The Fermi Thread - Part 3

[dogchainx]

This is the reason why I try not to get involved in these type of tech discussions, too many idiots making a huge fuss over different manufacturers' hardware.



This is almost as pathetic as the Picard vs Kirk debate.

[Sadasius]

Well the funny thing is that not only is it hot and uses way too much power but also the picture quality is still under the 5870. From Anadtech

Last but not least, there’s anisotropic filtering quality. With the Radeon 5870 we saw AMD implement true angle-independent AF and we’ve been wondering whether we would see this from NVIDIA. The answer is no: NVIDIA’s AF quality remains unchanged from the GTX200 series. In this case that’s not necessarily a bad thing; NVIDIA already had great AF even if it was angle-dependant. More to the point, we have yet to find a game where the difference between AMD and NVIDIA’s AF modes have been noticeable; so technically AMD’s AF modes are better, but it’s not enough that it makes a practical difference

[jaredpace]

they call it the apple equation.

rofl

[zerazax]

If you lie and people reward you for doing so by buying your products, then it must not be a real lie!! They call it the Apple Equation.

So true!

From TechPowerUp review:

NVIDIA first publicized its maximum board power as 295W, retracted it and posted it as "250W" probably fearing bad PR. We disagree with their 250W figure. Investigating maximum board power, we landed at the 320W mark, which is way off NVIDIA's claims.

Nvidia is just straight out lying about power figure?

Every review pinpoints the GTX 480 power load at around the 5970's... and the 5970 is rated at 294W. The 480 at 250W. Something obviously does NOT add up.

What's even more shocking is that the 470 uses almost 80-100W more than the 5850. It's one thing to say power doesn't matter when you're talking about the flagship, but now we're talking about the more budget-conscious 5850 vs. 470 bracket where the power consumption figures get noticed

[Chickenfeed]

Well the funny thing is that not only is it hot and uses way too much power but also the picture quality is still under the 5870. From Anadtech

Still to this date the AF on the evergreen cards causes a bigger performance hit than any other line has if I am not mistaken. Its not huge but bigger than what Id call ideal. Im an AF whore though so as a min I like 16af, I'll sooner omit the use of AA before AF but thats me. It is the best AF out (from a purely technical viewpoint ) granted but can I honestly say I notice the difference... not really.

Nvidia did improve their transparency AA however so that fact shouldn't be left out. Without a truely indepth article focusing on image quality alone (remember those nice G80 reviews that did this?), the verdict is still out on this I'd think.

As far as the whole power thing, it is quite odd. My understanding is due to the shader design (320shaders each with 4 simple shaders), it is very difficult to fully tax AMDs design (without very explicit programing from what I've heard) so in turn very unrealistic to reach their claimed TDP (reviews often have the 5870 well under its 188TDP ) Now Nvida on the other hand have a design which appears to actually reach its claimed TDP...

[***Deimos***]

This is a really expensive version of the George Foreman Grill.

looking back

Predictions that I made before...
- it would not be available in Nov.. probably not even Dec!! (they called me a heretic back then)
- Like PS3 CELL, was 99% sure wouldn't be 512SP. Based on super low yields, guessed 480-448SP for GTX480, and 416SP for GTX470. :*(
- compared power/die to R600 and Phenom. Regardless if broken or high power it WOULD launch.
- 750Mhz is way too high. Guessed 600-650Mhz max based on low GT240 clocks.
- Laughed at folks suggesting $399 or even $299. I guessed $600 - but, although MSRP is lower, surely with high demand there will be ridiculous markups.
- Like everybody else, it was assumed 100% true that power >200W (ie Charlie's 280W rumour). Yet, I thought its only for special case like folding.
- no 6 display EyeInfinity.. well that was pretty much a given.

nVidia surprised me:
- 700Mhz, and yet can OC a good 10% more!
- Surely nobody was naive enough to believe +60% on everything. But amazing performance in some games (FC2, BFDX11)tesselation, folding@home and few demos. Fearing only +10%, performance better than expected (depending on which review you read)
- *only* 250W.. better than 280W, but something around 220W would be better still.
- it actually works in all the games. Some 2560 and TRAA driver issues to fix, but overall pretty well polished. Nothing broken (ie Phenom,R600)

