The Fermi Thread - Part 3

[pentium777]

http://www.hardwareluxx.de/community...9-post364.html

8.7% over 5870? I don't see anyone with a 5870 buying a 480...

Seems like this makes sense why Nvidia would be so tight regarding releases of benchmarks other than the heavy tesselation stuff.

Just checked my old scores 2xGTX 280 SLI would get me 9200-9600 in 3dmark vantage extreme. So this card will not be as fast as 2xGTX 285s but seems to be as fast as 2xGTX 280s. If this was 6 months ago it would be something special but 5870 has been around and 5970 this isn't going to excite too many people.

[WeiT.235]

8.7% over 5870? I don't see anyone with a 5870 buying a 480...

Seems like this makes sense why Nvidia would be so tight regarding releases of benchmarks other than the heavy tesselation stuff.

Synthetic bench don't translate well the game performance, and this results are probably based on the 10.2 beta driver, not sure about the NV driver though.

[RPGWiZaRD]

Especially 3DMark. HD 5970 is also only 27.4% before GTX 480 here... yea right, like it would be this close to HD5970 in gaming benchmarks (where 5970 has working CF profiles) had been nice.

At least there's hoping for at least 20% performance advantage (480 vs 5870) in games looking at these scores though.

[pentium777]

Synthetic bench don't translate well the game performance, and this results are probably based on the 10.2 beta driver, not sure about the NV driver though.

True but it is a decent indicator of what we will likely see with in game benchmarks, each game will differ obviously.

Well sigh.... I never got the GTX 285s or 295, or HD5870 because I specifically waited for this release on my "aging" GTX 280 SLI setup. Now one of them is gone and I'm turning down graphics and having to run DX9 in BC2.

My hope I guess is that GTX 480 SLI will have 80-90% performance increase at 2560x1600 in DX11 for BFBC2 and hopefully driver improvements and SLI scaling doesn't make me regret the purchases.

[LiquidReactor]

No pictures or anything but it will be closer to G200 than G200b.

Is this based of some credible leaks/sources?

[Musho]

Synthetic bench don't translate well the game performance, and this results are probably based on the 10.2 beta driver, not sure about the NV driver though.

For nVidia's sake, I hope this isn't true. I hope that they are already using 10.3 drivers, because if they are not, the 5870 will be _very_ close to the GTX480. 10.3 gave some nice boosts in both benchmarks such as vantage, as real world games.

[WeiT.235]

For nVidia's sake, I hope this isn't true. I hope that they are already using 10.3 drivers, because if they are not, the 5870 will be _very_ close to the GTX480. 10.3 gave some nice boosts in both benchmarks such as vantage, as real world games.

Whatever, they are already 6 months later

[Coldon]

For nVidia's sake, I hope this isn't true. I hope that they are already using 10.3 drivers, because if they are not, the 5870 will be _very_ close to the GTX480. 10.3 gave some nice boosts in both benchmarks such as vantage, as real world games.

yeh, but later releases of the nvidia drivers will also offer nice performance boosts. If you really want to be fair you should compare the gtx480 with the 5870 using the launch ATI drivers, but that will never happen so just remember if the difference is 8~10% now with 10.3 drivers and NV beta drivers, given 4 months the gap will probably be more like 15%~20% with mature nv drivers.

the 10.3 drivers offer a substantial performance increase over the launch 9.12s, and the same can be expected over in NV land.

[mindfury]

Is this based of some credible leaks/sources?

[Musho]

yeh, but later releases of the nvidia drivers will also offer nice performance boosts. If you really want to be fair you should compare the gtx480 with the 5870 using the launch ATI drivers, but that will never happen so just remember if the difference is 8~10% now with 10.3 drivers and NV beta drivers, given 4 months the gap will probably be more like 15%~20% with mature nv drivers.

the 10.3 drivers offer a substantial performance increase over the launch 9.12s, and the same can be expected over in NV land.

