NVIDIA GF100 & Friends speculation

[KimB]

People know what PhysX is. If they want it they know that ATI doesn't support it and they don't need a review to tell them that.

True enough, but if PhysX adds performance, then I would expect that nVidia users would enable it by default. If it doesn't add performance, then it's actually hurting nVidia by only benchmarking with it enabled. So I don't exactly see the problem here.

 

[Kef]

Charlie's claims of clocks and 5% is toast.

Don't jump the gun ...

 

[Mintmaster]

Funny notion:
People dislike that PhysX is on in reviews, but I heard no such protest when reviews did stuff like this:
http://www.hardocp.com/article/2009/12/23/dirt_2_gameplay_performance_image_quality/4

I see. So what you're saying is that we should enable PhysX on NVidia cards (forcing a lower resolution) and keep it off on ATI cards, and just consider the differences in the gaming experience.

I have no problem with that.

 

[Lonbjerg]

I see. So what you're saying is that we should enable PhysX on NVidia cards (forcing a lower resolution) and keep it off on ATI cards, and just consider the differences in the gaming experience.

I have no problem with that.

Neither do I

 

[MfA]

Ok, so you tell me what you think "globallycoherent" does when all of the accesses to a UAV are unordered by definition? Why does it matter if you "flush the writes across the chip" if nothing else is guaranteed to be running at the same time and your groups can run in arbitrary order

Lets say we are inserting data into an ordered tree ... the fact that the groups aren't ordered is irrelevant, but each still needs coherent data if they are going to be inserting data in the same tree.

 

[Sontin]

Charlie's claims of clocks and 5% is toast.

20% faster with higher clocks and a lower TDP. Yes, he was very accurate in the last months.

 

[DavidGraham]

That's conclusion is a bit interesting - allows NV to say 20% faster with a reference when it's nowhere close to 20% faster unless you get up to 25x16 with AA and AF cranked. At more mainstream resolutions (19x12) it's more like 5-10% faster. I suppose one doesn't go after a 480 with a paltry 19x12 monitor?

No , at 19x12 the difference is the same too .. if not bigger ..
by the way GTX480 dominates heaven v2.0 at extreme tessellation !

 

[Kef]

20% faster with higher clocks and a lower TDP. Yes, he was very accurate in the last months.

Singing a bit too early aren't we?

 

[DavidGraham]

for those who can't access hexus , here is another source :

http://translate.google.com/transla...rabhardware.net/forum/showthread.php?t=165350

 

[Mize]

No , at 19x12 the difference is the same too .. if not bigger ..
by the way GTX480 dominates heaven v2.0 at extreme tessellation !

Well I went through the whole review before hexus.net took a nosedive and that's not my recollection.
It was faster in Crysis only at the highest res, faster in DIRT2 only with DX10 or the highest res, about the same in BFBC2...

In a few hours we'll have more conclusive data, but I sure didn't see 20% on most of those graphs.

I won't argue the tessellation as I didn't get to see the uniengine bench page (I always go to game benches first), but I suspect the 480 is a monster for tessellation.

 

[psychoticdream]

all the benches from HExus

http://www.computerbase.de/forum/showpost.php?p=7597425&postcount=3418

 

[eastmen]

That's conclusion is a bit interesting - allows NV to say 20% faster with a reference when it's nowhere close to 20% faster unless you get up to 25x16 with AA and AF cranked. At more mainstream resolutions (19x12) it's more like 5-10% faster. I suppose one doesn't go after a 480 with a paltry 19x12 monitor?

If true ati will drop prices a bit and make it look worse than 20% faster for 40% more money.

Also no one knows what drivers were used for benches or settings. I can't wiat to read it tho

 

[RobertR1]

all the benches from HExus

http://www.computerbase.de/forum/showpost.php?p=7597425&postcount=3418

Awesome! Hexus wasn't fast enough to pul the reviews

 

[Andrew Lauritzen]

Lets say we are inserting data into an ordered tree ... the fact that the groups aren't ordered is irrelevant, but each still needs coherent data if they are going to be inserting data in the same tree.

Right but this always requires atomics, which are always globally coherent regardless of the specifier by my reading of the docs...

 

[Lonbjerg]

http://forums.anandtech.com/showpost.php?p=29583762&postcount=21
http://forums.anandtech.com/showpost.php?p=29583764&postcount=22
http://forums.anandtech.com/showpost.php?p=29583768&postcount=23

 

[jimbo75]

20% doesn't nearly tell the whole truth, for example it loses to the 5870 in crysis in two benchmarks, but "wins" the last by 1.4fps or 15%....

 

[DavidGraham]

Well I went through the whole review before hexus.net took a nosedive and that's not my recollection.

May be you are right , I honestly didn't calculate the difference .. may be someone else can ..

 

[Kef]

Hexus has _always_ been biased to NV. Wait for the "real" benchmarks and not some teenagers dreams, will you? After Anand, Tech report and HardOCP we will have some more up to date figures.

 

[ChronoReverse]

20% doesn't nearly tell the whole truth, for example it loses to the 5870 in crysis in two benchmarks, but "wins" the last by 1.4fps or 15%....

I see now that Crysis will soon fall by the wayside as a benchmark because of that [/snark]

Hexus has _always_ been biased to NV. Wait for the "real" benchmarks and not some teenagers dreams? After Anand, Tech report and HardOCP we will have some more up to date figures.

If this is the result of "nvidia bias" then the GTX480 is going to be rather underwhelming.

 

[Mize]

May be you are right , I honestly didn't calculate the difference .. may be someone else can ..

Don't get me wrong - 15% faster for $75 more than a 5780 + monster tessellation is a perfect slot between the 5780 and the 5980.