[Chumbucket843]

Well the funny thing is that not only is it hot and uses way too much power but also the picture quality is still under the 5870. From Anadtech

with all due respect angle dependent AF isnt very useful if it is undersampled. its taking the correct samples it should but not enough, nvidia is the opposite. the net quality is about as good as nvidia's quality setting.
http://www.pcgameshardware.com/aid,6...eviews/?page=5

[zerazax]

- *only* 250W.. better than 280W, but something around 220W would be better still.

And yet, it pulls more power than the 5970, rated at 294W..., and the reviews consistently show it drawing 100W more than the 5870 which is rated at only 62W lower. Something doesn't add up with 250W

a 480 vs 5870 Collaboration of results, pls let me know of any errors. adding 5850 vs 470 atm as more reviews come in with it (470)

Thanks for the work. So at 1920 x 1200, roughly 8% faster without AA, 15% with AA.

Any 2560x1600 numbers?

[SimBy]

The 250W is obviously a lie. Reviews prove it's 300W or even more.

[***Deimos***]

Time to start another 100 page thread.

GF104 is reported to consist of 256 SP - half of GF100. 
same number of TMUs - 64; and only 33% less ROP - 32. 
Memory bus is down to 256-bit - the same as ATI's Cypress.

Are you saying nvidia fudged the Watts? *SHOCK!!*
Most reviews I think used Furmark. Maybe need to compare to 5970 across 5-10 games, before making conclusion.

10% OC headroom means slightly lower power version possible.
Realistically, 4-6 months for new PCB and new stepping/revision - or until next year for 28nm.
Till then, nVidia's gonna downplay the power and do the best trying to sell them.

[grimREEFER]

Nvidia did improve their transparency AA however so that fact shouldn't be left out.

well, i believe transparency AA is only needed in dx9 games actually.
so i dont know if its relevant anymore.

[zerazax]

Are you saying nvidia fudged the Watts? *SHOCK!!*
Most reviews I think used Furmark. Maybe need to compare to 5970 across 5-10 games, before making conclusion.

http://www.tomshardware.com/reviews/...0,2585-15.html

They ran the 480 & 470 across Unigine....

[Chickenfeed]

http://www.tomshardware.com/reviews/...0,2585-15.html

They ran the 480 & 470 across Unigine....

Oh wow I didn't notice that detail when I went over their review. Thats quite the statement right there... Perhaps we will see a 480 ship with a LN2 pot and personal backyard wind turbine

[zerazax]

Oh wow I didn't notice that detail when I went over their review. Thats quite the statement right there... Perhaps we will see a 480 ship with a LN2 pot and personal backyard wind turbine

I was shocked to see that it actually drew more power over the benchmark than the 5970 did!

[annihilat0r]

Is a single GTX480 able to game on three monitors ?

If not, this can be the source of the TDP discrepancy. Maybe Eyefinity'ing at 3 monitors draws significantly more power than a single monitor setup. And since TDP is supposed to be the maximum, AMD was forced to measure it at 3 monitors whereas Nvidia measured it a 1 monitor because a single 480 isn't able to game on three monitors.?

[Chickenfeed]

The video out logic is quite low power and the 4x0 cards still can output onto 2 displays and still have 3 outputs ( although they aren't always in use ). Then there is still the fact that the 480 is still pushing 250+watts with 1 display so regardless that doesn't change those numbers but I know where you are coming from with that. It remains that 188 is worst case with the 5870 where Nvidias 250 figure is suspect ( we've yet to see 3D vision surround numbers on a newer intensive game such as Metro 2033 so it might get even worse who knows ) Still this is the first time we actually will need a 1000watt psu for a 2 card system. In the past its only been needed with tri / quad configs or in high end bench settings ( ie highly clock LN2 benching / volt modding )

[SimBy]

Is a single GTX480 able to game on three monitors ?

