I change my mind about the GTX280

[john19055]

I kept saying that my two 9800GTx running in SLI was better and faster then my one BFG GTX280 OCX.I ran my two 9800GTx in SLI for about a month and it would get 16,350 in 3DMark06 running at 780/1900/2200 and running Crysis at very high settings at 1920x1200 would get 37FPS ,plus it was a few frames higher in other games.But I decided to put my BFG GTX 280 OCX back in and see if I could tell that if it is any better.It maybe just my eyes but even 2D look better.Running 3DMark06 I would get 16,006 and in Crysis I would get 35FPS ,but it look so much more sharper and in Crysis my lows did'nt dip as low as it did with my two 9800GTx running in SLI.It may be a placebo effect but ever game I ran just looked so much better.So I will be sending my two ASUS EN9800GT ULTIMATE/HTDP/512M back to newegg.com,I got lucky and they are not going to charge me the 15% restocking fee.But I can't make up my mind if I want to get another GTX 280 or wait or buy something else.I would like to get a Q9550 but then I would have to get a new mother board.I thought about buying another Q6600 since I would more then likely get a GO and be able to overclock a little more.I have a B3 that I run all the time at 3240gigs at 1.37v.I just don't know if getting another GTX 280 running at 1920x1200 would be worth it without a faster CPU.

[Madpistol]

You have 2 9800 GTX's AND a GTX 280?!?!?

Overkill much? :headexplode:

To answer your question, a faster CPU would probably be a wise investment if you're thinking about going SLI GTX 280's.

[Lfctony]

What else is new mate? This has been reported by a lot of users with 9800GX2 cards after upgrading to a single 280. Average FPS never offer the real picture. SLIed cards always suffer lower minimums versus a slower on average single card that is faster than a single of the 2 SLIed cards... It has been said time and again, 2 cards does not equal double the frame rate. That's why I said in another post, that if you are faced with a choice between two solutions of the same speed, but one is a single card and the other is two cards together (even if the two cards are slightly faster), the single card will always offer a better experience, in terms of less problems and better gameplay...

[john19055]

Lfctony do you have two GTX 280's running in SLI on the Q6600 running at 3.2gigs and if you do how much does running two help.

[Lfctony]

Well, its less smooth than one card. There's severe stuttering from time to time. The motherboard does make a difference in smoothness etc. The CPU takes a hit from SLI, 200-300mhz I'd say for syncing the cards... Also, you only notice a difference where you are GPU limited. In CPU limited scenarios, your system is slower because of the extra burden on your CPU. If I had a choice to get 500e back and return one of these cards I would. I prefer a single card experience with lower details...

Say Crysis(Crysis scales quite bad, suffers from stuttering due to the load it places on the cards and your system in general) on High looks good and plays good.Add a second card and run on High, performance is actually lower (at 1680x1050 with one 280 the game is CPU limited).Take it up to Very High, the game looks better, plays better than with one 280, but still, it doesn't play as smooth or as fast as I would like. Warhead would dip to 19fps in DX9 from time to time, it would have been worse with DX10...

[aceflier]

Well, its less smooth than one card. There's severe stuttering from time to time. The motherboard does make a difference in smoothness etc. The CPU takes a hit from SLI, 200-300mhz I'd say for syncing the cards... Also, you only notice a difference where you are GPU limited. In CPU limited scenarios, your system is slower because of the extra burden on your CPU. If I had a choice to get 500e back and return one of these cards I would. I prefer a single card experience with lower details...

Say Crysis(Crysis scales quite bad, suffers from stuttering due to the load it places on the cards and your system in general) on High looks good and plays good.Add a second card and run on High, performance is actually lower (at 1680x1050 with one 280 the game is CPU limited).Take it up to Very High, the game looks better, plays better than with one 280, but still, it doesn't play as smooth or as fast as I would like. Warhead would dip to 19fps in DX9 from time to time, it would have been worse with DX10...

Say you are getting dips to 19fps in warhead at 1680x1050? Something is wrong with that man my 260's smoke warhead especialy in dx9 at 1920x1080 I never saw below 34fps in the final battle with the boss. Id say this is the most demanding spot in the game?

[Bman212121]

Actually I'd bet it's more due to your RAM limitations on the 9800GTX cards versus the GTX 280. In 3dmark it's not going to touch your ram, where Crysis turned up the way you have it will likely tank because of the 512MB vram.

Also, I don't think a CPU upgrade is going to get you very far. Unless it's to a E8600 or something you can get to 4.0Ghz, it will probably be a moot point.

