GTX580 reviews thread

[Skurge]

GTX 480 to 580 is not very much like 280 to 285 at all. The GTX 285 had all of the same internals of the 280, and the only differences were the clocks bumped up 10% and 55nm instead of 65nm. It was more of a die-shrink with a small bump in clocks.

For those who had 280s and 285s (myself included - I had 2x280s and 2x285s due to free step-up at the time) they were exactly the same. The 280 was already very overclockable, and I actually regretted "upgrading" because I lost over-volting ability due to NV using a cheaper voltage tech on the 285. Also, it wasn't really any more power efficient either (same with 55nm 260s if you recall).

The 580 not only is speed-bumped, but has more architecture enabled. The amazing thing is this WASN'T a die-shrink at all, AND it manages to be more power efficient. I know it's not hard to get worse than GF100, but you get my drift....

I will say again that a 20% speed bump is not trvial. We get cooler, quieter, more overclockable on non-exotic methods, plus slightly more efficient. Thats a big win for NV compared to where they have been. If Fermi was this good from the get-go, I probably could have gotten my 5870 for $50-75 cheaper.

I gues that why its called GF110 and not GF100b. I really wish TSMC didnt cancel 32nm. We might have seen a much more impressive launch for both the 6000series and 500series.

 

[DaveSimmons]

The 580 is a nice improvement over the 480 in every way, and the best single GPU card if you're willing to spend $500. Good job, nvidia.

 

[notty22]

F1 is a very unpredictable game. Its not what i would call polished.
Considering, it seems to be the Dirt 2 engine.
They can't officially release the dx11 functionality, unless they recently have.
Though it is somewhat fun, and racing in the rain is VERY cool.
I saw in a recent Top Gear, a official F1 race track , or F1 themselves has built a game/simulator around this very game. The courses are officially or whatever created to scale. So drivers race the game/course to familiarize themselves, cool ? yeah.

 

[Daedalus685]

GTX 480 to 580 is not very much like 280 to 285 at all. The GTX 285 had all of the same internals of the 280, and the only differences were the clocks bumped up 10% and 55nm instead of 65nm. It was more of a die-shrink with a small bump in clocks.

For those who had 280s and 285s (myself included - I had 2x280s and 2x285s due to free step-up at the time) they were exactly the same. The 280 was already very overclockable, and I actually regretted "upgrading" because I lost over-volting ability due to NV using a cheaper voltage tech on the 285. Also, it wasn't really any more power efficient either (same with 55nm 260s if you recall).

The 580 not only is speed-bumped, but has more architecture enabled. The amazing thing is this WASN'T a die-shrink at all, AND it manages to be more power efficient. I know it's not hard to get worse than GF100, but you get my drift....

I will say again that a 20% speed bump is not trvial. We get cooler, quieter, more overclockable on non-exotic methods, plus slightly more efficient. Thats a big win for NV compared to where they have been. If Fermi was this good from the get-go, I probably could have gotten my 5870 for $50-75 cheaper.

Well yes, but my point is that this is not a great refresh because the 580 is great (though it is a nice product) but more so because of the failings of the 480 that the contrast is so pronounced. All I am saying is that if the 480 didn't have the leakage and yield issues the 580 would only bring better clocks and the tweaks with the z-stencil. I( guess we are just a bit skewed on what refresh means given the rather poor showing of them in recent years. No more radeon 9700 to 9800 jumps at least.

If all of the cores worked on the 480 this would be nearly the same as the 280 to 285. As it is it is more like the 7800 to 7900, or 1800xt to 1900xt in the boost of performance.

Though the 280 to 285 really seemed like a shrink to make money and more or less clone the card as opposed to improving the card in the eyes of the purchaser.

Regardless though, the 580 is a great halo card. It could have been named the 485, it could have been named a lot of things. But it is much more akin (imo anyway) to 480 done right rather than a new series of card.

 

[Daedalus685]

F1 is a very unpredictable game. Its not what i would call polished.
Considering, it seems to be the Dirt 2 engine.
They can't officially release the dx11 functionality, unless they recently have.
Though it is somewhat fun, and racing in the rain is VERY cool.
I saw in a recent Top Gear, a official F1 race track , or F1 themselves has built a game/simulator around this very game. The courses are officially or whatever created to scale. So drivers race the game/course to familiarize themselves, cool ? yeah.

The [H] review was done on the dx11 patch, or at least Kyle mentioned that the patch just came out and linked where to find it. Come to think of it it wasn't clear to me if he meant he didn't use it as it came out during testing or that it came out just in time... I'll have to actually read instead of skim.

