Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
OpenMark results on PowerPC Macs
- Thread starter zen.state
- Start date
- Sort by reaction score
btw, did someone somewhen compare the two 12" ibooks and powerbooks with the highest features?
12" ibook G4 1,33GHz, ATI 9550, 32MB, 1,5GB RAM
12" Powerbook G4 1,5GHz, GF5200, 64MB, 1,25GB RAM
The 5200 is always said to be poorer than the 9550, but on the otherhand it has the bigger graphics Ram.
12" ibook G4 1,33GHz, ATI 9550, 32MB, 1,5GB RAM
12" Powerbook G4 1,5GHz, GF5200, 64MB, 1,25GB RAM
The 5200 is always said to be poorer than the 9550, but on the otherhand it has the bigger graphics Ram.
It keeps hanging for me after the 139K Triangles are finished....
All I can tell you is that I'm averaging about 1.1MTriangles/s, 1.2 when I overclock the 183 MHz core/mem, and 1.3 when I overclock to 200/200.
It gives me the same score of 165 no matter what the FPS rates are, though. This is only in the log... I never get to the screen with the score because it hangs before that.
Beige G3 w/1 GHz G4... and Original Radeon Mac Edition 32 MB DDR PCI. I've tried this with and without PCI Extreme enabled; crashes both ways in the same spot. Any thoughts?
I'll get around to benching the Quad G5 w/7800 GT sooner or later...
OpenMark started...
hardware detected:
cpu family: 0x10c
cpu type: 0x111
opengl vendor: ATI Technologies Inc.
opengl renderer: ATI Radeon OpenGL Engine
opengl version: 1.3 ATI-1.4.18
test parms:
surface 1280x1024 32bpp
textures: on
lighting: on
keep in sync: off
test started
FPS 478.0 0 triangles
FPS 309.0 1152 triangles
FPS 181.0 4608 triangles
FPS 99.0 10368 triangles
FPS 54.0 18432 triangles
FPS 39.0 28800 triangles
FPS 28.0 41472 triangles
FPS 21.0 56448 triangles
FPS 16.0 73728 triangles
FPS 13.0 93312 triangles
FPS 11.0 115200 triangles
FPS 9.0 139392 triangles
final score: 165
AGP 2x vs. 4x actually doesn't matter with GPUs this old; they don't come close to saturating even the 1 GBps throughput of AGP 2x. There was a definite improvement in performance when Macs turned over from PCI to AGP 2x; the Radeon 32 MB DDR performed often 10-15% faster when unencumbered by the poor bandwidth that the PCI offered (as well as AGP's other features; caching, special writes, etc), but 2-4x made little difference until you got into higher-end GPUs. Very little saturated AGP 4x during its lifetime, and nothing saturated 8x. Today's GPUs would, but we're onto PCIe.
I can't speak to the Mobile versions of those, I only really ever followed the DT versions, and I know at that point ATI especially was playing rather fast and loose with its namings of Mobile chips (i.e. a 9550 =/= necessarily equal a 95xx DT version...); but I can tell you two things.. #1, VRAM, like regular RAM, doesn't matter unless you run out of it.. and #2, the GeForce5200 was a horrific chip... actually, the whole GF5 series was a joke. The 9500 on the ATI side would have blown holes the size of platters in a 5200, although I can't tell you how the mobile counterparts would perform. They were probably 4 pixel pipes... and very possibly 64-bit memory interfaces.
All I can tell you is that I'm averaging about 1.1MTriangles/s, 1.2 when I overclock the 183 MHz core/mem, and 1.3 when I overclock to 200/200.
It gives me the same score of 165 no matter what the FPS rates are, though. This is only in the log... I never get to the screen with the score because it hangs before that.
Beige G3 w/1 GHz G4... and Original Radeon Mac Edition 32 MB DDR PCI. I've tried this with and without PCI Extreme enabled; crashes both ways in the same spot. Any thoughts?