Far Cry 2 has already shown it likes 4 cores, so for future games you're already sitting pretty well.

[Lfctony]

Say you are getting dips to 19fps in warhead at 1680x1050? Something is wrong with that man my 260's smoke warhead especialy in dx9 at 1920x1080 I never saw below 34fps in the final battle with the boss. Id say this is the most demanding spot in the game?

Well, my mate was playing yesterday (so my eyes were free), and during the game there was a lot of action at the beginning level and the frame rate would drop below 20fps at several points and stay there... It could have been a background program causing it though, don't remember what I had open at the time.

SLI is working fine, the cards are also fine as far as I can tell since they are working in all games as they should. My P Vantage score is above 20000. Warhead does utilise SLI somewhat, my average FPS are higher with two cards vs one. Can you run a couple of benches for me?

Download the Warhead Benchmark tool and run the Frost (plain) level and the Airfield Flythrough at Enthusiast DX9 1680x1050, let me know what it scores.

http://downloads.guru3d.com/Crysis-WARHEAD-Benchmark-Tool-BETA-download-2072.html

Thanks,
Tony.

[aceflier]

Well, my mate was playing yesterday (so my eyes were free), and during the game there was a lot of action at the beginning level and the frame rate would drop below 20fps at several points and stay there... It could have been a background program causing it though, don't remember what I had open at the time.

SLI is working fine, the cards are also fine as far as I can tell since they are working in all games as they should. My P Vantage score is above 20000. Warhead does utilise SLI somewhat, my average FPS are higher with two cards vs one. Can you run a couple of benches for me?

Download the Warhead Benchmark tool and run the Frost (plain) level and the Airfield Flythrough at Enthusiast DX9 1680x1050, let me know what it scores.

http://downloads.guru3d.com/Crysis-WARHEAD-Benchmark-Tool-BETA-download-2072.html

Thanks,
Tony.

ENTHUSIAST SETTINGS

Frost

!TimeDemo Run 0 Finished.
Play Time: 38.38s, Average FPS: 49.03
Min FPS: 15.77 at frame 62, Max FPS: 59.46 at frame 180
Average Tri/Sec: 28846766, Tri/Frame: 588349
Recorded/Played Tris ratio: -0.32
TimeDemo Play Ended, (1 Run Performed)

Airfield Flythrough

!TimeDemo Run 0 Finished.
Play Time: 39.93s, Average FPS: 50.09
Min FPS: 19.88 at frame 90, Max FPS: 81.37 at frame 1457
Average Tri/Sec: -40817188, Tri/Frame: -814832
Recorded/Played Tris ratio: 1.21
TimeDemo Play Ended, (1 Run Performed)

I would say the min. FPS is caused from loading all the textures into memory as its in the first 100frames each time more along the lines as 28-30fps lows.

Guess i should add i normaly run my shadows on gamer as they are the system killers.

[mailman2]

ONE CARD!!!!!

All dual card solutions, SLI, 9800GX2, 4870x2 have micro stutter which make good frames seem choppier, less smooth, etc. Stick with a single GPU, problem solved!!!

[aceflier]

ONE CARD!!!!!

All dual card solutions, SLI, 9800GX2, 4870x2 have micro stutter which make good frames seem choppier, less smooth, etc. Stick with a single GPU, problem solved!!!

Says you

I have not seen this myth on my system as of yet. :headexplode:

[Xion X2]

Says you

I have not seen this myth on my system as of yet. :headexplode:

Neither have I, but that won't keep the single GPU fanboys from shouting it from rooftops constantly.

[Lfctony]

ENTHUSIAST SETTINGS

Frost

!TimeDemo Run 0 Finished.
Play Time: 38.38s, Average FPS: 49.03
Min FPS: 15.77 at frame 62, Max FPS: 59.46 at frame 180
Average Tri/Sec: 28846766, Tri/Frame: 588349
Recorded/Played Tris ratio: -0.32
TimeDemo Play Ended, (1 Run Performed)

Airfield Flythrough

!TimeDemo Run 0 Finished.
Play Time: 39.93s, Average FPS: 50.09
Min FPS: 19.88 at frame 90, Max FPS: 81.37 at frame 1457
Average Tri/Sec: -40817188, Tri/Frame: -814832
Recorded/Played Tris ratio: 1.21
TimeDemo Play Ended, (1 Run Performed)

I would say the min. FPS is caused from loading all the textures into memory as its in the first 100frames each time more along the lines as 28-30fps lows.

Guess i should add i normaly run my shadows on gamer as they are the system killers.