 

[BrentJ]

The [H] review was done on the dx11 patch, or at least Kyle mentioned that the patch just came out and linked where to find it. Come to think of it it wasn't clear to me if he meant he didn't use it as it came out during testing or that it came out just in time... I'll have to actually read instead of skim.

We used the patched version of F1 2010 running DX11 effects.

You have to log into LIVE to get the patch, even if you have the STEAM version of the game. Steam won't update the game automatically, it requires logging into LIVE to get the patch. We did so, and used it, so our tests are done with the brand new DX11 features the patch has implemented in F1 2010.

 

[blastingcap]

PingviN, please sit there and tell me that you didn't know about AMD's recent massive misinformation campaign? You thought I was making fun of AMD and so you take a mocking stab at me? WTF dude?

Wow, I guess there are some people who didn't know. chewietobbacca included according to his seemingly clueless post above about it. I thought you guys especially would be apprised of that situation. Well, you can ask Silverforce11 or BlastingCap about it then if you don't believe this coming from me.

RE: Barts:

I'm not sure how I got dragged into this, but I can guess. I was stuck in IGP for 2 months (my backup, a 6800XT, broke on me days after I sold my 5850 in anticipation of replacing it with Barts) and kept up to date on Barts rumor due to my painful IGP situation, but I never had much interest in Cayman, so I am not as up to date about that. Barts had massive misinformation going on, some or even most of it intentional. The naming was mixed up (Southern, Northern islands), AMD had a change of heart re: number of shaders vs. clockspeed (1280 turned into 1120 for Barts XT, but clockspeed ramped up to compensate), and even naming was indeterminate, bouncing back and forth between 67x0 and 68x0. Silverforce claimed to have inside knowledge, something I never had since all my info was coming from the rumor mill. So please don't lump me with him, especially since some of his predictions were way off base despite his alleged insider info.

RE: Cayman:

We know these facts about Cayman, and I don't use the word fact loosely:

1) It is one-half of the Cypress fork (Cypress spawned Barts in the lower branch, Cayman in the higher branch). Source: Barts launch slides, made public at review sites.
2) 40nm TSMC chip
3) It was scheduled to launch soon after Barts and definitely before the end of the year (Barts launch slides; David Hoff)
4) Two Cayman chips power one Antilles dual-GPU card (Barts launch slides)
5) Price will be above $250 (Barts launch slides)

The following are VERY LIKELY true:

1) It will be a 4D and not 5D setup (leaked slides, David Hoff's comment about a new-ish architecture, leaked slides, AMD employee talking about taking greater risks with Cayman)
2) It was scheduled to launch in November (Hoff comments about rolling out multiple SKUs prior to Christmas, leaked slides, history of Evergreen rollout)
3) Cayman XT performance will be somewhere between 30% and 50% faster than Cypress XT
4) Some sort of tessellator change (leaked slides, Barts launch slide implying a greater emphasis in this area, AMD employee talking about taking greater risks with Cayman)
5) Connectivity same as Barts reference (e.g., 2 mini-DP, HDMI, etc.)
6) Price competitive with NV's offerings

As for misinformation, anything's possible but I doubt AMD is trying to mislead about launch dates because they were accurate with the Barts launch date despite all the FUD swirling around about how Barts was going to be delayed. But there is a real possibility of Cayman delays, due to driver issues from going to 4D.

As for yield rumors, Fudzilla seems to have been the origin for the "Cayman is delayed due to poor yields" rumor which correct me if I'm wrong was picked up by Techeye without attribution. From there it kept spreading. Fudzilla has historically been an awful source for info, yet some sites parrot it anyway, perhaps to share in the pageview revenue.

I don't keep track of Cayman much, but according to Chinese leaks (historically more reliable than Fudzilla, especially people like nApoleon on Chiphell), if Cayman is delayed it will be due to drivers/BIOS tweaking. Remember that Cayman is 4D, not 5D, so this is not like all the other 5D AMD cards. I'd say the probability of Cayman being delayed longer than November 30 is ~33%. (Note that even if it launched on Dec. 1, that'd still be a "delay" by this metric.) That's just a wild-arsed guess based on gut feeling.

By the way, remember that Fudzilla and parrot sites claimed Barts was going to be paper launched or not launched at all till November. Remember that Fudzilla had a series of "Fermi is just around the corner" articles from last fall to last winter--predictions that they quietly deleted as fall turned into winter turned into spring. They have no journalistic integrity.

