NVIDIA GF100 & Friends speculation

Hmm then why would they state time and time again, its going to be the fastest graphics card :), I would think its more then just Unigine where they are leading.

I said where the lead is "significant". Leading by 5~10% is still leading, but I wouldn't call it significant.
 
I said where the lead is "significant". Leading by 5~10% is still leading, but I wouldn't call it significant.


That's true. But I'm still inclined to believe what we have seen so far all official numbers, are the gtx 470, outside of that one video of Heaven. So if the gtx470 is truly right around the HD5870 guess what, 5-10% is not where the 480 is going to be.
 
It doesn't matter what you should have said :LOL:, I'm talking about that part of the video where tesselation is turned on and off. It may allow to assess the performance drop for this obscenely tesselated case for both architectures.

So anyone out there dares to tell us how a HD5870 fares there (scene from video at 1:48 - 1:58) with tesselation on and off?

I tried, but have problems finding the time of day. It seems like the day is not moving (as it should, some shade moving) in the video. The time of day is heavy on FPS if AO and all is on.
 
Last edited by a moderator:
That's true. But I'm still inclined to believe what we have seen so far all official numbers, are the gtx 470, outside of that one video of Heaven. So if the gtx470 is truly right around the HD5870 guess what, 5-10% is not where the 480 is going to be.

Depends on clocks and whether the 480 is indeed 512 or 480 units, as current rumors are now picking up steam on
 
Did Gigabyte already give away their secret to that? :p
Haha, Dave ...

Dont burn the trolls!

Well 3rd player won't see Intel for a while :D,

I don't think the gtx 480 is only 10-15% in real games, probably will be a little bit more, since the gtx 470 looks to be around the hd5870. To me what we saw in all those white papers was actually the gtx470 benchmarks, if we look at the unigine benchmarks in the white paper it showed 1.5 to 1.8 times faster, but in this latest one, it looks to be 1.5 to 2.0 possibly a little bit more times faster in those same 60 seconds.

Also if you aren't pushing these cards what the use of even looking at them.
Thats quite a changed tone there .. I dont like it.

I think the die argument is pointless - its better than before means squat when you dont know the yield numbers.

And I think NV harping on the Uniengine benchmark is getting tiring. The other down side to it is probably ATI is working their asses off improving performance in this synthetic benchmark rather than the game. So its a two pronged issue that I have here.
 
That's true. But I'm still inclined to believe what we have seen so far all official numbers, are the gtx 470, outside of that one video of Heaven. So if the gtx470 is truly right around the HD5870 guess what, 5-10% is not where the 480 is going to be.

I'm not sure what benchmarks you're referring to, since we have seen several coming from different sites, but most of them point to the GTX 470 being a bit slower than the 5870 on average, Heaven aside*.

As for the gap between the 470 and the 480, well, the GTX 280 was 10~15% faster than the GTX 260-216, so if history is any indication...

Plus, for what it's worth, 280-260 = 20; 480-470 = 10, so that would point to a smaller gap, which is consistent with the gap between the GTX 275 and 285. Then again, NVIDIA's naming scheme seldom makes any sense, so I probably shouldn't be reading that much into it.


*Heaven aside. That sounds nice, could be a good name for a band, a video game or something.
 
Version 1.0 of Heaven:

http://unigine.com/download/

NVidia is using the unreleased version 1.1 of this benchmark. So it turns out we cannot make a direct comparison.

Jawed

so wait.. iirc the numbers being quoted for the ATI 5870 are the same from Decemberish when 1.0 was used.. so are the numbers for the GTX480 depicted in the video from Dec (1.0) or using the updated 1.1 demo ? Also of note is no mention of drivers.. (though I don't recall any recent Catalyst performance reviews.. ie 9.10 v 9.11 v 9.12/HF v 10.1 v 10.2 and if any subsequent numbers have increased due to driver updates, though it wouldn't make that big a difference (if +20% perf .. ie 40FPS min now 48FPS vs NV's 78FPS it's still "losing").

One area I am in agreeance with Razor is that while uengine (may not represent "real" world performance) it should show potential performance..
 
Plus, for what it's worth, 280-260 = 20; 480-470 = 10, so that would point to a smaller gap, which is consistent with the gap between the GTX 275 and 285. Then again, NVIDIA's naming scheme seldom makes any sense, so I probably shouldn't be reading that much into it.
That gap's small because there's a heavily cut-down GTX460 coming along :runaway:
 
I'm not sure what benchmarks you're referring to, since we have seen several coming from different sites, but most of them point to the GTX 470 being a bit slower than the 5870 on average, Heaven aside*.

As for the gap between the 470 and the 480, well, the GTX 280 was 10~15% faster than the GTX 260-216, so if history is any indication...

Plus, for what it's worth, 280-260 = 20; 480-470 = 10, so that would point to a smaller gap, which is consistent with the gap between the GTX 275 and 285. Then again, NVIDIA's naming scheme seldom makes any sense, so I probably shouldn't be reading that much into it.


*Heaven aside. That sounds nice, could be a good name for a band, a video game or something.