Run the pass 3 times, meaning choose 3 in the runs... That will eliminate the loading of textures as it won't count the first run. And yes, the first run is always slower than the rest. :) No wonder those minimums are that low.

[Shocky]

Look at [H]'s latest Slalker performance review comparing the GTX280 and HD4870X2, the HD4870X2 got better benchmark results but the framerate wasn't consistent enough, the GTX280 even with a lower average framerate offered a more consistent experience. This is fairly typical when comparing multi gpu to single gpu's, has been since sli and crossfire arrived and still is, just some people are obsessed with having the best will defend their SLI and Crossfire solutions to their grave, problem is the only way we can directly compare is the framerate and when comparing a single gpu to a multigpu setup the average doesn't tell the whole story.

[Xion X2]

Look at [H]'s latest Slalker performance review comparing the GTX280 and HD4870X2, the HD4870X2 got better benchmark results but the framerate wasn't consistent enough, the GTX280 even with a lower average framerate offered a more consistent experience.

This notion that single GPU cards offer smoother overall framerates is hogwash. Stalker Clear Sky offers better performance (higher minimum framerate) on 4870 Crossfire as compared to a single 4870. At 1680x and maximum settings, a minimum of 22fps on Crossfire vs 13fps on a single 4870:

http://www.driverheaven.net/reviews.php?reviewid=642&pageid=7

How "consistent" something is doesn't really matter when you have drops to 13fps.

The key component is the minimum framerate. This determines how smoothly a game plays. In my own tests I've found a large difference in minimum framerates between single/multi-GPU setups as well. Look at the difference in the minimum framerate on Oblivion between a single GPU and Crossfire:

http://img296.imageshack.us/img296/9821/oblivion2pv0.jpg

Now which is going to be smoother? 28fps or 49fps?

And Mass Effect, a minimum of 31fps on a single GPU vs 54 in Crossfire:

http://img234.imageshack.us/img234/5986/masseffectus3.jpg

And Call of Juarez.. the minimum more than doubles going from 29 --> 64:

http://img296.imageshack.us/img296/6296/callofjuarezkr2.jpg

None of these tests above are run at high resolutions, either. At higher resolutions, this is magnified even further since multi-GPU shines at those settings and a single GPU tanks.

I just don't understand this mindset that you and some others have by claiming a single GPU setup offers a better overall gaming experience. It's nonsense. If you have a setup that is running games at a minimum of 64fps vs. a setup that's running them at 29fps, you're going to have a much smoother gaming experience in multi-GPU.

[ravinrobbie]

Most guys that complain about microstutter these days are those with low video memory solutions like the 9800GX2 or 9800GTX's in SLI running at resolutions of 1920X1200 or even higher.

Some people simply dont understand how the Vram is a limiting factor at very high resolutions and then blame all of the stuttering on SLI because they simply dont know any better. Fact is even a single 9800GTX would stutter at very high resolutions (maybe not as bad as it would with SLI) but it would still be there because of the low Vram.

For the most part cards like the 4870X2 with 1GB of usuable ram is pretty much flawless, I have seen it in action for myself and it was smooth as butter, I would also think that GTX 260's or 280's in SLI would also be smooth as well due to the high Vram.


All in all I have to say guys that I really think ATI has a one up on Nvidia with dual card solutions, this might change with the X58 board but for now I would not touch SLI with a 50 mile long pole due to the need for a NV chipset.

[ChrisRay]

A single GPU does offer a more consistent frame time pattern than a multi GPU at the "Same" framerates. If your getting 30 FPS on a single GPU and 30 FPS on a multi GPU. Then the single GPU static frame times will be better. This is easily documented via fraps. This is why comparing solutions like dual low end cards to high end cards like 9800GTX SLI/9800GX2 or HD 4850 to a single high end GPU is hogwash. Even if the GTX 280 were to performa a few percent slower. The experience would be more consistence.

I dont think anyone has ever said that If your getting 90 FPS on a multi GPU that your experience wouldn't be better than a single GPU at 50 FPS. Its when they are within a few percent of each other I cant fathom going the multi GPU choice given the option if you were buying "now".

[Lfctony]

Good post ChrisRay...

[Xion X2]

A single GPU does offer a more consistent frame time pattern than a multi GPU at the "Same" framerates.


I don't think anyone was arguing against this, either. Shocky didn't say the "same" framerates, though; he said at LOWER framerates, and it was he who I was replying to:

Look at [H]'s latest Slalker performance review comparing the GTX280 and HD4870X2, the HD4870X2 got better benchmark results but the framerate wasn't consistent enough, the GTX280 even with a lower average framerate offered a more consistent experience.