RE: HARDOCP:

Russian, HardOCP reviews' hallmark is that they play through games rather than relying on canned benches because canned benches can be misleading. You should know that already. Their games selection is also partly dictated by members; they polled members recently about which existing or upcoming games that members would like to see benched. That said, yes, in an ideal world HardOCP would review more games, especially the heavy hitters where you need every last fps to stay above 30, but they have limited resources and would look like jerks to their readers if they ignored their own polling of readers' requests.

 

[Daedalus685]

We used the patched version of F1 2010 running DX11 effects.

You have to log into LIVE to get the patch, even if you have the STEAM version of the game. Steam won't update the game automatically, it requires logging into LIVE to get the patch. We did so, and used it, so our tests are done with the brand new DX11 features the patch has implemented in F1 2010.

Hehehe, sorry Brent, you wrote the article not Kyle .

Also thanks for the confirmation.

 

[exar333]

Well yes, but my point is that this is not a great refresh because the 580 is great (though it is a nice product) but more so because of the failings of the 480 that the contrast is so pronounced. All I am saying is that if the 480 didn't have the leakage and yield issues the 580 would only bring better clocks and the tweaks with the z-stencil. I( guess we are just a bit skewed on what refresh means given the rather poor showing of them in recent years. No more radeon 9700 to 9800 jumps at least.

If all of the cores worked on the 480 this would be nearly the same as the 280 to 285. As it is it is more like the 7800 to 7900, or 1800xt to 1900xt in the boost of performance.

Though the 280 to 285 really seemed like a shrink to make money and more or less clone the card as opposed to improving the card in the eyes of the purchaser.

Regardless though, the 580 is a great halo card. It could have been named the 485, it could have been named a lot of things. But it is much more akin (imo anyway) to 480 done right rather than a new series of card.

I would argue it's a bigger change than the 280->285, but I get where your coming from here. The 480 was a pretty good card in terms of performance, but it really fell short in a lot of other areas. As for refreshes, I completely agree we get better refreshes now than what we got from say 9700-9800 or 7800 to 7900.

 

[Seero]

To make it sound ridiculously simple for everyone. 580 is "Fermi done right!"

We knew there was something wrong with the GF100 down to its design. This may be a wild assumption, but I don't think heat is the reason why 480 has to have SM disabled. I recalled that JHH said that those parts runs fine individually, but ain't able to talk to each other. If it was due to technical reason rather than physical (heat) barrier, then the 580 design must not be the same compare to 480.

Regardless of what you compare it to, it is indeed a good card. This time everything on board are being used, not like 480 where the card is only running at 80%. Maybe the iteration does not actually has speed increase per core, but there are now more active cores with less electricity. Clearly, it isn't brute force.

I like to see Charlie's opinion about this "unmanufacturable" chip.

Oh here it was Charlie's vision from semiaccurate

 

[exar333]

To make it sound ridiculously simple for everyone. 580 is "Fermi done right!"

We knew there was something wrong with the GF100 down to its design. This may be a wild assumption, but I don't think heat is the reason why 480 has to have SM disabled. I recalled that JHH said that those parts runs fine individually, but ain't able to talk to each other. If it was due to technical reason rather than physical (heat) barrier, then the 580 design must not be the same compare to 480.

Regardless of what you compare it to, it is indeed a good card. This time everything on board are being used, not like 480 where the card is only running at 80%. Maybe the iteration does not actually has speed increase per core, but there are now more active cores with less electricity. Clearly, it isn't brute force.

I like to see Charlie's opinion about this "unmanufacturable" chip.

Oh here it was Charlie's vision from semiaccurate

Charlie sure must be uncomfortable with his foot in his mouth about now...I see 580s on Newegg right now!

 

[Daedalus685]

Charlie sure must be uncomfortable with his foot in his mouth about now...I see 580s on Newegg right now!

I doubt he cares. His page hits are already tallied.

 

[Arkadrel]

Here is Linus taking apart a 580 down to the Vapor Chamber.
Very cool. http://www.youtube.com/watch?v=ncEU1mrN2oc&feature=channel

+1 respect for that. Awesum video.

It had actually never occurded to me that you could change the thermal paste on the GPU to some more expensive type if you wanted. Ive never personally done that to a grafics card Ive owned. I usually do it for CPUs if they come with that dark thermal pad type stuff, and buy some decent thermal paste.

Even if the vapor chambers cost abit more than heat pipe system, they are so much more awesum. Im kinda hopeing every AIB takes on this approce for future cards so we as consumers can have cooler cards in the future too (reguardless of brand you buy).

 

[Skurge]

+1 respect for that. Awesum video.

