Well, it only feels like six months and twelve days since I had my last bout of madness and, against many recommendations, went and bought three 480s. Now the new and improved 385 485 580 is here and has been released to rave reviews. I weighed up my finances and current happiness levels and decided to take the plunge. Am I glad? Read on... Prologue Of all the Nvidia board partners, Palit was the only one to offer a factory overclocked variant. Not only that, it was quite a significant overclock (835MHz Core vs 772, 4200MHz RAM vs 4008), which for £20 more is a bargain, no doubt about it. Unfortunately, on the day of delivery, Palit had to recall all their Sonic cards due to unstable settings. In the end, I ended up with three KFA2 cards. First impressions Say what you want about the 480, it has character. Never before has a reference card had heat pipes sticking out the side! Unfortunately, the 580 just looks, well, boring in comparison. Have a look: Don't tell me that the 480 doesn't look more awesome! You also may have noticed that KFA2 forgot to put the stickers on. Personally, I'm rather glad they did! Here are the 580s in tri SLI: Again, rather boring if you compare them to the 480s in my original review. Noise The 480 was originally lambasted for being a loud son of a gun; since then, I think Nvidia have refined the thermal issues a bit and tweaked the fan profile for the better. Temperatures for my 480s certainly got better, dropping from close to 100C to 93C, with an 800MHz core (stock voltage). So what of the 580? Well, good news. One 580 is inaudible at idle (though so was the 480) and doesn't have any change in noise when under load! On the other hand, SLI is still a noisy affair - it's important to get at least one gap between cards if you can. SLI'd 580s are quieter than SLI'd 480s but for a reason that'll be covered in the overclocking section. Heat I didn't monitor heat until I had all three cards installed but from the fact that one 580 didn't make any noise, I'm guessing it ran at ~80-85C. In tri SLI, the primary GPU reached temperatures of 91C for 'normal' benching and 93-94C for Furmark and the Heaven benchmark. However: When I was benching, I noticed that even though the cards ran cooler, the room itself seemed to get warmer that when I ran my 480s flat out. This implies that the new cooling is better at getting heat away, rather than the actual card itself being cooler. Just a thought... Driver Maturity This is probably best discussed after seeing the results, but I feel the drivers need a bit of refining; they were okay but lacked a certain...something. Benchmarks You've probably already skipped ahead to this bit, haven't you? So, all tests at 2560x1600, all game benchmarks with highest possible settings in-game (and 16xAF, high quality textures in Nvidia control panel), all benchmarking software on highest settings, Nvidia control panel settings at defaults. GO GO GADGET GRAPHS! A nice increase over the 480, and probably the only game that actually has better scaling with a third card (go figure). This is using the GPU test. As you can see, 60FPS mean is achieved! Also using the GPU test. Over 40FPS - not much more to say, really. I wish the maximums could be brought down for some better minimums. Yeah...the 580s are beaten by the 480s. WHAT THE HELL NVIDIA!!! I really should get around to playing this game. Scaling seems to be an issue with the current drivers. Overclocking Allow me to sum up my experience of overclocking the 580s with a smiley: Overclocking my 480s was remarkably easy using Afterburner - I downloaded the new beta and gave the 580s a go. I use OCCT for error checking, which in my opinion is the only good bit of software for checking core stability. So, I fired up OCCT and having read other reviews, gave the core a modest bump to 800MHZ (shaders 1600MHz). After a few minutes, a couple of errors had shown up! I thought this was a bit rubbish, so I switched to Furmark, which allowed me to get to 850MHz (I actually stopped here as I knew it was lying). Sure enough, at 850MHz, no game would start. Even 800MHz couldn't be maintained for long. A bit annoyed, I gave in to voltage tweaking. I raised the core voltage to 1.088V and managed to squeeze 842MHz out of the core, which survived five Heaven benchmarks in a row, so good enough. I then turned my attention to the memory but after doubting results, I left it alone. Anyhoo, look at the pretty graph: When overclocked, my 480s show a 10% improvement in mean frame rates whereas the 580 shows a 5% improvement. Admittedly the 480s have a 14% overclock (with a tiny memory tweak) as opposed to the 580s' 9%, but still, it's quite disappointing. Minimum frame rates fare a little betterI could push the voltages higher but I won't for reasons below. Temperatures rose to 98C during benchmarks when overclocked, which leads me to another thing; I discovered that the 580 has its fan locked to a maximum of 85% (at least through Afterburner). I really hope this is an Afterburner limitation as I would like to run the fan a little faster. Update: it turns out that an 842MHz overclock wasn't stable either, so the cards are now running at 830MHz with a 1.1V core. Maximum temperatures still get to 98C. Conclusions 1) The 580 as a single card is truly quiet. 2) Is a single 580 £100 better than a 480? Not really (in my most humble opinion) but those who really do value a quieter card will. 3) Is replacing SLI'd 480s with 580s worth it? In my opinion, absolutely bloody not! Given the extra overclocking headroom of the 480, the 580 doesn't actually offer much more. On the other hand, if you really, REALLY like Crysis, I guess it's worth it. FAQ Are you happy with your purchase? NO!!! If I'd upgraded straight from my GTX 260s to 580s then I would, but the extra money to go from 480s to 580s is not worth it. Still, I should've known better. Why, oh why didn't you wait for the 69xx series?! I dunno. I must be an Nvidia fanboi in denial over his condition. So, what's next? Not spending money on computer bits for another 3-4 years, that's what!