And that just isn't true most of the time.

We also should be careful to distinguish what we're talking about. There is a difference in what "frame time"/frame delay is and "framerate." Frame delay, as Chris and many others know, defines this "microstutter" or response variation that many claim exists on multi-gpu setups and doesn't on single-GPU. For multi-GPU, the frame time or frame delay is a very key component in how smooth your framerate appears on-screen.

And this is where I think our opinions differ a little from reading Chris Ray's last post. I don't believe, at least from an ATI perspective, that there is a huge gap in frame time / frame delay between a single/multi-gpu solution of this gen. For example, in looking at the consistency between single/multi-GPU when going the ATI route, a 4870X2 offers more consistency in frame response times than a 4870 does in both DMC4 and Call of Juarez at high settings. You can see both the 3870 and 4870 fluctuating wildly in comparison:

DMC4:
http://i103.photobucket.com/albums/m135/_DragonAvenger/frametimegeral.png

CoJ:
http://i103.photobucket.com/albums/m135/_DragonAvenger/graf-2.png

Notice especially the peak.. how much higher the variation is on the single-gpu solutions (3870 = black; 4870 = red; X2 = green):

http://img119.imageshack.us/img119/7465/66120608qa7.jpg (http://imageshack.us)


One possible argument against this would be something that I think Chris hinted at. My interpretation is that he's saying if you take a multi-GPU solution with two low-end GPUs that are performing around the same (within a few %) as a single GPU that the single GPU will have a lower frame response time, and that is probably true. I believe that a 280 would offer a better overall experience than 4850 Crossfire, for example--in the titles that scaled within a few percentage points of each other. But I think the gap has been lessened quite a bit this gen and is overexaggerated by just about everyone.

[Lfctony]

Damn Xion, the 3870x2 was a mess...

[Xion X2]

Damn Xion, the 3870x2 was a mess...

That it was. I think it was similar to the 9800GX2 in that it suffered from a lack of VRAM and PCI-e bandwidth.

You can see here that the total interconnect benefits from the PCI-E 2.0 and nearly doubles the bandwidth of prior gen--even when not factoring in the sideport, which is currently disabled (sorry, not trying to load this topic w/ ATI information, but it's where the majority of my recent experience is):

http://img157.imageshack.us/img157/1668/20080808e691c4a1f58ba4fmb2.jpg (http://imageshack.us)

Add to this extra bandwidth another 512MB of VRAM and you have a much better overall dual-GPU solution.

[Blacklash]

I've been experimenting with an X2 myself.

It's wonderful in AoC @ 1920x with full shadows, bloom, full quality ground radius, the rest set to high and 12xCFAA|16xAF. I haven't noticed any horrid micro stutter and my min frames are more than double over my overclocked GTX 260.

So far, I have yet to regret the purchase.

My GTX 260 did fine in the above title and certain scenarios dropped it down to 18FPS with 4xAA active @ 1920x. That annoyed me to no end and made the game sluggish in said scenarios.

One of the best price vs performance options going @ 1920x with AA is HD 4850 Crossfire for about 300usd. On a similar note, I feel that res and up with AA is best for the X2.

DX 9 Warhead results are below for the X2 and others from Hardware Canucks-

http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/10671-crysis-warhead-hardware-performance-review-5.html

[hell_of_doom227]

Well, its less smooth than one card. There's severe stuttering from time to time. The motherboard does make a difference in smoothness etc. The CPU takes a hit from SLI, 200-300mhz I'd say for syncing the cards... Also, you only notice a difference where you are GPU limited. In CPU limited scenarios, your system is slower because of the extra burden on your CPU. If I had a choice to get 500e back and return one of these cards I would. I prefer a single card experience with lower details...

Say Crysis(Crysis scales quite bad, suffers from stuttering due to the load it places on the cards and your system in general) on High looks good and plays good.Add a second card and run on High, performance is actually lower (at 1680x1050 with one 280 the game is CPU limited).Take it up to Very High, the game looks better, plays better than with one 280, but still, it doesn't play as smooth or as fast as I would like. Warhead would dip to 19fps in DX9 from time to time, it would have been worse with DX10...


I wonder what's wrong with your system. Crysis scales amazing and it's stuttering free. I would like to see wtf stuttering is anyway. I think this is all related to Monitor.

[Lfctony]

I wonder what's wrong with your system. Crysis scales amazing and it's stuttering free. I would like to see wtf stuttering is anyway.

Yes my system is broken. Yours works perfect...

[meddilion]

I would like to see wtf stuttering is anyway.



It happens every time you open your mouth.