NVIDIA GF100 & Friends speculation
- Thread starter Arty
- Start date
DavidGraham
Veteran
I don't see how that game performs better on Geforce hardware ?
Far Cry 2 :
HD 4870X2 is better than GTX 295
HD 4890 is equal to GTX 285
I sense an unbelievable amount of prejudice toward Nvidia and Fermi , I don't know why .. but people are trying to strip the card of any expected advantage , I think that should stop .
Far Cry 2 :
HD 4870X2 is better than GTX 295
HD 4890 is equal to GTX 285
I sense an unbelievable amount of prejudice toward Nvidia and Fermi , I don't know why .. but people are trying to strip the card of any expected advantage , I think that should stop .
Hm, you are the guy who missrepresenting the benchmark. I see a very nice performance increase over GTX285. And this is more than AMD delivered with Cypress in this farcry 2 benchmark.
And you don't show all of your tricks in a preview.
Shoot me for being a consumer
I don't care so much about the PR spin of it. The Fermi benchmark represents a product that will be available later. Farcry 2 is a game I would expect that Fermi would shine in, even if it otherwise would be equal to the 5870. I had higher expectations towards what Fermi could do compared to existing offerings like the 5870. GTX285 isn't even a viable option in my opinion.
Nvidia probably don't want to show all its tricks in a preview, but I would expect that it had something to show. I mean, they called in for this preview, they selected the game and the setup and I didn't see much that I found impressive. Were you impressed?
Its only 1 benchmark and there might be more tomorrow that will change this picture. At least I hope so.
Mintmaster
Veteran
Well, obviously I didn't look everywhere, but:
http://firingsquad.com/hardware/ati_radeon_hd_5870_performance_preview/page18.asp
http://techreport.com/articles.x/17618/12
In any case, you get my point.
So, 70% more performance over GTX285 is not enough? Hm, i hope you don't like Cypress.
Yes, i like the performance increase.
DavidGraham
Veteran
HD 4890 is faster than GTX 285 in HAWX because it employs DX10.1 render path which is faster than DX10 render path that GTX 285 uses .
However , GTX285 is faster than HD4890 in most games .
Blazkowicz
Legend
it's an arcade game with an extremely detailed ground, where you bomb stuff on the surface with missiles. not your average Secret Weapons of the Luftwaffe from 20 years ago
Cities with buildings actually drawed, that kind of stuff.
arguably you still have that empty top half but for the other part of the frame it must be grinding pretty hard.
Mintmaster
Veteran
Are you sure that PCPer is doing "Ranch Small"? The Fermi videos show a demo that's ~51 seconds long. PCPer's FPS traces are 145 seconds long.
Anyway, we see that the 5870 is anywhere between 22% (Techreport) to 46% (Hexus) faster than a 285 at 1920x1200. I'll just take fellix's word that he benched the right settings and that this Fermi card is 12% faster.
Nah, was talking about backface culling on the GPU ... without hierarchical backface culling you are always going to get huge spans of vertex shading producing bugger all for the pixel shaders. Even if setup wasn't a problem that isn't a nice thing to do.
That is true. Everything I am talking about is a leaked video of Far Cry 2 benchmark, which I found disappointing. Perhaps you find it inappropriate that I state my opinion of a leaked video of a benchmark of GF100 in a pre-release speculation thread?
You might believe that the video is fake, or that Nvdia didn't handpick the games themselves, but I think its real and that they did pick the games.
So, 70% more performance over GTX285 is not enough? Hm, i hope you don't like Cypress.
Its not enough. 70% more performance over the GTX285 is historically interesting, but not so interesting from a consumer point of view unless I am upgrading from a GTX285. Then its more interesting to see how a GTX360 (and even more a GTX380) performs towards other offerings within the same price class.
