NVIDIA GF100 & Friends speculation

Thanks, it was just a typo, and we have almost the same saying german (Tool = Werkzeug). But in this case, both, tool or fool would have made sense IMHO, because there's another saying: Never attribute to malice what can be explained by stupidity - me obviously not excluded.

 

Hang on, so turning on 16xAF and 4xMSAA has totally killed GTX480, and now it is mostly slower than HD5870 when tessellation is heaviest?

What tessellation setting was that? It seems there are 3 settings for tessellation in Heaven 2.0.

According to:

http://www.hardwareluxx.de/index.ph...tx-480-im-test-nvidias-comeback.html?start=25

at those settings GTX480 is 32% faster than HD5870.

Why does the HD5870 graph spend long periods on exactly 30fps?

Jawed

 

Fishy, very fishy. Seems some kind of fps lock on cpu side to me. Though can't figure out which?

 

Or VSYNC.

 

But why would it happen only for one IHV's part, and that too, occasionally.

 

When every frame takes 1/60th to 1/30th second to render you'd see 30 FPS. Maybe GTX480 just fluctuates more?

 

Not sure. Could have been they messed up the driver settings or something.

 

From that graph, I get the feeling that if 5870 weren't capped at 30 fps for most of the run it would have ended up being faster than Fermi at those settings. Especially when the lows were never as drastic as 480.

Interesting bit of an oddity there.

Regards,
SB

 

That definitely looks like fps clamping. Very strange.

 

Seems like GTX480 is doing extreme tesselation and 5870 lower tesselation mode. Going by pretty much all other benches GTX480 smokes 5870 and puts itself head to head with 5970 in Uningine with tesselation enabled in any tesselation mode mode.

http://www.hardwarecanucks.com/foru...s/30297-nvidia-geforce-gtx-480-review-21.html

1920x1200 4xAA, 16xAF, tesselation on, GTX480 min:26 avg:36.9 and 5870 min:14 avg:25.3

http://www.pcgameshardware.com/aid,...Fermi-performance-benchmarks/Reviews/?page=16

1920x1200 4xAA, 16xAF, tesselation on (extreme), GTX480 min:18 avg:30.3 and 5870 min:7 avg:17,2

PCGH also has a diagram over how the framerate unfolds during bench time.

 

Must have been the secret NV edition version.

 

Other sites have replicated the previous results, though. Guru3D appears to be an outlier.

 

Yeah, I even posted links to 2 reviews showing how GTX480 greatly pulls ahead in tesselation regarding Uningine (pretty much all reviews shows the same outcome) in post above. One review even has a diagram over framerate during bench. Would be good if a summary is made of all review sites results for Uningine + settings to not be overwhelmed with FUD and cherry picked images. Min, avg, max and if possible diagrams for bench time.

 

It's so easy to do it using Fraps that it should be more common really. I can only think of PCPerspective off the top of my head that does it consistently. Everybody else relies on line and bar graphs.[H] tries but their approach is too muddled and inconsistent to be reliable.

 

http://www.hardwarecanucks.com/foru...s/30297-nvidia-geforce-gtx-480-review-21.html

Spoiler

http://www.bjorn3d.com/read.php?cID=1831&pageID=8804

Spoiler

http://www.hardwareheaven.com/reviews.php?reviewid=950&pageid=22

Spoiler

havent gone through all reviews but just included those showing atleast min, avg and if possible diagram (previous page for one of them). Tesselation seems to be Fermi's strong point, no?

 

Metro 2033 is interesting being a game with lots of state of the art tech and cappable of bringing high-end GPUs to their knees. Quite a difference in most cases.

http://www.tomshardware.com/reviews/geforce-gtx-480,2585-12.html

Spoiler

http://www.hardwarecanucks.com/foru...s/30297-nvidia-geforce-gtx-480-review-20.html

http://www.hardwareheaven.com/reviews.php?reviewid=950&pageid=13

Spoiler

 

Large die, high leakage = high static power.

 

Likely not enough VSS/VDD pins for the memory controller. The VSS/VDD pin matrix does two things, isolates the signals/prevents coupling and provides more stable power with less ripple/overshoot/droop/etc for the analog side. But yeah, something is seriously wrong with their MC design.

 

not leakage but coupling between signals/VSS/VDD and ripples/droops/overshoots. Hopefully, given his background, Nvidia's new CTO can teach them the black magic of high speed signaling its going to be painful when they try to do the next gen of PCIe.

 

That cannot be the only reason, because GTX470 is only consuming ~60% of GTX480's idle power, while the die itself is obviously the same.