I did read on another forum that the drivers are not yet mature enough for the 580's would you agree with that.
What's the CPU use look like? Possible one of the other factors (bus lanes as an example) are limiting the FPS? I've also got the E760 and I know it doesn't have the nVidia 200 chip like the E759 does for the extra bandwidth (albeit with slightly higher latency)...
I hope the trend of home-made reviews will last Maybe with some driver improvements, you will see a bigger performance increase?
Wow...two of those graphics cards combined cost more than my entire computer. I salute you and your spending power.
Honest unbiased review, enjoyed reading it! supposedly gainward have a factory oc'd golden sample that's identical to pallit's. This hasn't materialised in the real world either yet so maybe it's suffered the same fate. I've decided to jump back to nvidia this round and have preordered the gainward card I mentioned. I'm thinking this card with good drivers will at least match the 5970 on new games, so I'll happily swap my 5870's to not have to wait for drivers to unleash the crossfire potential. I've a Gtx 280 to throw in as a physx card with it so it'll be nice to not have to "fix" nvidia drivers to activate physx... Guess I'm getting lazy and loosing interest in spending hours fiddling and searching the net trying trying to get the most out of my rig, it's factory oc and pure drivers for the next 10 months (sips piña colada)
I absolutely agree with that. The fact that 480s beat 580s in Just cause 2 is evidence enough. I wasn't watching CPU usage the whole time, but it definitely wasn't busy! As for the bandwidth, I don't think it's too much of an issue unless you're running quad SLI I like home grown reviews too, they're more realistic! Not only that, I've probably spent more time achieving a truly stable overclock than most review sites have; yes, I can run my GPUs at 850MHz but they'll fail on a second attempt at a benchmark. As for the drivers, see above. You're welcome! I wouldn't be surprised if the gainward card mysteriously disappears for the moment. And don't get me started on Nvidia's absolutely fantastic idea of disabling PhysX with an ATI/AMD card present .
How hard was it to get the 580s? I remember before when you bought the 480s you were all over the forums looking for stock. What you planning on doing with the 480s? Nice review BTW, agree with Ph4ZeD That user reviews are very useful.
Nice review! It's a shame initial results aren't quite what you were looking for, but as other have already said, hopefully with driver updates things will improve.
I ordered the original cards (Palit) at about 9.15 on the 10th; there were four left but due to bank problems I had to call back, which I did 15 minutes later. By that time, the other one had been grabbed. Come the next day (when Scan informed me that Palit were recalling the cards), there were four 580s left! And I've already sold 2 480s on these very forums, the third is lined up with a buyer.
Great review Pete, thanks for that. It's a shame you havn't seen a greater improvement. Looking at your scaling results it does look like you are cpu limited (to a degree) at 2560x1600. Could you possibly run Crysis with maxed out AA and AF (higher than 4xAA) to see if the results improve your scaling?
Doesn't this one count? http://www.evga.com/products/moreInfo.asp?pn=015-P3-1589-AR&family=GeForce 500 Series Family&sw=
I can sort of see what you're getting at. Anyone want to send me a 980x so I can confirm? AF was already forced through drivers at 16x, though I think due to the way Crysis does it's 'custom DX11' thing that AF gets disabled. As for running more scaling tests, I'm loathe to take the cards out again! I might get away with running an 8xAA test but 16x will definitely cause the cards to go over their RAM - and that results in nice single figure stats! Good spot!
sweet.. the 85% fan should be unlocked with fermi bios editor lite- sure someone on guru3d or xtreme forums would unlock it for you once the writer figures it out
sweet cards, definitely some improvement over old 480. it's clear which game are bottlenecked by shader core counts while which game are bottlenecked by other elements of graphics card. Batman AA, Stalker CP are both clearly bottlenecked by core count, and saw a good minimal FPS improvement on 580.
An interesting read, I'm thinking of also going up from 480's to 580's. I'm curious about your core clocks though, from what i've read most people are hitting 1000+mhz on the cores on H2O and most around the 940 mark on stock, while the Ln2 boys are floating around 1200mhz. Could your clocks possibly be a power issue ? Obviously it's easier to get clocks higher and more stable on single card jobs whereas multi-card setups tend to not clock so high, i did find i was able to hit over 900 on my cores though with tri-sli 480's after adjusting my IOH and ICH voltages. Personally I run my 480's off of a completely independent 1200w silverstone decathlon PSU and the rest of the rig runs from a 1250w enermax revolution, are you just using the thermaltake on it's own ? 3 480's can pull like 1200w by themselves when overclocked, I wouldn't think that 1500w is leaving you much headroom especially with an overlcocked CPU and other hardware in your system. The 580's seem to be also drawing slightly more power than the 480's, well depending on the review you read.
Bummer, I am sorry to see that your not happy with your 580's. Hopefully with driver revisions you will see an improvement and i hope knowing that you have the fastest GPU setup avaiable will give you some comfort.