1.0

i bet that the 9800gx2 will be the first card with lower performance on pcie 1.0 mobos

[takamishanoku]

I really don't understand why all the benches are without AA. I mean does anyone on here play with no AA if they could?
Cool simulation regardless. Nvidia are only competing with themselves at the moment.

[OBR]

benches updated ...

[saaya]

but no effect on performance, just look at the 8800gtx, pcie1.0 and similard scaling as 8800gt in SLI

pcie x8 2.0 = pciex16 1.0

i bet that the 9800gx2 will be the first card with lower performance on pcie 1.0 mobos

so you mean pciE 2.0 16 is too much bandwidth hence no scaling and pciE 2.0 8x is the same as 16x pciE 1.0 so it should be enough?

i dont think so... dont forget that the top scores with sli are with oced pciE bus to increase the b/w... so 16x pciE 1.0 is definately not enough to supply enough bw to even 1 card, let alone 2! hence 8x pciE 2.0 wont be anough for 1 card either... so 16x 2.0 wont be anough for 2.

all speculation, but thats using logic looking at results we know.
im pretty sure 9800gx2 will b slower than 2 8800gt/gts cards with the same specs and clockspeeds but each card having full 16x b/w

[OBR]

all speculation, but thats using logic looking at results we know.
im pretty sure 9800gx2 will b slower than 2 8800gt/gts cards with the same specs and clockspeeds but each card having full 16x b/w

Yeah, because that i underclocked cards ...

[jas420221]

so you mean pciE 2.0 16 is too much bandwidth hence no scaling and pciE 2.0 8x is the same as 16x pciE 1.0 so it should be enough?

i dont think so... dont forget that the top scores with sli are with oced pciE bus to increase the b/w... so 16x pciE 1.0 is definately not enough to supply enough bw to even 1 card, let alone 2! hence 8x pciE 2.0 wont be anough for 1 card either... so 16x 2.0 wont be anough for 2.

all speculation, but thats using logic looking at results we know.
im pretty sure 9800gx2 will b slower than 2 8800gt/gts cards with the same specs and clockspeeds but each card having full 16x b/w

According to several reviews, the 8x pcie 1.0 versions have plenty of bandwidth except for o/c'd Ultras and SLI configs. 16x and 16x X2 are PLENTY of bandwidth for every card under every condition possible. The best part about pcie 2.0 right now is the increase in power to the slot. Until cards start maxing out 16x bandwidth, the extra it gives is useless.

Kind of like ATI's decision to bring dx10.1 and sm4.1 in their cards now when its only a minor update......and an update that wont even be out until SP1 is officially released....and devs wont even be using these features until at least the end of 08 into 09.

Can anyone tell I HATED that ATI tried to market those features?!?!?