Hello all. I'm in the process of developing a new TG2 benchmark to be featured on a benchmarking site I am helping to develop. In order to make the benchmark relevant and accurate for the largest number of machines possible, I'd like to test the absolute fastest machines available today to make sure they don't render the scene too fast. Thus I need someone with TG2 running on a Core i7 940, or better yet Extreme 965, overclocked better still. PM or email me "tgbench AT oshyan.com" if you want to volunteer. The tests shouldn't take long.
Thanks!
- Oshyan
Quote from: Oshyan on February 15, 2009, 07:13:39 PM
I'd like to test the absolute fastest machines available today to make sure they don't render the scene too fast.
;D
Quote from: Oshyan on February 15, 2009, 07:13:39 PM
to make sure they don't render the scene too fast.
Perish the thought.
John
I can help with Q6600 @ 3.2 GHz.
It ain't speed daemon right now but if you would like to build a database of some kind or compare Kentsfield with Nehalem - I'm good to go.
Q9450 @ 2.67 Ghz here, wont be getting an i7 till July (which will be OC'd and water cooled Mmmm) but if I can help let me know... :)
ruichard
I'm only using the i7 920, but if you don't find users with overclocked 940 / 965EE, I can test it with ~3GHz.
Thanks for the replies guys. I've got several quad Core 2's of my own here to test on. What I really need is a top of the line Core i7. But, failing that, a 920 (overclocked even better) would be good, at least. So if you're ready to try, go ahead and contact me privately via email Kranky. I'll arrange to get you the benchmark files and away we go. :)
- Oshyan
I think I'm gonna cry :-(
I have access to a Core i7 @ 3.6GHz if you still need someone. It's only on Windows XP32 right now though. I'm trying to install Mac OS 10.6 on it though.
The benchmark is published now so go check it out! I'd love to have your result. http://www.3dspeedmachine.com/ :)
- Oshyan
Quote from: Oshyan on March 06, 2009, 01:02:03 AM
The benchmark is published now so go check it out! I'd love to have your result. http://www.3dspeedmachine.com/ :)
- Oshyan
Done. I put the results of a few machines on there. I got the highest score at 3.6GHz but I think I could do better with a CPU cooler and more RAM.
Awesome, thanks! Pretty impressive, although surprising it's not a whole lot faster than Frank's result. Here's something to try. Run the benchmark again, except this time before you do the actual render, change the Minimum Threads setting to 8 and see what happens. Apparently Core i7's "hyperthreading" implementation is actually quite good and gives a notable speed increase, whereas with older hyperthreading (Netburst, P4 architecture) the "virtual cores" were comparatively slow and so by default TG2 doesn't recognize and try to take advantage of them. We may have to change that strategy for Core i7's.
Also if you can properly indicate the speeds of the systems you entered, instead of using the default clock speed for your CPU, that would be great. You should have a *separate* system entry for each clock speed for overclocking results.
By the way, we now have more Terragen 2 benchmarks than any other application on the site. Go Terragen community! ;D
- Oshyan
I changed the number of cores override in the preferences to 8 to get that render time. I tried putting the actual clock speed, but the benchmark site reset it back down to 2.66GHz. I had it running at 3.6GHz. I'm sure I could get it higher with a better CPU cooler. People have gotten them up to 4.5GHz on air. I just have a stock cooler right now. I also need to get more RAM, but I'm saving for a Mac Pro and she doesn't really want to spend more money on computer stuff right now.
Check out the speed of the new Mac Pros @ 2.93GHz compared to the Core i7 I was using a@ 3.6GHz.
The Core i7: http://browse.geekbench.ca/geekbench2/view/114772
Mac Pro: http://browse.geekbench.ca/geekbench2/view/115605
Strange. What was your result with the default minimum of 1 thread? I'm just surprised your system is so much slower than Frank's. It's possible that Vista or x64 provides some benefit I suppose, but that seems pretty significant. In your posted results, are both at 3.6Ghz overclocked, or is your XP SP3 result at default 2.6Ghz?
I'll look into the issue with entering the clock speed correctly.
