My benches so far. Adding GTX480 SLI benches.

[RussianSensation]

HD5970 is also $200 more expensive

I found a Diamond 5970 for $602: http://www.ncixus.com/products/index.php?sku=46685

Check this out:
http://www.legionhardware.com/articles_pages/nvidia_geforce_gtx_480_fermi_arrives,4.html

BF:BC2 1920x1200 4AA/16AF
GTX480 = 50 frames
5970 = 88 frames (+76%)

2560x1600 the difference is a whopping +97%

Metro 2033 1920x1200 AAA/4AF
GTX480 = 35 frames
5970 = 48 frames (+37%)

STALKER: Call of Pripyat 2560x1200
GTX480 = 59
5970 = 78 (+32%)

So you pay 20% more for 30-70% performance increase in most demanding games (other than Crysis which is a PITA on any card)!
...and GTX480 is NOT available for sale yet....will consume more power, be louder and slower by a lot more than 15% champ!

 

[apoppin]

Thats about enough. FERMI is hot Loud and a power pig. Thats just the facts. As for disrupting BS. Were discussing Fermi. This is what Fermi Is. No its not fun yet.

Fact: That is your 2nd-hand *opinion* of Fermi which you got from reading what you wanted to see.

So you pay 20% more for 30-70% performance increase in most demanding games (other than Crysis which is a PITA on any card)!

If you are looking to spend $600 on a HD 5970, at least consider HD 5850 CF.
- i think you get the same performance for about $500.

It's possible to get the fan going that fast with a compressed nitrogen or argon cylinder and proper gun. Flying apart is definitely a concern. I remember reading about someone spinning up a fan in this method and it flying apart with blades stuck in the wall!

i believe i read the same thing.

 

[RussianSensation]

If you are looking to spend $600 on a HD 5970, at least consider HD 5850 CF.
- i think you get the same performance for about $500.

Ya that's a good point apoppin! 5850s in CF are an amazing high-end "value" right now, even more so than a 5970.

 

[Rubycon]

I'll be curious to see what the GTX 495 looks like.

 

[Unkle_Tar]

Is this thread about HD5970?

We are talking about the GTX480 and how it performs in todays games. It was benched against a HD5870 as a reference, because keys doesnt have a HD5970 although I dont know which is more appropriate to compare against. HD5970 is also $200 more expensive according to newegg where as the HD5870 is $80~0 cheaper.

If you look at the performance, overall its about 10~15% faster, but in games where Fermi seems to like, those leads become 20~50%.

That's the point, it will look less flattering against a 5970 so it won't be tested against one. If the 480 was slower then a 5870 this thread wouldn't exist. The price thus far dosn't matter, and will continue to be a moot point until you can actually pick one up for $499.

 

[apoppin]

Ya that's a good point apoppin! 5850s in CF are an amazing high-end "value" right now, even more so than a 5970.

Just like 4850, 5850 carries on the performance/watt/price tradition.
-They are easy to find; they are often on sale near $500 a pair and i understand they overclock well .. and they run pretty cool also ... unlike 5970 which only OCs well but is almost never on sale and is hard to find.

Frankly, i would love to test a pair of 5850s and GTX 470 (or two) also

 

[Zebo]

Werd. Once they get pwr under control it's the card to have. Everyone knows this.

Hey Mark I thought you quit using Smilies. Have to admit they've slowed a bit.

Ditto on 5850 best mainstream card $200-$300 and best higher end too w/ xfire.

 

[Unkle_Tar]

It's possible to get the fan going that fast with a compressed nitrogen or argon cylinder and proper gun. Flying apart is definitely a concern. I remember reading about someone spinning up a fan in this method and it flying apart with blades stuck in the wall!


Impossible without breaking trademark law. Cool n' quiet is an AMD thing.

I don't know if running it too far out of spec is a good idea, might end up with little pieces of plastic embedded in your face when it explodes.

 

[Rubycon]

5970s overclocking well is a farce. Sure the chips can do it but power delivery is the big culprit and cannot be easily rectified. (no pun intended!)

 

[apoppin]

5970s overclocking well is a farce. Sure the chips can do it but power delivery is the big culprit and cannot be easily rectified. (no pun intended!)

Well you can overdraw the PCIe spec; we did it with GTX 480 you just need a really stable PSU. Nvidia's spec for GTX 480 is 250 W TDP in ***gaming***
... not the maximum power draw of the card

Didn't AMD engineer the 5970 with overclocking in mind; i thought they planned for it?