If not, this can be the source of the TDP discrepancy. Maybe Eyefinity'ing at 3 monitors draws significantly more power than a single monitor setup. And since TDP is supposed to be the maximum, AMD was forced to measure it at 3 monitors whereas Nvidia measured it a 1 monitor because a single 480 isn't able to game on three monitors.?

1. no
2. eyefinity setup has same powerdraw at load as single monitor, only difference is idle where it consumes more due to higher clocks

[annihilat0r]

The video out logic is quite low power and the 4x0 cards still can output onto 2 displays and still have 3 outputs ( although they aren't always in use ). Then there is still the fact that the 480 is still pushing 250+watts with 1 display so regardless that doesn't change those numbers but I know where you are coming from with that.

Where I was coming from was that the rated TDP of the E6 edition was actually in the 220W TDP range AFAIK. A gig of RAM can't make that much difference.

[sxs112]

EVGA 480/470 HC FTW

[Chickenfeed]

Where I was coming from was that the rated TDP of the E6 edition was actually in the 220W TDP range AFAIK. A gig of RAM can't make that much difference.

The additional video out logic ( adding another 3 outputs is more than a basic change ) does boost the TDP obviously and the vram does have its effect. My understanding is the PCB on the eye6 card is different.

I wonder how long it will take for EK to have their blocks out in the retail channels... I'm sure they will sell very well as these cards are going to end up widely watercooled no doubt.

[YukonTrooper]

*snip*

They say right in that quote that they have yet to come across a game where the AF handling is noticeably different between Nvidia and ATI GPU's.

[E30M3]

Anand

"For whatever reason AMD can’t seem to keep up with NVIDIA when it comes to the minimum framerate, even at lower resolutions. Certainly it’s obvious when the 1GB cards are video RAM limited at 2560, but if we didn’t have this data we would have never guessed the minimum framerates were this different at lower resolutions.

Finally we have a quick look at SLI/CF performance. CF seems to exacerbate the video RAM limitations of the 5000 series, resulting in the GTX 480SLI coming in even farther ahead of the 5870CF. Even at lower resolutions SLI seems to be scaling better than CF."

[Boissez]

BTW, Charlie "i'm an idiot" dedknefuhrefuohreofmerijian was wrong in most of the hate he posted. The product has one flaw and that is power draw but all in all it's high clocked and a real great performer. Eat that C.

Yes... one flaw only no reason to consider these:

- noise (loudest cards ever according to anandtech)
- Time to market (6 months after 5870)
- Availability (this is a paper launch)
- Manufacturability (castrated 512 core)
- Manufacturing cost (larger chip, more ram, beefier cooling, pwms, etc...)

... those are all features right?

[annihilat0r]

Yes... one flaw only no reason to consider these:

- noise (loudest cards ever according to anandtech)
- Time to market (6 months after 5870)
- Availability (this is a paper launch)
- Manufacturability (castrated 512 core)
- Manufacturing cost (larger chip, more ram, beefier cooling, pwms, etc...)

... those are all features right?

Let Nvidia think about manufacturing cost and "manufacturability". What, aren't you going to buy it for $500 because it is costly to manufacture? Consumers don't care about manufacturing cost.

How do you know this is a paper launch? From everything I have heard, the availability of Fermi might be more than 5800s. Don't forget that while there were some 5800s you could buy at launch date, even after 2 to 3 months you still couldn't buy any. 5870 was only a little better than a paper launch.

[cegras]

Let Nvidia think about manufacturing cost and "manufacturability". What, aren't you going to buy it for $500 because it is costly to manufacture? Consumers don't care about manufacturing cost.

How do you know this is a paper launch? From everything I have heard, the availability of Fermi might be more than 5800s. Don't forget that while there were some 5800s you could buy at launch date, even after 2 to 3 months you still couldn't buy any. 5870 was only a little better than a paper launch.

http://ncix.com/search/?categoryid=0&q=gtx480

I see preorders.