Maybe NV is planning on a further reduced (sp/clock) GTX460. so they 480 -10 = 470 -10 = 460 and when B2 arrives we'll see the GTX485 ;)

edit: damn it Jawed types faster then me

edit2: so then we can look forward to g92d GTS350, GT213 GTS 345, GT215 340, G92c GT330 (gotta get rid of these somehow ;) ).. etc
 
HD5850 Unigine scores

Well, just download the Unigine Heaven benchmark version 1.0 (forgot I could do DX11 now I'm on Win 7, criminal I know :p).

I got the following using free roaming camera looking at the Dragon at approx same distance as "Nvidia Guy".


With tessellation : 17 fps steady (rotated around the dragon didn't seem to change)

Without tessellation : 41 fps steady (again rotated around the dragon and no change to fps).


So thats roughly a 59% hit when enabling tessellation (right?? my high school maths fails me here)


Settings where no AF or AA, all other settings as high as they could go and at 1920x1200 resolution.

HD5850 was also OCed to max ATI overdrive settings (775hmz core & 1125 mhz memory).

Rest of system :

Catalyst version : 10.2
Core i7 920 @ 3.4ghz
6GB PC3-10666 RAM @ 1500mhz (or thereabouts)
2 x 36GB raptors in RAID 0
 
so wait.. iirc the numbers being quoted for the ATI 5870 are the same from Decemberish when 1.0 was used.. so are the numbers for the GTX480 depicted in the video from Dec (1.0) or using the updated 1.1 demo ?
Good question. Someone volunteer to pick that apart... It might be that 1.0 and 1.1 actually have the same performance...

Jawed
 
With tessellation : 17 fps steady (rotated around the dragon didn't seem to change)

Without tessellation : 41 fps steady (again rotated around the dragon and no change to fps).


So thats roughly a 59% hit when enabling tessellation (right?? my high school maths fails me here)
GTX480 has 43fps with tessellation and 56fps without. So about a 23% drop.

Settings where no AF or AA, all other settings as high as they could go and at 1920x1200 resolution.
The GTX480 was tested at 1920x1080, but I doubt resolution makes a notable difference.

HD5850 was also OCed to max ATI overdrive settings (775hmz core & 1125 mhz memory).
So a stock HD5870 with tessellation off would be about the same speed as GTX480 with tessellation on. That's pretty impressive.

Just a minor quibble remains, now, over version 1.1 versus 1.0...

Jawed
 
Does this apply to both tesselation enabled and disabled? Would mean the 5870 could about keep up with GTX480 without tesselation in that benchmark. Since it's claimed cat 10.3 should boost tesselation performance we should probably wait for this before drawing any conclusions. That said, I've no doubts that the GTX480 _is_ faster with tesselation, if not the whole distributed geometry processing would be completely pointless...
 
GTX480 has 43fps with tessellation and 56fps without. So about a 23% drop.


The GTX480 was tested at 1920x1080, but I doubt resolution makes a notable difference.


So a stock HD5870 with tessellation off would be about the same speed as GTX480 with tessellation on. That's pretty impressive.

Just a minor quibble remains, now, over version 1.1 versus 1.0...

Jawed

That looks quite good yes.

From what we've seen so far, AA might be the rod that breaks Fermi's back though. Whether the 470 or 480, there does appear to be a raw power in Fermi that supercedes Cypress, but the 470 was seen to fall down at 8xAA in unigene and showed what can only be called a capitulation in the Crysis benchmarks shown earlier in this thread (faster than xfire 5870's at 0xAA and slower than a 5850 with 8xAA). I have this feeling that the 480 will stumble in a similar fashion.

I'd also be a lot more impressed if both running benchmark versions were the same.
 
Last edited by a moderator:
From what we've seen so far, AA might be the rod that breaks Fermi's back though. Whether the 470 or 480, there does appear to be a raw power in Fermi that supercedes Cypress, but the 470 was seen to fall down at 8xAA in unigene. I have this feeling that the 480 will stumble in a similar fashion.
Hmm, well AA performance prolly hurts on GTX470 due to the bandwidth deficit. GTX480 should have 60%-ish more bandwidth I think.

I'd also be a lot more impressed if both running benchmark versions were the same.
It does seem Petersen's risking quite a scandal if the graph has HD5870 on version 1.0 and GTX480 on 1.1.

Jawed
 
It's terribly vague, and I'd ignore the number and just concentrate on the fact the new bench is, hopefully, some measure quicker than the current one on everyone's hardware. Could be overall, could be in one situation in one portion of one subset of frames on one driver for one SKU on one day of the week.

I won't get to talk to Dan at GDC (I'm not going sadly), but they'll announce it there so it's not long to wait.

And while I'm posting, the thread's way too Heaven-centric at the moment (it's not a game!) and too conversational. Stick to something with cred, not NV's cherry-picked stuff on non-final configs on a version of a non-game you can't determine.

Will lock the thread if needs be, to clean it up. Help me not waste my time doing that.
 
Back
Top