Eventually, a video card with a power brick and cord plugged into the wall would solve the PCIe power delivery issue and the cards can get as power-hungry as they want and we can go back to 250 W PSUs for the rest of the PC.

Yes, Zebo .. less emoticons ..
.. here at this forum

 

[Rubycon]

The problem is with VRMs on the card rather than insufficient PSU not supplying sufficient power.

Well you can overdraw the PCIe spec; we did it with GTX 480 you just need a really stable PSU. Nvidia's spec for GTX 480 is 250 W TDP in ***gaming***
... not the maximum power draw of the card

Didn't AMD engineer the 5970 with overclocking in mind; i thought they planned for it?

Eventually, a video card with a power brick and cord plugged into the wall would solve the PCIe power delivery issue and the cards can get as power-hungry as they want and we can go back to 250 W PSUs for the rest of the PC.

Yes, Zebo .. less emoticons ..
.. here at this forum

 

[MagickMan]

Yep, the VRMs are the culprits. My 5850 core runs @80C when OCed to 1090MHz and 1.35v, but the VRM temps go nuts during heavy testing, as high as 100C in Furmark.

Not to mention, at those speeds @100% load it's loud, really loud. It just isn't practical.

 

[ArchAngel777]

Yeah I know these things factor in, I posted it mainly because cookie said he was getting sick of performance threads turning into arguements about power and noise. If thats the case and we throw out all the drawbacks, the 5970 would be the card to buy because its faster then the 480.

Ahhh, I understand now. I agree with you completely.

 

[blanketyblank]

Fact: That is your 2nd-hand *opinion* of Fermi which you got from reading what you wanted to see.


If you are looking to spend $600 on a HD 5970, at least consider HD 5850 CF.
- i think you get the same performance for about $500.


i believe i read the same thing.

Where are you getting 5850s for 250? Cheapest I've seen them so far is $280 of course not counting bing cashback which could also apply to a 5970. That would mean 2x 5850 would cost around 560, but probably more like $600 since $300 is really the low normal price.
That being said I'd have to say 5970 for $600 would be a better deal since it's a single card, and I believe slightly faster than 2 x 5850s.

 

[apoppin]

Where are you getting 5850s for 250? Cheapest I've seen them so far is $280 of course not counting bing cashback which could also apply to a 5970. That would mean 2x 5850 would cost around 560, but probably more like $600 since $300 is really the low normal price.
That being said I'd have to say 5970 for $600 would be a better deal since it's a single card, and I believe slightly faster than 2 x 5850s.

Well, i was looking at the best sale i had seen. if you take $560-$600, then you take 5970 for near $700. 5970 costs a bit more was my point and it is hard as heck to find.

i don't know if 5970 is faster; it doesn't OC as well, i believe

 

[anthonee1230]

Werd. Once they get pwr under control it's the card to have. Everyone knows this.

You talking about Fermi II or what? How do you think they are going to get power under control with the GTX 480/470? With a simple driver update?

Not flaming, just wondering how you think that's going to be accomplished with the GTX 480.

 

[evolucion8]

Seconded. ATI seemed to score better in Futuremark benches, but in the 2900XT's case, didn't really translate into real world performance. This is pretty much the very reason why I didn't bother with Vantage. Doesn't really tell any significant story.

Totally true, when nVidia launched the GTX 260+ series, it scored higher than the HD 4870 1GB, and yet, they traded blows in games with the slight edge going to ATi in newer, shader intensive games while the GTX 260+ offered better performance in current (At the time of launch) and older games.

 

[evolucion8]

IIRC, in Dirt 2 all you get is tessellation of the water puddles on the track so they behave more realistically and sprinkled character details in the crowd. For these small visual differences 5xxx series gets massively penalized (see 5770 and 5850) compared to GTX275 and 4890 in a very similar DX10 mode: http://www.xbitlabs.com/articles/video/display/radeon-hd5830_11.html#sect1
At this point hardly anyone will accept that kind of a performance for only better water effects. Speaking of water effects, this isn't like comparing the difference in water Half-Life 2 with DX8 vs. DX9 paths: http://www.anandtech.com/show/1549/2 This better not be the direction of DX11 games.

Didn't know that Dirt 2 could be run in DX10 mode. . .

 

[Keysplayr]

Ok, some o/c results. I went to 830 core, 1660 shaders, and 2211 Memory.

Ran Stalker Bench again. Same settings as my original bench.
Numbers rounded to nearest 10th. 55.4 equals 55. 55.5 equals 56.