I'll get around to benching the Quad G5 w/7800 GT sooner or later...
OpenMark started...
hardware detected:
cpu family: 0x10c
cpu type: 0x111
opengl vendor: ATI Technologies Inc.
opengl renderer: ATI Radeon OpenGL Engine
opengl version: 1.3 ATI-1.4.18
test parms:
surface 1280x1024 32bpp
textures: on
lighting: on
keep in sync: off
test started
FPS 478.0 0 triangles
FPS 309.0 1152 triangles
FPS 181.0 4608 triangles
FPS 99.0 10368 triangles
FPS 54.0 18432 triangles
FPS 39.0 28800 triangles
FPS 28.0 41472 triangles
FPS 21.0 56448 triangles
FPS 16.0 73728 triangles
FPS 13.0 93312 triangles
FPS 11.0 115200 triangles
FPS 9.0 139392 triangles
final score: 165
AGP 2x vs. 4x actually doesn't matter with GPUs this old; they don't come close to saturating even the 1 GBps throughput of AGP 2x. There was a definite improvement in performance when Macs turned over from PCI to AGP 2x; the Radeon 32 MB DDR performed often 10-15% faster when unencumbered by the poor bandwidth that the PCI offered (as well as AGP's other features; caching, special writes, etc), but 2-4x made little difference until you got into higher-end GPUs. Very little saturated AGP 4x during its lifetime, and nothing saturated 8x. Today's GPUs would, but we're onto PCIe.
Cox Orange said:
I can't speak to the Mobile versions of those, I only really ever followed the DT versions, and I know at that point ATI especially was playing rather fast and loose with its namings of Mobile chips (i.e. a 9550 =/= necessarily equal a 95xx DT version...); but I can tell you two things.. #1, VRAM, like regular RAM, doesn't matter unless you run out of it.. and #2, the GeForce5200 was a horrific chip... actually, the whole GF5 series was a joke. The 9500 on the ATI side would have blown holes the size of platters in a 5200, although I can't tell you how the mobile counterparts would perform. They were probably 4 pixel pipes... and very possibly 64-bit memory interfaces.
Last edited:
It doesn't quite work like this. If you run out of bandwidth on your interface, it doesn't matter what card you're running. You can have a Radeon 6990 HD tweaked and LN2 cooled to -50C overclocked to 8 GHz, but if you're running it on a bus interface that's too slow, you'll saturate the bus, and it won't make a lick of difference vs. any other card that saturates the interface.
Think of it this way... if you can only feed so much water through a tube per second, it doesn't matter how powerful the pump that feeds it is; that pipe can only sustain so much pressure. Superman could be doing the pumping and it wouldn't speed things up
OpenMark is useless for benchmarking since it seems to determine its score based on the description string of the graphics card.
Another note on 7800GT & X1900GT on Dual Core G5s: I benchmarked both gfx cards in my 2.0DC in Leopard (10.5.8) & Tiger (10.4.11). Both are flashed PC cards (the X1900GT being a former X1900XT). In UT2004, Botmatch scores only vary between max and min details and not between different resolutions. It's ~80fps at min details, regardless if res is 640x480 or 1600x1200. In Flyby, the 7800GT performs only marginally better than the X1900 (by about 5%). Fps vary more between high and low resolutions and detail levels, but not substantially. In ioquake3, the X1900 performs better, at 1600x1200x32 max details it scores 280fps. In 640x480x16 min details it's like 350fps.
