Some notes on 3D performance on the new iMacs

[psonice]

I've played around with my new imac for a while now, and have come across some issues with 3d performance, so thought I'd share what I've found - it may explain the poor performance seen at times. I've been using mainly windows for 3d, but I've seen exactly the same thing in osx, so I guess the drivers are based on the same code for both.

First - performance overall is very good. When people say the radeon 2600 is a slow card - this is only true when you compare it to the really high end cards. It's actually quite fast, and should handle the latest games at reasonable speed + quality (I'm not a gamer though, so I've not tried it). The 3d stuff I use has all run better than I expected.

Now - the catch. I've noticed that I can run things at full resolution (1920x1200) with 4x AA, and it runs fine - until new textures are loaded, and then it stutters badly for a second or so. I reckon this would have a pretty big impact on frame rate figures, bench marks etc. where it happens often.

I asked a friend (who wrote one of the bits of software in question), and his reply was "Yes, the ATI 2xxx series drivers break the cache warming". What happens is this - when a new texture is needed, it's prepared and sent to the card so it's there when it's needed. If that isn't done, then the nice smooth running 3d grinds to a halt while the card waits for the texture to be loaded. Exactly what I've seen.

The good news is that this is a driver issue, and should be fixed soon with a bit of luck - expect a good performance boost when it is. In the meantime, I think there are work-arounds, so the companies making games etc. could provide a patch (I saw that there was one for bioshock - perhaps this is what it fixes? Anyone care to try it before applying the patch, see if there's bad stuttering with smooth patches between?)

 

[AlexisV]

It might even have been fixed already in the last driver update, which is unavailable to users on Bootcamp.

I expect Leopard will have very much improved drivers for both OS X, XP and Vista. Bootcamp is still only beta until then after all.

 

[MIDI_EVIL]

That's really great information thanks.

I'm sure a lot of us guessed it was drivers, but it's good to have some more confirmation.

 

[DMHB]

Thanks alot for information!

 

[psonice]

It might even have been fixed already in the last driver update, which is unavailable to users on Bootcamp.

I expect Leopard will have very much improved drivers for both OS X, XP and Vista. Bootcamp is still only beta until then after all.

I'm using the bootcamp 1.4 drivers, no updates were available after that unless there are new ones now. As far as I'm aware, ATI haven't fixed it for windows yet though, and I'd guess that they'll update the standard windows drivers before they get around to the apple bootcamp ones. There is a bootcamp drivers section on ATI's driver download page by the way - it's worth keeping an eye on that, as I think new drivers will appear there before apple updates bootcamp. Also, the mobility radeon 2600 series drivers might work, but ATI have none on their site as yet.

 

[RichardI]

Now - the catch. I've noticed that I can run things at full resolution (1920x1200) with 4x AA, and it runs fine - until new textures are loaded, and then it stutters badly for a second or so. I reckon this would have a pretty big impact on frame rate figures, bench marks etc. where it happens often.

I am playing a rather simple little 3D adventure game called AGON and I have seen this issue. I, too had assumed it was a driver issue. I've been checking Software Update every day . Looking forward to new ones.

Rich

 

[RichardI]

Now - the catch. I've noticed that I can run things at full resolution (1920x1200) with 4x AA, and it runs fine - until new textures are loaded, and then it stutters badly for a second or so. I reckon this would have a pretty big impact on frame rate figures, bench marks etc. where it happens often.

I am playing a rather simple little 3D adventure game called AGON and I have seen this issue. I, too had assumed it was a driver issue. I've been checking Software Update every day . Looking forward to new ones.

[edit] Very bad form on the double post. My apologies gang.

Rich

 

[PAC2007]

I ran the below tests on a new iMac 2.8GHz with 2GB Ram...under Vista 64bit.

Steam Counter Strike Stress Test...

Now these are with High settings, which the system recommended...That is also No AA and uses Trillinear

I ran the tests three times:

1024x768

1st: 186
2nd: 195
3rd: 196

1600x1200

1st: 108
2nd: 108
3rd: 107

Downloaded and installed 3DMark 03, 05 and 06 and also PCMark05...

Now, since I'm using Vista I decided to run these programs in XP SP2 mode under the properties...

3DMark 03: 10,834

3DMark 05: Crashes (Tried reinstalling etc)

3DMark 06: 3820

and Finally PCMark 05: 6868

Hope this helps?

 

[psonice]

I ran the below tests on a new iMac 2.8GHz with 2GB Ram...under Vista 64bit.

...

Hope this helps?

Did you notice a stuttering effect? I.e. you get roughly constant speed/fps, then it jerks a lot for a short time, then goes smooth again? Usually as something new appears on the screen, but not always. I'd like to know how much effect that has on some of the scores, but I guess we can't test that until it's fixed.

 

[Pressure]

It could also be part of the memory sub-system acting up.

Say, whenever something new comes on the screen it fetches textures etc. from the slow hard drive, thereby causing the dreaded jerkiness.

 

[psonice]

Yes, memory could cause the same effect, but in this case I can rule it:

It won't run out of system RAM, as I have 4 gig, and know for a fact that the bit of software I first noticed this on (and several others) run flawlessly with 2 gig.

It could also be video ram, which could run low if you use lots of textures and have high resolution. I can rule that out because on one program it runs fine on a pc with a 128mb card, but stutters on the imac with 256. Also, it happens pretty much exactly the same at any resolution.