It had actually never occurded to me that you could change the thermal paste on the GPU to some more expensive type if you wanted. Ive never personally done that to a grafics card Ive owned. I usually do it for CPUs if they come with that dark thermal pad type stuff, and buy some decent thermal paste.

Even if the vapor chambers cost abit more than heat pipe system, they are so much more awesum. Im kinda hopeing every AIB takes on this approce for future cards so we as consumers can have cooler cards in the future too (reguardless of brand you buy).

Its on the 5970, sapphire used it on the 3870 and still use it today. XFX use it on the 6850. So, they tech was there, they just chose not to use it. Vapor chambers aren't that special to begin with. coolers like the Twin Frozr are equally as good. It's just that reference coolers generally suck.

 

[Seero]

I have a question about the vapor chamber design. Doesn't water accelerates towards the center of the earth. If the GPU is parallel to the surface of the earth and the heatsink is on top of it, then I can see why it works, but doesn't GPU being placed underneath the GPU with respect to earth? So then whatever liquid inside the chamber is actually sitting away from the GPU?!?

 

[tviceman]

Its on the 5970, sapphire used it on the 3870 and still use it today. XFX use it on the 6850. So, they tech was there, they just chose not to use it. Vapor chambers aren't that special to begin with. coolers like the Twin Frozr are equally as good. It's just that reference coolers generally suck.

I'm guessing the twin frozr II is probably a better cooler. I've got a 33% overclock on my card with the twin frozr II and it's still pretty damn quiet.

 

[T2k]

I doubt he cares. His page hits are already tallied.

Exactly. OTOH you have to give it to him that he was dead right on the GTX480 all along the way... I bet NV managed to figure out who was his source inside mfr'ing and made sure it does not leak anymore. :twisted:

 

[badb0y]

15-20% performance gain over GTX 480 on the whole and lower power consumption with less die space?

I am impressed. Now show me a GTX 595 .

 

[Seero]

Exactly. OTOH you have to give it to him that he was dead right on the GTX480 all along the way... I bet NV managed to figure out who was his source inside mfr'ing and made sure it does not leak anymore. :twisted:

Please explain the "dead right" part.

 

[tviceman]

Please explain the "dead right" part.

He must be referring to the part where charLIE said there was only going to be 10,000 total GF100 cards in existence because it was unmanufacturable.

 

[Arkadrel]

Charlie was wrong about no 580s being out in november/december.
Charlie was right about it being about ~20% faster than a 480.
Charlie was right about it still being obscenely huge (nvidia say 520mm^2)
Charlie was right about it not being faster than the hemlock (5970) (close call tho)

by no means was charlie dead right on this one... the papir launch didnt happend, its a soft launch.

 

[T2k]

Please explain the "dead right" part.

He was right about JHH lying to the public.
He was right about card not arriving in the Fall.
He was right about JHH showing a fake wooden card to fool investors.
He was right about He was right about Fermi not arriving in December.
He was right about the multiple, endless respins.
He was right about the design defects that caused the long-long delay.
He was right about Fermi not arriving in January-February.
He was right about Fermi not coming in 512 cores flavor.
He was right about Fermi being unmanufacturable with original clocks and cores.
He was right about Fermi being late, even beyond March.
He was right about Fermi being out-of spec both in terms of power and thermal with its original clocks and specs.
He was right about having horrible yields.
He was right about Fermi being cut down to 4xx cores to make it manufacturable.
He was right about lowered clocks to make it manufacturable and standard-compliant -> marketable whatosever.
He was right about the final shipping date.

Did I miss something..?

 

[T2k]

Charlie was wrong about no 580s being out in november/december.
Charlie was right about it being about ~20% faster than a 480.
Charlie was right about it still being obscenely huge (nvidia say 520mm^2)
Charlie was right about it not being faster than the hemlock (5970) (close call tho)

by no means was charlie dead right on this one... the papir launch didnt happend, its a soft launch.

Exactly and this is why I think this time he only had AIB sources to rely on for leaks.

 

[happy medium]

Charlie was right about it being about ~20% faster than a 480.
Charlie was right about it still being obscenely huge (nvidia say 520mm^2)
Charlie was right about it not being faster than the hemlock (5970) (close call tho)

This was news repeated by another source , not Charlie though.
The one thing he predicted (Nvidia won't release till DEC,) he got wrong.

 

[Arkadrel]

@T2K

Did I miss something..?

Yes you only took notice of the times he was right...
you missed all the times he was wrong ^-^

Granted I think hes right more often than hes wrong.