GF100 evaluation thread

[Ethatron]

1. If ATI disregarded power and heat, we would have the same performance and a smaller chip to boot.

I once measured Amps for OCs, giving an idea about the clock/Wattage curve (of course Volts are always to make it stable):

900 1.100V 39.5A
900 1.125V 41.8A
950 1.149V 44.1A
960 1.187V 48.8A/49.9A
970 1.212V 51.1A/52.3A
980 1.250V 55.7A
990 1.274V 58.1A/59.2A
1000 1.300V 61.5A

As it's really long ago I can't reproduce the Amps for the regular 850 (I don#t even remember the setting, just that all number came from the same setting), let's pull it out of the arse^H^H^H^H cognitive extrapolator:

850 1V 35A = ~35W
1000 1.3V 62A = ~80W

I did not write down anything else, and I can't measure the cards entire Watts I think. Maybe someone of you can make something out of it. Is it enough data to get an idea about how much core-clock we would have with a hypothetical 300W 5870?

 

[Florin]

Aaron, if that's so, why aren't any OEMs selling such OC cards? (are there any?) You'd think it would be a good market differentiator.

Anandtech offered this nugget a few weeks ago:

Sapphire will not be producing a 5870 Toxic, and the reason for that is that AMD won’t let them (or anyone else) offer a factory-overclocked card that runs significantly faster than their existing Vapor-X card (875MHz). This apparently isn’t a huge secret, but this is the first time we’ve heard this.

When we asked AMD about this, they told us that this all boils down to what AMD believes is safe operation for their chips. AMD allows vendors to factory overclock their chips to whatever point AMD feels is as high as they can safely go, and no higher. If any significant number of them could go higher, then AMD would have released them as a higher-end bin.

 

[eastmen]

I doubt it. Saw a review with a HD5870 overclocked to ~1GHz, it was still getting beat by the GTX 480 (default clocks).

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_480_Fermi/35.html

Cod4 1920x1200 4xaa 16x af
5870 115.1 fps
480 119.3 fps
480oc 129.4fps 770mhz core overclock 1048 memory best stable oc they could get
5870oc 132.3fps 1ghz core 1225 memory


5870 oc temp

Idle 30c
load 66c
Load+ oc 70c

gtx 480 oc temp

Idle 65c
load 96c
load oc 96c

While idle temperatures are OK, load temperature is close to what I would consider a limit for any graphics card.
Due to NVIDIA's fan profile in the BIOS, the card will reach 96° maximum at which point the fan speed will ramp up some more to bring the card down to 91°. My guess is that at 90° the fan would slow down again, but it never reaches that level while load is applied. This is also the reason why overclocking does not increase the temperatures any further, the fan goes up to 96°, spins faster and stays at that speed.

The 5870 has alot of head room in terms of heat.. Even overclocked at 1ghz its usoly 5c higher than the gtx 480 stock at idle !


I wonder what ati can put out 6 months later with higher voltages and a better cooler ?

 

[Kaotik]

Is it just me, or did no-one test the new AA modes? Everyone for what I checked quickly seemed to just quote nV numbers without actually testing them

 

[ninelven]

Congratulations, you found 1 test in all the internets to support god knows what idea...

 

[Florin]

I dare anyone to try. They are already right at the edge of the power and thermal envelope. They already carry the biggest coolers and the most powerful fans. They already generate the most heat and noise. There's no room for overclocking even on the cherry picked review parts.

Did we seen any review tell us of overclocking potential? I suspect it was banned by the Nvidia reviewer guidelines.

Here's a few. There are probably many more but eh.
Guru3D: Overclocking the GTX 480
Overclockersclub: 480 GTX
Legitreviews: Overclocking GTX 480
Hardware Heaven: GTX 480 overclocking

Oh, Nvidia apparently provided an updated version of EVGA Precision along with the cards so reviewers could play with it.

You guys

 

[Dave Baumann]

MSI's Lightning 5870 board at 900Mhz engine is already listed and there are others coming with 900MHz or higher engine clocks.

 

[eastmen]

Congratulations, you found 1 test in all the internets to support god knows what idea...

We are discussing that if ati didn't care about power and heat they could put out a competing part.