So, it seems like the 970MP is too slow for those gfx cards (as seen by comparing cpu-intensive botmatch scores to graphics-intensive flyby scores). Knowing this, I can only recommend getting the X1900XT and flashing it to the X1900GT, since the only PCIe 7800GT with a factory 128K BIOS I know of is the one from Albatron. Any X1900XT will work with the 64K Mac ROM
Another note on 7800GT & X1900GT on Dual Core G5s: I benchmarked both gfx cards in my 2.0DC in Leopard (10.5.8) & Tiger (10.4.11). Both are flashed PC cards (the X1900GT being a former X1900XT). In UT2004, Botmatch scores only vary between max and min details and not between different resolutions. It's ~80fps at min details, regardless if res is 640x480 or 1600x1200. In Flyby, the 7800GT performs only marginally better than the X1900 (by about 5%). Fps vary more between high and low resolutions and detail levels, but not substantially. In ioquake3, the X1900 performs better, at 1600x1200x32 max details it scores 280fps. In 640x480x16 min details it's like 350fps.
So, it seems like the 970MP is too slow for those gfx cards (as seen by comparing cpu-intensive botmatch scores to graphics-intensive flyby scores). Knowing this, I can only recommend getting the X1900XT and flashing it to the X1900GT, since the only PCIe 7800GT with a factory 128K BIOS I know of is the one from Albatron. Any X1900XT will work with the 64K Mac ROM
I get an error saying "unable to choose right pixel format for device" when I try to run Openmark on my Digital Audio G4 with a Rage 128 Pro. Anybody know a way around this?
Maybe it is too low in performance to be shown with a score in the ranking of OpenMark (e.g. the base score OpenMarks sets is a better card and all other a placed accordingly). My iMac G3 DV 400MHz does refuse it, too and it has an ATI Rage 128.
@ thorns
Thanks for your input, now we have real evidence. I thought that this would be the case, because, even if the cards were extremely well made (which zen.state argued as the reason for all scores being the same), one should see at least a difference of 1 tiny point, when running the test on the equal cards.
Now, has anyone found an alternative test and has made several tests?
on the albatron 7800GT, I never heard of a limitation for flashing (I think one should have read it here and there, since quiet a few people do flassh cards), I thought virtually all 7800GT could be flashed for Mac (or aren't they 128k and so one will get a different one, though flashing will work?).
@ thorns
Thanks for your input, now we have real evidence. I thought that this would be the case, because, even if the cards were extremely well made (which zen.state argued as the reason for all scores being the same), one should see at least a difference of 1 tiny point, when running the test on the equal cards.
Now, has anyone found an alternative test and has made several tests?
on the albatron 7800GT, I never heard of a limitation for flashing (I think one should have read it here and there, since quiet a few people do flassh cards), I thought virtually all 7800GT could be flashed for Mac (or aren't they 128k and so one will get a different one, though flashing will work?).
No, virtually none of the PC 7800GT are flashable out-of-the-box, since they all carry 64K ROMs by default. Some Asus cards should also have a 128K ROM, but I managed to get my hands on one (the one with the green PCB) and it has a 64K ROM. You can replace the 64K ROM with a 128K ROM, but this is where it leaves the territory of hassle-free. The only cards which are guaranteed to be falshable are the quadro fx4500, the 7800gtx 512 and the radeon x1900xt.
Oh, come on! Why is that so bad for you! I like you and your experience and the information you give a lot; you are always there, when there is a specialized question on PPCs - but why can't you just for a moment question the open mark results.
"own delusion", what is my own delusion, when several people have done the test and there is another person that compares the power of the cards by testing them in games?
1. Tell me, where in the world is a piece of hardware made so identical, that all measurements will turn out the same? I know of two german manufacturers of microphones for music studios (Sennheiser and Beyerdynamics). They test the microphones and they have a range in which the results have to be. Beyerdynamics for example sorts out all pieces, that do not meet the tolerance range and sell them as a low cost series. They also do that to be able to sell matched pairs with nearly the same frequency diagram, but they won't be identical in the very meaning of the word, still! This was told to me by an engineer of Beyerdynamics.
2. If you run several graphics extensive tasks at the same time, while running OpenMark, should there not be a different result? or will it just block all other tasks?
And now, when you answer, please do not attempt to be mean to me, cause I am not to you and do not look down on me just because I do not have that much experience as you have. I am just questioning the results, because it is my nature to not finally exclude other possible answers.