I see where you are coming from, but that's not entirely true in my opinion. I'd rather to see what performance I can get right now, than what I _might_ get later on. Also, due to all the delays, the driver team should have had quite some time optimizing the drivers, so we can't really be sure how much more performance they can squeeze out of these cards compared to the drivers they will launch with. I'd love it to happen, though. Competition is good for the market!

[LiquidReactor]

With a chip that big nVidia just has to match 5970 in order to compete with ATI in the performance/cost war.

Cypress is what, 334mm^2?

[ajaidev]

With a chip that big nVidia just has to match 5970 in order to compete with ATI in the performance/cost war.

Cypress is what, 334mm^2?

I will not go against the 5970, the x2 is suppose to come out in a few months tough.

[saaya]

A huge cost is R&D. You worry how much the wafers and heatsinks cost, but the real point to consider is R&D. The Fermi architecture, just like the G80 architecture, was a long time in development and expensive to produce.

But consider a few points:

• The architecture is being made into three seperate high-end products, the cheapest of which is the GeForce part. Quadro parts are often priced twice (or more) the price of GeForce parts, and Tesla parts even more.
• The architecture is extremely modular. Scaling up for the next generation after GTX 480 is almost as simple as CTRL+V. Scaling down is just as easy.
• Longevity. Fermi has it. It's an extremely advanced architecture with a ton of features built around the best part of DX11, tessellation.

jeez... why dont you post the nvidia pr slides here right away? you seem to be repeating what nvidia pr tells everybody... :P

and scaling down is easy? gf100 taped out when? and gf104 is coming when? if its that easy wouldnt you expect gf104 to come out just months if not weeks after gf100? look at ati, they released rv870 and rv840 at almost the same time, in fact the rv840 team managed to slightly pull ahead of the rv870 team in the rnd process! THATS what id call easy scaling...

I haven't seen anything that's made me believe that NVIDIA will be selling GPUs at a loss.

fair enough...

Consider that even with HD5800 out performing NVIDIA parts, NVIDIA is gaining more of the market, selling a ton of GPUs and making a lot of money.

uhmmm and you base this on what exactly?

Even IF they were selling GF100 at a loss, they aren't in AMD's financial shoes.

exactly, amd has financial backing from abu dhabi and they have all the ip they need for the next decade, x86, gpu, chipsets...

hey guys have you seen this?
http://www.tweaktown.com/articles/31...is/index3.html

10.3 gives ati cards almost a 10% boost in unigine and hawx... impressive...
and a 20% boost on minfps for the 5870 in fc2 as well...
if you ask me, this is atis fermi welcome party hehe
im sure they had those tweaks before, but didnt integrate them into the past few driver releases to have something for the fermi launch...

and the timing is perfect, they released them just in time for fermi reviews...

[ElSel10]

exactly, amd has financial backing from abu dhabi and they have all the ip they need for the next decade, x86, gpu, chipsets...

I hope you aren't implying that AMD is better off financially than Nvidia

[WeiT.235]

Fermi yields still under 50 percent

According to Digitimes, Nvidia's Fermi is still plagued by poor yields, but the company is still expected to have enough cards at launch.TSMC's yields are apparently still under 50 percent and it is unclear whether they can or will improve anytime soon.

However, as we said earlier, Nvidia opted to reduce the number of cores from 512 to 480 on the flagship GTX 480, while the GTX 470 will have even fewer cores, 448 to be exact.

Should TSMC manage to improve the yields, Nvidia could probably launch a faster version with more shaders, putting more pressure on ATI. However, ATI has shipped a few million DirectX 11 cards since October, so this is a small consolation for Nvidia.

Can't believe it after the TSMC's announcement of 40 nm production 1+ year ago...

[ajaidev]

....
10.3 gives ati cards almost a 10% boost in unigine and hawx... impressive...
and a 20% boost on minfps for the 5870 in fc2 as well...
if you ask me, this is atis fermi welcome party hehe
im sure they had those tweaks before, but didnt integrate them into the past few driver releases to have something for the fermi launch...

and the timing is perfect, they released them just in time for fermi reviews...