 

[Pressure]

Yes, memory could cause the same effect, but in this case I can rule it:

It won't run out of system RAM, as I have 4 gig, and know for a fact that the bit of software I first noticed this on (and several others) run flawlessly with 2 gig.

It could also be video ram, which could run low if you use lots of textures and have high resolution. I can rule that out because on one program it runs fine on a pc with a 128mb card, but stutters on the imac with 256. Also, it happens pretty much exactly the same at any resolution.

Have you tried running the same game on a RAID 0 setup and see if the jerkiness persists? I usually got stutter in a MMORPG I used to play when new characters were getting on the screen from the drawing distance, as the hard drive were loading up the textures from itself to the memory. A RAID setup took care of it.

 

[psonice]

Have you tried running the same game on a RAID 0 setup and see if the jerkiness persists? I usually got stutter in a MMORPG I used to play when new characters were getting on the screen from the drawing distance, as the hard drive were loading up the textures from itself to the memory. A RAID setup took care of it.

It's not games I've been using to test - infact I haven't got any games at all. The bits I've tested are all "demoscene" productions (that most of you have probably never heard of, look it up They're all non-interactive, and generally tend to be optimised a lot to ensure that they run smoothly. They also normally run from RAM so the hard disk would make no difference.

If you want to compare, try these (Windows only unfortunately, i'll try and find some mac ones later but there are fewer of them and they're not generally as good):

http://www.pouet.net/prod.php?which=31571 (this should run fine on a 3 year old machine with average specs, it runs perfectly on my work PC which is a FAR lower spec than the imac)

http://www.pouet.net/prod.php?which=30244 (this is more demanding, but it shows the cache issue perfectly - it runs smoothly, then stutters a lot when the scene changes) (Note that this one is only 170kb too, for 7 minutes odd of audio and video =) )

Download links for both are at the bottom right of box at the top with the screenshot in.

 

[RRK]

Makes sense. This series of processors had high expectations and are just not living up to their potential.

 

[Pressure]

It's not games I've been using to test - infact I haven't got any games at all. The bits I've tested are all "demoscene" productions (that most of you have probably never heard of, look it up They're all non-interactive, and generally tend to be optimised a lot to ensure that they run smoothly. They also normally run from RAM so the hard disk would make no difference.

If you want to compare, try these (Windows only unfortunately, i'll try and find some mac ones later but there are fewer of them and they're not generally as good):

http://www.pouet.net/prod.php?which=31571 (this should run fine on a 3 year old machine with average specs, it runs perfectly on my work PC which is a FAR lower spec than the imac)

http://www.pouet.net/prod.php?which=30244 (this is more demanding, but it shows the cache issue perfectly - it runs smoothly, then stutters a lot when the scene changes) (Note that this one is only 170kb too, for 7 minutes odd of audio and video =) )

Download links for both are at the bottom right of box at the top with the screenshot in.

Tsk, I was at the demo party where Farbrausch released their famous 64k demo back in the days (called "The Party")

Could be a driver thing going on but I could test them on my old MacBook Pro with an Mobility Radeon X1600.

 

[psonice]

Tsk, I was at the demo party where Farbrausch released their famous 64k demo back in the days (called "The Party")

Could be a driver thing going on but I could test them on my old MacBook Pro with an Mobility Radeon X1600.

Nice, which one? The Product? FR did a few 'famous' 64ks, but that's the most famous I guess. Oh yes.. it was released at the party, so it'll be that one. Must have been quite an experience! I've only been to the one party, Assembly this summer. It was a good year for releases, although the 64ks weren't so hot.

Yes, I think it's a driver issue, that's what my coder friend told me and everything seems to point to exactly that issue. Those demos should run fine by the way on an x1600, and should run pretty smoothly.

 

[Tom Sawyer]

I ran the below tests on a new iMac 2.8GHz with 2GB Ram...under Vista 64bit.

Steam Counter Strike Stress Test...

Now these are with High settings, which the system recommended...That is also No AA and uses Trillinear

I ran the tests three times:

1024x768

1st: 186
2nd: 195
3rd: 196

1600x1200

1st: 108
2nd: 108
3rd: 107

Downloaded and installed 3DMark 03, 05 and 06 and also PCMark05...

Now, since I'm using Vista I decided to run these programs in XP SP2 mode under the properties...

3DMark 03: 10,834

3DMark 05: Crashes (Tried reinstalling etc)

3DMark 06: 3820

and Finally PCMark 05: 6868

Hope this helps?

Thanks for sharing the numbers... I tried to install 3dMark 05 and it crashed on me as well... I was worried it was something with my machine but apparently not. I wonder what it could be doing that would cause it to hang/crash like that? Anyway... fairly decent numbers though.

 

[AlexisV]

3D Mark 06 has a known bug in it regarding the new Radeon GPUs (I had to rename a dll file to get it to run), so I'm not surprised 05 falls over.

 

[RRK]

It didn't get any attention in the 2600xt thread so I will ask again here.

So, could the new linux drivers from AMD mean something to us?

http://www.phoronix.com/scan.php?page=article&item=821&num=1

Some interesting quotes:
"How much faster? In many cases it is about 50% faster while in some configurations it may go as high as 90% or more. In fact, in some benchmarks the Mobility Radeon X300 was over 10x faster!"

"The fglrx 8.41 driver is focused primarily for the R600 series and the 8.42 driver will contain additional fixes and optimizations for the older generation of ATI graphics processors."