1ghz 5870 runs only 5c above an idle gtx 480.

Its obvious that there is room for cypress to grow. 1-1.1ghz core with 2gig framebuffer of 1.3-1.4ghz ram should make for a very competetive place if they wanted to go that way.

 

[Ethatron]

Yes, 900 ...

I expect something like 1350MHz+ core within 300W. Where is the review comparing 300W 5870 vs. 300W 480? Or a 190W 5870 vs. a 190W 480? I guess you have to go LN2 on 5870 to even hit 300W.

perf/watt, perf/mm2, perf/dollar, cost^time ... where just does the 480/470 yield?

 

[ninelven]

We are discussing...

No. We are not discussing anything. I was addressing aaronspink's specific claim that a higher clocked 5870 would likely outperform a GTX 480 most of the time.

 

[eastmen]

No. We are not discussing anything. I was addressing aaronspink's specific claim that a higher clocked 5870 would likely outperform a GTX 480 most of the time.

i provided a case where it does. Do you have other cases where a 1ghz ne doesn't ?

 

[ninelven]

Read my OP until you get it; if you don't get it, that is your problem not mine.

 

[Doomtrooper]

An overclocked 5870 with 2 Gigs of memory would close the gap on a 480 GTX, the minimum framerate is most certainly affected at those crazy resolutions with high IQ settings by video memory that maybe 1% of PC gamers play at.

 

[Jaaanosik]

That's not fair man. It has enough problems without having to compete with hypotheticals too. You may be proven right soon though depending on what AMD's schedule looks like.



Yeah that 10 bucks would go a long way

You mean a month, right?

 

[eastmen]

Read my OP until you get it; if you don't get it, that is your problem not mine.

So your unable to share this review that you claim to have seen ?

I at least backed up my claim.

 

[XMAN26]

I dare anyone to try. They are already right at the edge of the power and thermal envelope. They already carry the biggest coolers and the most powerful fans. They already generate the most heat and noise. There's no room for overclocking even on the cherry picked review parts.

Did we seen any review tell us of overclocking potential? I suspect it was banned by the Nvidia reviewer guidelines.

no, because Nvidia cards are ALREADY BASICALLY AT THE POWER WALL! Nvidia has already admitted the quoted TDP isn't actually the TDP! In other words, the 480 is certainly using more than 250W. OTOH, ATI TDP for the 4870 so far has proven to be conservative which means that they certainly have significant power headroom if they wanted to trade off power for performance in a standard vendor part. Nvidia has already made the power/performance trade off and doesn't really have any margin to increase power for performance.

http://www.hwgurus.com/testovi/grafike-karte/152.html?start=10

Yeah, sure looks like it is at the end of its clocking rope to me.

 

[XMAN26]

That was obviously more than one frame then.

Multiple times in various time frame was 2-3 second, which if you look at the [H] charts, those monster dips are in the 2-3 sec range. Just enough to cause a slide show.

 

[compres]

What's the difference between 'should've stayed in the labs' and 'product shouldn't exist'?

Notice that I never said that?

 

[aaronspink]

http://www.hwgurus.com/testovi/grafike-karte/152.html?start=10

Yeah, sure looks like it is at the end of its clocking rope to me.

Within a retail envelope it certainly is, at 825 MHz, even if the real TDP for the base card was only 250 (which nvidia has admitted that it is higher), that card is pulling north of 300W. If you believe that the real TDP is likely in the range of 275-300W for the 480, then that card is easily north of 350W.

OTOH, ATI can increase their power by at LEAST 75W and still be within the PCIe limit. In other words, ATI could if they wanted to, likely release a ~950 MHz card right now with their current 5870 parts at roughly the same power envelope as the 480.

 

[DemoCoder]

An overclocked 5870 with 2 Gigs of memory would close the gap on a 480 GTX, the minimum framerate is most certainly affected at those crazy resolutions with high IQ settings by video memory that maybe 1% of PC gamers play at.

Yes, memory could be an issue at ultra-high resolutions and AA, but this doesn't explain Fermi's apparent speed advantage at low resolutions. It could be that triangles are too small at the lower resolutions, and bumping the clocks might not improve this much.