Paradigm Shifters Surround Review GTX460 SLI 1GB & 2GB

[skipclarke]

Discuss the review here.

[Paradigm Shifter]

Love the graphs - look better than mine did! - although I do like to see the numbers (particularly on the VRAM charts) on each bar... :)

What's up with the WSGF front page? It says it was posted in July!? :?

Edit: differences in X58 board are minor to be honest. The effect of having an eVGA X58 vs. a Gigabyte X58 will be at most a few fps. However, the 12GB of RAM you have might make a difference since games swap to system RAM when they run out of VRAM. If I can get another 6GB of RAM, I might rerun a few tests (and do Batman, Battleforge etc) so there is something to compare. :)

I want to do some 4GHz testing too, but right now I just don't have the patience to rebench all that lot. :lol:

[skipclarke]

Bad copy/paste for the D/T stamp... Fixed.

I don't remember if the 12GB made a big difference over 6GB, but that was back with older games. More changes of something making a difference now. I totally understand the time involved. If you can get the RAM, then try one or two of the titles with an automated demo.

[scavvenjahh]

Great stuff - both the article and its presentation. :twothumb

I usually pick the bigger GPU and thus bought 480s for the sheer horsepower, but 460s getting 2GB VRAM and DP outputs are much more exciting stuff. Since I'm not likely to upgrade my monitors anytime soon, I may go for 2x 460 2GB DP + TripleHead2Go DP, and maybe try 3-way SLI later on*. Lower temps, more RAM, no forced SLI, no Bezel Compensation worries... Might be an interesting alternative for 5040x1050 gaming.

Thanks for this and here's hoping we'll be discussing 3GB vs. 6GB cards soon enough ;)

*Edit: argh, the 460 is not 3-way compatible, I forgot yet again !!

[Paradigm Shifter]

*Edit: argh, the 460 is not 3-way compatible, I forgot yet again !!

Haha... yeah, if the GTX460 could do tri-SLI, I'd probably look at that before thinking about switching to 470/480s.

But before I switch, the 480's need to have 3GB of VRAM... :(

[mattsimis]

Excellent review, though Im still cautiously optimistic my 5040 res just comes in under the 1GB VRAM cap. :)

[Paradigm Shifter]

Haha, as long as you don't try going for high levels of AA, I'd hope it does too.

I'm still waiting for a driver that fixes the GRID bezel-management crash issue. And hopefully it'll also fix my inability to run at non-native Surround resolutions. I'd quite like to try 5040x1050 just to see how much VRAM it does chew...

[tineras]

This is making me regret purchasing another GTX 275. I should have just taken the plunge and moved to the 4xx series. :doh

[ThE_MarD]

Heyyo,

I'd like to say for the GTAIV benchmark there is no memory leak. It's how every engine for GTA was designed since GTA 3. They only load the essentials, and then as you travel the city it loads what it needs.. quite a bit like Unreal 3 Engine will do playing games like Gears of War (mainly on xbox where there is a lack of RAM compared to PC's). They do this to lower the loading times cause that's a constant complaint for most games. They did it in a smart way for GTA though.. the way you start in a small secluded area (a house, garage, etc) that way only the stuff in that small space is loaded, then it starts up the rest as you leave the starting area.

I do find it odd the save games would not load in surround.. maybe the game was not fully tested to work with it?

[Paradigm Shifter]

Given that Surround/EyeFinity weren't around when GTA4 was released, I'd say that it not being tested was quite probable.

However, the game failing to load in Surround resolutions with 1GB cards is fully reproducible so I suppose it's a texture/paging issue.

[Skid]

I figgered it was worth mentioning here, the price of GTX 460s have fallen though the floor, you can pickup EVGAs 1GB model for less then £160 now and the 2GB models for just shy of £200.

Also the price of 470 have fallen as well, with EVGAs standard model setting you back just shy of £200 as well. The price drop is largely down to NVidia trying to keep up with ATI on price point as well as the the new 6000 cards that actually compete closer to the GTX 460 and 470 then they do the ATI 5850 and 5870.
http://www.tomshardware.co.uk/radeon-hd-6870-radeon-hd-6850-barts,review-32028-23.html

[Paradigm Shifter]

I was lucky when I got my 2GB GTX460s, obviously, as they launched really cheap, then increased in price. While that 2GB Palit card has dropped again, it's still £4 more than I paid. Granted, £4 ain't a lot when you're playing in the £190 region, though. :D

Actually quite tempted to see how a pair of GTX470s does in Surround at those prices... the VRAM drop might be worth the extra GPU grunt...

[Paradigm Shifter]

Since a 12cm flexible SLI bridge arrived and I can now move my second GTX460 down a PCI-E slot, I did it today.

So from going to a 16x/16x arrangement to a 16x/8x arrangement resulted in... no framerate decrease in Just Cause 2 or DiRT2 (only two games I've tested so far aside from Civ 5, which I didn't test in this review).

But... my GPU1 temps have dropped 25*C from the move - it's getting extra air, instead of being jammed right in the back of GPU2.