- Oshyan
The one with XP I think was at the default clock and the default 4 cores detected. I suspect Frank had his set to render on all 8 core and/or was overclocked too in order to get that time. The XP I used was 32-bit and the Windows 7 was 64-bit. The machine however only has 3GB of RAM, so I don't know if the 64-bitness helped any. It installed Terragen into the 32-bit Programs folder.
Ironically, I also got the two lowest scores on the benchmark list too. Go 500MHz G3 iMac!
Any chance you could compile the Mac OS X version of Terragen with the architecture set to x86_64? Right now it's only running in 32-bit mode according to activity monitor. It's sad having one of the most powerful applications on my computer still being 32-bit in relation to almost all the processes in Mac OS 10.6 being 64-bit. I'd personally love to compile Terragen myself so I could set the cflags to something like this for my laptop: -arch x86_64 -march=core2 -mtune =core2 -msse4.1.
Well, considering that Terragen isn't an open source project, you probably won't get to compile it yourself.
As for 64-bit, the Terragen interface is based off of the Carbon API, which Apple never released 64-bit versions for. Until Planetside can migrate away from Carbon towards Cocoa, they won't be able to make a 64-bit version of TG2 for the Mac.
I don't think simply compiling TG2 with 64 bit flags will really change anything. We need to make fundamental architectural changes to take proper advantage of a 64 bit environment. We plan to do that after the coming release.
As for Frank's benchmark, I'm pretty certain it was not overclocked. When you say "The one with XP I think was at the default clock and the default 4 cores detected." it's a bit worrying. It's important that the results there be verifiable and clear. So I would ask that you re-run the results and more carefully note the environment details, if possible. We're working on allowing you to easily specify that a machine is overlocked. I would also ask that you run the benchmark with default settings and use that for your on-site benchmark report. Alternate configurations can be discussed here until we have a better system for noting special environment details.
Thanks!
- Oshyan
I will test it with my AMD Phenom2 x4 940 @3.2 ghz. If it is possible to render it with a preview version.
Quote from: Oshyan on March 11, 2009, 12:05:16 AM
I don't think simply compiling TG2 with 64 bit flags will really change anything. We need to make fundamental architectural changes to take proper advantage of a 64 bit environment. We plan to do that after the coming release.
As for Frank's benchmark, I'm pretty certain it was not overclocked. When you say "The one with XP I think was at the default clock and the default 4 cores detected." it's a bit worrying. It's important that the results there be verifiable and clear. So I would ask that you re-run the results and more carefully note the environment details, if possible. We're working on allowing you to easily specify that a machine is overlocked. I would also ask that you run the benchmark with default settings and use that for your on-site benchmark report. Alternate configurations can be discussed here until we have a better system for noting special environment details.
Thanks!
- Oshyan
The benchmark running XP was definitely on 4 cores. I just don't remember what it was clocked to exactly. I think it was 3.2GHz. I was having memory issues on XP32 running all 16 cores.
Quote from: echrei on March 11, 2009, 03:54:15 PM
Quote from: Oshyan on March 11, 2009, 12:05:16 AM
I don't think simply compiling TG2 with 64 bit flags will really change anything. We need to make fundamental architectural changes to take proper advantage of a 64 bit environment. We plan to do that after the coming release.
As for Frank's benchmark, I'm pretty certain it was not overclocked. When you say "The one with XP I think was at the default clock and the default 4 cores detected." it's a bit worrying. It's important that the results there be verifiable and clear. So I would ask that you re-run the results and more carefully note the environment details, if possible. We're working on allowing you to easily specify that a machine is overlocked. I would also ask that you run the benchmark with default settings and use that for your on-site benchmark report. Alternate configurations can be discussed here until we have a better system for noting special environment details.
Thanks!
- Oshyan
The benchmark running XP was definitely on 4 cores. I just don't remember what it was clocked to exactly. I think it was 3.2GHz. I was having memory issues on XP32 running all 16 cores.
Ok, it's important to have the specifics though. Please re-test when you get a chance and remove any results you're not certain of for now.
Thanks!
- Oshyan