Yep 10.3 look specially made to combo up with 5870 2GB and give a sort of a painful box to the Nvidia's part. If 10.3 is official by the time fermi gets out and reviewers use that as reference 5870 driver, GTX 4x0 will not seem that sweet.

In anycase when ATi gets the new 5870 2GB's out reviews are bound to use 10.3 "If final" and also GTX 4x0

I hope you aren't implying that AMD is better off financially than Nvidia

The fact that abu dhabi is interested in AMD and in a situation where AMD fails he will either buy majority "May not be allowed by Gov." or alternatively he will enter a some ratio ownership.

EDIT: BTW this is someone who bailed out Dhubai for $10bn

[azza21]

yeh, but later releases of the nvidia drivers will also offer nice performance boosts. If you really want to be fair you should compare the gtx480 with the 5870 using the launch ATI drivers, but that will never happen so just remember if the difference is 8~10% now with 10.3 drivers and NV beta drivers, given 4 months the gap will probably be more like 15%~20% with mature nv drivers.

the 10.3 drivers offer a substantial performance increase over the launch 9.12s, and the same can be expected over in NV land.

Whining about no dual monitor option, whining about a so called flicker problem with 5800 series, whining about the fact its unfair to compare a newly launch fermi with the latest ATI driver (oh you think nVida would do a fair bench)...get a grip man.
You complain about all the ATI fanboys and ranting/bashing on in these threads. Your just the same, clearly an nVidia fanboy, trying to find any stupid of reason to bash an ati product or use an excuse not to buy and ati product. You don't like ATI and your an nVidia fanboy thats why you won't buy a 5800 series card, don't use lame excuses like there is a flicker problem.

Its absolutely shocking the way some of you go on, so delusional.
Its a bloody graphics card company, its hardly Manchester United vs Liverpool rivalry is it, i cannot understand why you lot get so worked up.

[zalbard]

http://www.hardwareluxx.de/community...9-post364.html

Interesting.
I assume ATI driver is 10.2.
So damn, they are very close.
I suppose Nvidia also has a lot of driver work to do, though. So we'll see 10% performance difference between 5870 and GTX480.
Vantage isn't a very realistic benchmark, though...

[sxs112]

CPU i7 965
MB RAMPAGE II EXTREME
RAM 3G DDR3

http://www.mobile01.com/topicdetail....&t=1475786&p=1

[Picao84]

Its absolutely shocking the way some of you go on, so delusional.
Its a bloody graphics card company, its hardly Manchester United vs Liverpool rivalry is it, i cannot understand why you lot get so worked up.

LOL! Im not a buyer from either. I just like to watch like a football match really, I admit

[ajaidev]

CPU i7 965
MB RAMPAGE II EXTREME
RAM 3G DDR3

http://www.mobile01.com/topicdetail....&t=1475786&p=1

That reminds me too much of GTX 295 anyways here is 5870's score with 10.3 "Idk about the settings i guess they are the same i could be wrong"

[Coldon]

Whining about no dual monitor option, whining about a so called flicker problem with 5800 series, whining about the fact its unfair to compare a newly launch fermi with the latest ATI driver (oh you think nVida would do a fair bench)...get a grip man.
You complain about all the ATI fanboys and ranting/bashing on in these threads. Your just the same, clearly an nVidia fanboy, trying to find any stupid of reason to bash an ati product or use an excuse not to buy and ati product. You don't like ATI and your an nVidia fanboy thats why you won't buy a 5800 series card, don't use lame excuses like there is a flicker problem.

Its absolutely shocking the way some of you go on, so delusional.
Its a bloody graphics card company, its hardly Manchester United vs Liverpool rivalry is it, i cannot understand why you lot get so worked up.

i'm not whining just pointing out a flaw in his logic, saying that he hopes they used the 10.3 drivers as that will show that fermi isn't that much faster (which may very well be), i was just pointing out that nvidia's driver also have to mature or do you disagree? So basically drivers don't affect performance in your opinion and driver versions don't really matter.