Not at all. Just pointing out the futility of drawing far reaching conclusions about whether or not Fermi is a disappointment based on that one video. Though it is strange that you would be so flabbergasted by a 60% improvement over GT200 on a stripped down Fermi part. I've seen folks commenting that if it doesn't topple the 5970 it's a big disappointment, guess the goal posts have moved
What are you upgrading from by the way?
PSU-failure
Newcomer
It's not that setup bound if you look at RV790/Juniper/Cypress.it's an arcade game with an extremely detailed ground, where you bomb stuff on the surface with missiles. not your average Secret Weapons of the Luftwaffe from 20 years ago
Cities with buildings actually drawed, that kind of stuff.
arguably you still have that empty top half but for the other part of the frame it must be grinding pretty hard.
Cypress Pro is just above RV790 while setup rate is 15% lower, Juniper XT is ~20% slower while having the same setup rate and the same ALU throughput. Nothing comparable to Heaven setup dependancy. Bandwitdh on the other hand seems to be a real concern with this game.
I'll try a 1920x1200 4xAA test with 5770 and 5870 at various clocks, it could give some good indications.
Sorry if this is a stupid question but what exactly is hierarchical backface culling? If a polygon is backfacing it's gonna get culled no matter what so how does the hierarchy even come into play?
One practical optimization would be to cull backfacing HOS primitives to avoid the unnecessary tessellation hit. But obviously you need higher setup anyway if the number of visible polys is going to increase substantially with tessellation.
leoneazzurro
Regular
You have to compare the same architecture with the same limitations.
Cypress Pro can be faster while having lower setup rate, but have higher ALU throughput.
Juniper XT has the same setup rate of RV790 but has much lower bandwidth and shaders must perform additional work (interpolation).
As there are multiple bottlenecks when running the game (setup, bandwidth, ALU, TMU).
Real comparison here is between Cypress XT and Juniper XT, the former being 2x the latter in all except the setup rate. In that game, 2x Juniper XT are 20% or more faster than the single Cypress. This hints about a bottleneck somewhere, it could some internal limitation but it is possible indeeed that the setup rate is somewhat involved in this.
Now, it would be interesting to see if Fermi will exhibit the same behaviour or not.
No, you were drawing your own conclutions out of my opinions. I'm sorry if my opinions doesn't fit your agenda, but seeing 1.1x to 1.2x increase of FPS over the 5870 in a game that traditionally favors Nvidia cards is not what I hoped to see once Nvidia presented the first benchmarks.Thats something I could expect even if both cards were on par in all other cases. I don't find it strange to be disappointed, but as mentioned I still have hope for something more impressive tomorrow when we see the rest (provided that TGdaily was correct about the date).
How it competes towards the GT285 isn't that interesting.
Btw, has it been confirmed that it was a stripped down Fermi part, since you mentioned it?
Nao linked this one a while back (slightly shallow hierarchy, but still a hierarchy).
NVM bad seed
Last edited by a moderator:
Huh? Your abject disappointment was explicitly laid out in several of your posts. There wasn't much room for misinterpretation
If you're looking for an agenda I think you'll have more luck elsewhere.Of course it is. But don't worry, there will be lots of comparisons to both GT200 and Cypress forthcoming.
Nope, all we have is that translated chinese forum post and its suggestion that the GTX 360 is 20% faster than Cypress in FC2. If the other card in that video is a GTX285 then the 55fps lines up with the numbers here - http://www.pcgameshardware.com/aid,...test-DirectX-11-graphics-card/Reviews/?page=4. That would put 84fps at 24% faster than HD5870. Of course, you can also use fellix's numbers if they make more sense to you (I don't know why since we don't know how a GTX 285 fares on his machine vs HD 5870).
How do you figure a 12 core variant has 32 TMUs? The math doesn't make sense.
DavidGraham
Veteran
Hey man , Do you know about the card shown in far cry 2 video ? is it GTX 360 or GTX 380 ?
Also were the rumors true ? GTX 380 would be castrated to 448 cores ?
Any idea about the frequencies and the Rasterizer ?
Similar threads