My comment wasn't directed at you at all, Cox. You are one of the top 3-5 people IMO on this forum.
I base my trust in this benchmark because highly respected Mac organizations that do a lot benchmarks (ie. MacWorld, BareFeats and others) as well as Mac video driver developers do also. People who constantly run benchmarks wouldn't waste their time if they came to the same conclusions as some here.
Consistency in benchmarks is a very good thing and some act like thats bad. Results that are always changing cannot be actuate and certainly not trusted.
Cinebench is more a CPU/GPU combo benchmark and it gives the same results for the same GPU's on different Macs. It's easily one of the highest if not the most respected benchmark in the Mac industry.
I base my trust in this benchmark because highly respected Mac organizations that do a lot benchmarks (ie. MacWorld, BareFeats and others) as well as Mac video driver developers do also. People who constantly run benchmarks wouldn't waste their time if they came to the same conclusions as some here.
Consistency in benchmarks is a very good thing and some act like thats bad. Results that are always changing cannot be actuate and certainly not trusted.
Cinebench is more a CPU/GPU combo benchmark and it gives the same results for the same GPU's on different Macs. It's easily one of the highest if not the most respected benchmark in the Mac industry.
I do not disagree that consistency in benchmarks is a good thing. But I think that it's not very useful to benchmark a system's graphics performance with an application that only assesses the GPUs performance (btw, my X1900GT scored 16040 points in 1600x1200x32) when the 3D or gaming performance is a sum of all parts of the computer system (CPU speed, FSB speed, etc).
@ zen.state:
ok, then everything is fine. Thanks and excuse me for imputing you.
I saw that too, that many organisations (including japamacs, who's webpage about graphics card I like very much) use OpenMark a lot and I agree that would not, if the results were not reasonable or acurate to them.
I got try of an explanation, now: I do not know how GPU-benchmarks work and what they actually measure (other than FPS). If it measures FPS only, then the FPS should stay the same, if the graphical object displayed stays the same in every test, which is the case in this test. Maybe there we have the reason, why it gives the same scores over and over again.
If it measures throughput, I could imagine variability. If there is variability, the it would result in slightly different scores. (!) But wait! If the raw data of the test is being converted to a score number, then we could have the same results over and over again, as there may be the case, that a variability range of data gives 1 point. (Example: 1. time it gives a throughput of 100, next time it gives a throughput of 101. The score 1000 includes all results from 100-110, the score 1001 includes all results from 111-120. - no real world data taken for my example!)
ok, then everything is fine. Thanks and excuse me for imputing you.
I saw that too, that many organisations (including japamacs, who's webpage about graphics card I like very much) use OpenMark a lot and I agree that would not, if the results were not reasonable or acurate to them.
I got try of an explanation, now: I do not know how GPU-benchmarks work and what they actually measure (other than FPS). If it measures FPS only, then the FPS should stay the same, if the graphical object displayed stays the same in every test, which is the case in this test. Maybe there we have the reason, why it gives the same scores over and over again.
If it measures throughput, I could imagine variability. If there is variability, the it would result in slightly different scores. (!) But wait! If the raw data of the test is being converted to a score number, then we could have the same results over and over again, as there may be the case, that a variability range of data gives 1 point. (Example: 1. time it gives a throughput of 100, next time it gives a throughput of 101. The score 1000 includes all results from 100-110, the score 1001 includes all results from 111-120. - no real world data taken for my example!)
Sigh
I'm super bummed my G5 2.5 died - been trying to resurrect it but the 7800GS I have in that box made for one zippy machine.
Some amazing scores in this thread!
I'm super bummed my G5 2.5 died - been trying to resurrect it but the 7800GS I have in that box made for one zippy machine.
Some amazing scores in this thread!
You've got to go to the OpenMark menu and select "Run Benchmark" (or hit Command + R)