I don't have a preference to either camp and will buy whatever is the best value for money and will give me the least problems. Prices in south africa are insane (around $650 for a 5870) and I was prepared to buy a 5870 at that price till i saw reports of this *imaginary" flicker problem. I obviously must have been delusional that day and imagined those hundreds of posts confirming it and obviously in my state of delusional missed the driver change log that confirmed that it was fixed (i've been checking each new driver change log for that to happen cause then i can pretty much go out and buy a 5870 - but i sure as hell am not paying $650 for a card that flickers and then try explain why i want a refund to the idiot techguys at the suppliers here), but that's not the issue here is it. My initial comment was just a joke (you know something that's meant to be funny).

ironically enough my posts haven't been aggressive or overbearing, i wasnt even whining just pointing out a valid fact, which obviously got your panties in a bunch, so maybe instead of getting all upset give me a valid counterpoint to why the driver version maturing wont improve performance over time.

just to clear things up in case you missed my point like you did in my last point: at launch they SHOULD use the latest driver versions from each camp for the reviews and compare performance on those but it is naive to not think that performance will improve as the driver matures, a followup review in a few months with a more mature driver from both sides might yield different results. So the argument that a driver is going to bridge the gap between the two products is *at this point* incorrect as ATI have had nearly 6 months time to improve their driver.

[RaZz!]

yeh, but later releases of the nvidia drivers will also offer nice performance boosts. If you really want to be fair you should compare the gtx480 with the 5870 using the launch ATI drivers, but that will never happen so just remember if the difference is 8~10% now with 10.3 drivers and NV beta drivers, given 4 months the gap will probably be more like 15%~20% with mature nv drivers.

the 10.3 drivers offer a substantial performance increase over the launch 9.12s, and the same can be expected over in NV land.

new gpus are always benched with release drivers against older cards that have matured drivers. that's not unfair at all. comparisons like that can only be done at a point in time. what you can expect from these comparisons are results that are only valid for this specific point in time as well. nobody can predict how the performance will evolve in the future. sometimes matured drivers increase performance by a significant amount, but then there are times where matured drivers don't deliver better performance at all.

but that doesn't really matter, as you buy the graphics card a this point in time as well, so the comparison is valid. when you want to buy such a card, you care about its current performance most, and not what the performance will look like in 5 months (where the performance of the competing product might be better as well...).

so, yeah, it's pretty hard to say

[Coldon]

new gpus are always benched with release drivers against older cards that have matured drivers. that's not unfair at all. comparisons like that can only be done at a point in time. what you can expect from these comparisons are results that are only valid for this specific point in time as well. nobody can predict how the performance will evolve in the future. sometimes matured drivers increase performance by a significant amount, but then there are times where matured drivers don't deliver better performance at all.

but that doesn't really matter, as you buy the graphics card a this point in time as well, so the comparison is valid. when you want to buy such a card, you care about its current performance most, and not what the performance will look like in 5 months (where the performance of the competing product might be better as well...).

so, yeah, it's pretty hard to say

+1 I fully agree as I said in my followup post, cant help it if people misunderstood my point, obviously you will test with the latest drivers but you cant expect that test to be valid 2/3 months down the line, point in case the 10.3 catalysts. The gap between the 285 and the 5870 has grow a lot larger since the 5870 launch just due to the driver updates.

[Sentential]

new gpus are always benched with release drivers against older cards that have matured drivers. that's not unfair at all.

Not only that but they have had this card for 3 respins, so to say that they dont have a "mature" driver is somewhat disingenious, their "dedicated" driver team has had ample time to write something acceptable for the card given the sheer amount of delays.

Really the drivers dont concern me; my issue with Fermi is as follows:

#1 Its the first time that I can think of that a "harvested" part is the best they can do, NV30 was not like this, R600 was not like this, R520 was not like this.

#2 Power efficiency 250W vs 185W-ish for the ATi equivalent.

#3 Lack of taped out parts, sure G108 just supposedly did but what about 104 and 106? Is this going to be a repeat of the GT200?