Map: Min/Avg/Max @ Stock. 701/1402/3696
Day: 25/48/64
Night: 33/50/70
Rain: 41/52/70
Sun: 26/34/46

Map: Min/Avg/Max @ 830/1660/4422
Day: 27/56/73 (+8%| +18% | +15%)
Night: 37/59/82 (+12% | +18% | +18%)
Rain: 44/62/82 (+8% | +19% | +18%)
Sun: 31/41/54 (+20% | +20% | +18%)

Ran fan at 100% to play safe, but not needed. Monitored temps shows a high of 66C at it's peak. I probably could have been fine setting the fan to 65 to 70%.
I went right to 830/1660/2211(4422). Didn't try anything lower, or higher. Just a quickie.
The o/c on core and shaders and memory is about 18.5%.

 

[Dark4ng3l]

Safe to assume you left the voltage alone? I haven't read anywhere what the stock voltage on these things is so I don't know if there is much room for upping it. I know evergreen can take up to 1.35v safely on air so I would assume that these being on the same process would have similar limits?

 

[Keysplayr]

Safe to assume you left the voltage alone? I haven't read anywhere what the stock voltage on these things is so I don't know if there is much room for upping it. I know evergreen can take up to 1.35v safely on air so I would assume that these being on the same process would have similar limits?

Did not touch the voltage. So what do you think? near 20% o/c on stock voltage ok?
Like I said, I rarely o/c anything for normal usage, so I'm not that great at anything except moving sliders. Voltage tweaks, I'll leave to the pros.

 

[Dark4ng3l]

Ok? thats really good! Though considering the power numbers I have seen up to now you probably won't want to up the voltage too much because you don't want to outdraw the limits of what you can feed the card and cause your system to crash... But since a huge part of the problem is chip power leakage if you can keep the temps down (like with watercooling) then you could get monster overclocks on this thing.

 

[Keysplayr]

Ok? thats really good! Though considering the power numbers I have seen up to now you probably won't want to up the voltage too much because you don't want to outdraw the limits of what you can feed the card and cause your system to crash... But since a huge part of the problem is chip power leakage if you can keep the temps down (like with watercooling) then you could get monster overclocks on this thing.

Well, the temps I observed running the Stalker bench didn't get past 66C but the fan was at 100% (very noisy). If this card can run comfortably at 90C+, I could have run the fan considerably slower. I ran it at 100% only because I didn't know what to expect for temps. But was surprised it was that low. HSF design seems really good. probably better than most aftermarket coolers. Once voltage is messed with though, temps would go up rather radically I would imagine. But to be honest, unless you're a die hard overclocker, like the type of user who would overclock their toaster if they could, I don't really see any reasons to overclock the 480. At least not in the DX11 benches I did. Maybe guys running F@H to squeeze that last bit of PPD out of the GPU.
When I have more time, I'll try running at the o/c speeds again but lower the fan incrementally until I get to a reasonable operating temp and fan speed and report back on those findings.

 

[3DVagabond]

I'm surprised that with all the noise about Fermi's power draw nobody's tried undervolting it? Just a thought.

 

[Dark4ng3l]

Well, the temps I observed running the Stalker bench didn't get past 66C but the fan was at 100% (very noisy). If this card can run comfortably at 90C+, I could have run the fan considerably slower. I ran it at 100% only because I didn't know what to expect for temps. But was surprised it was that low. HSF design seems really good. probably better than most aftermarket coolers. Once voltage is messed with though, temps would go up rather radically I would imagine. But to be honest, unless you're a die hard overclocker, like the type of user who would overclock their toaster if they could, I don't really see any reasons to overclock the 480. At least not in the DX11 benches I did. Maybe guys running F@H to squeeze that last bit of PPD out of the GPU.
When I have more time, I'll try running at the o/c speeds again but lower the fan incrementally until I get to a reasonable operating temp and fan speed and report back on those findings.

I wonder if you can find out what the default voltage is. There was speculation a couple of weeks ago that nvidia upped the voltage at the last minute to boost clock speeds a bit to make sure that they would beat the 5870 during reviews. 20% O/C on stock voltage is truly massive. Maybe they upped the voltage .15v just to play it safe to give it 5% more umph when they could have left it lower and thats why you have all this headroom?

Anyways you shouldn't base your temps on a single game. I think it was mentioned in the anand article that Crysis ran the card even hotter than furmark so if your sample only hits in the 60s with max fan speed that particular game might not be using the whole chip at close to 100%(considering the flexible architecture the driver might not be able to utilize all the hardware at the same time in certain situations... could be a lot of room for improvement in drivers though).