Watercooling? Who needs it? ;)

Having said that, before I did the move, I saw Civ 5 (in Surround) max out both GPUs constantly, and get GPU1 to 98*C! :shock: It probably would've gone higher, but I turned Surround off and played it single screen and temps didn't break 80*C. Civ 5 DX11 uses DirectCompute (apparently) to do some of the AI calculations... end result it hits up a DX11 GPU something fierce...

[scavvenjahh]

This is not exactly related with VRAM usage, but there's something fishy about Dirt 2...

First, the framerate is not capped at all :? so you shouldn't hit that 60-fps wall in all res's.

I also have a major problem with water on Marina Del Rey (L.A):
5220x1050, Preset: Ultra, 4x MSAA (max VRAM usage: 1380 MB)
Setting Water to High or Ultra immediately cuts my framerate by 4... and there is no water on this track :shock: In fact, with High water the framerate crawls while I'm in the arena (as low as 22 fps and very inconsistent), then is butter smooth (100+ fps) on the rest of the track, until I reach the arena again. With Medium water everything is fine across the whole track.
I tried with the Ultra Low preset, same thing: Medium water is okay, High water makes the arena part all sluggish.
In DX9 mode the framerate drop inside the arena is less of a problem, but it happens nonetheless (framerate drops to 75fps).

I say again: inside the arena, the Water setting alone has a massive impact on my framerate. Crowd, Particles, Post Process... all have a minor impact compared to Water being set to either Low/Medium or High/Ultra.

So I don't know what to think. There is no bloody water in the arena !!! And driving through Ultra water on other tracks has no significant impact on the framerate, anyway.
If you can test that as time permits, I'd like to know your results and what you think of mine.

While I'm here, there's also some minor editing to do on the Dirt 2 results page... Stuff about Figure 15 & 16...
Oh and Ibrin, the x axis on each page's last chart should read "VRAM usage" on the x axis, not "Frames Per Second". ;)

[Paradigm Shifter]

I don't know why DiRT2 was capped. It's a bit weird. I've checked that Vsync was off in nVidia control panel (forced off) and in game (off). But it's still capped at 60fps. On another system, the vsync toggle works correctly. I'll check out that track and see what happens.

The Ultra water thing was something I noticed during testing IIRC. On 1GB cards, set water to high or ultra, no matter what the other settings are, and it was an instant framerate bomb down to single digits from mid/high 30/40s.

I had DiRT2 hard crash on me yesterday. Have no idea why - I was bombing round a rally track, about 200 yards from the finish, and suddenly *bang* - black screen, a fraction of a second of audio on permanent repeating, the three finger salute (Ctrl-Alt-Delete) did nothing - eventually had to hard reset. Had been playing for about an hour. Was something like the ninth time I'd tried that particular rally and managed to get round a hairpin at a half decent speed, so who knows, maybe the game had got sick of me hitting 'Menu - Restart' time after time after time... :lol:

Haven't had the guts to load it up again since.

Forgot to set Afterburner to log to file when I updated it, so no idea what GPU usage, temps etc were like. Shift works fine, as does Civ 5 and Just Cause 2 (Shift and JC2 hit both GPUs 100% - Civ 5 only hits the first 100% and the second about 30% but I'm running 258.96 drivers which don't do Civ 5 SLI well.

Thanks for the error pointouts - Ibrin'll have to fix them, I'm lost in the wiki... :oops:

[Paradigm Shifter]

Am not able to replicated an "enormous" framerate drop on my 460s. Although VRAM usage goes sky high on that track for some reason.

Will try again. Driver version, mate?

[scavvenjahh]

260.99 WHQL (480s) & 263.09 WHQL (580s), clean driver installs, same thing.
I'll try with one/two/three cards at 2560x1600 to make sure it's not an SLI issue on my end...

[Paradigm Shifter]

I can't get Surround working on any driver other than 258.96. Well, I say that, but I tried it with 260.something and it crapped out - not had the enthusiasm for trying it again yet.

[scavvenjahh]

Been doin' some overdue benchmarkin' and I'm now pretty sure that the issue is SLI-related, and may have been caused (or revealed) by my PCI-E config. I still have to remove a card and test 16x/16x SLI, but both 16x/8x SLI and 16x/8x/8x tri-SLI have the "water problem", whereas single-card gameplay is completely unaffected. I'll post my results on the Nv Surround board (don't expect fancy charts !) :P

Sorry to have brought this up.

[Paradigm Shifter]

It's alright, no worries.

I'm going to shoot down (at least partially) your theory about it's being PCI-E bandwidth - I'm running my 460s 16x/8x at the minute, rather than 16/x16x as putting them in slots 1 and 3 rather than slots 1 and 2 gives GPU1 a lot more air and stops it getting to 98*C in DiRT2. And there has been no noticeable performance alterations going from 16x/16x to 16x/8x.

Of course, that doesn't rule out the idea of motherboard specific issues... but still. ;)