Another round of benchmarks
Moderator: Graf Zahl
- Graf Zahl
- GZDoom Developer
- Posts: 7148
- Joined: Wed Jul 20, 2005 9:48
- Location: Germany
- Contact:
Another round of benchmarks
I am starting another round of benchmarks. My main interest is the performance development of the current drivers and the performance of the new Geforce GTX400 series.
http://grafzahl.drdteam.org/other/gz_benchmarks.zip
Instructions:
- Download GZDoom 1.5.3 and extract it into an empty directory.
- Start GZDoom and set it up to your preferences. Then set the following GL options:
* shader options: all on (if available in the menu)
* fog mode: Radial
* Light mode: Doom
* Vertical sync: off
* Rendering quality: Quality
- Download the savegames and extract them into the same folder.
- Bind a key to the 'bench' console command to a key. You can bind another key to 'exit' so that you can exit quicker.
- download all mods to be tested and copy them into the same folder. Here's the links:
Hellcore 2.0
Phobos: Anomaly Reborn
Knee-Deep in ZDoom
Phobia: The Age
ZPack
Ultimate TNT
The Darkest Hour
Super Sonic Doom
BGPA Missions Liberation
Operation Overlord
EPIC
Cheogsh
Sunder
- when everything is set up, start 'bench.bat'. This will load all test savegames sequentially. In each level press the key you bound to 'bench' and wait until a message appears in the top left corner. After that quit, wait for the next level to load and repeat. Please make sure that the console is closed at all times! If you call the 'bench' command from the console the results are useless.
Post the numbers here, along with the output you get from starting 'gzdoom -logfile log'.
Note: This test only works on Windows because the low level timers are not implemented for the Linux build.
http://grafzahl.drdteam.org/other/gz_benchmarks.zip
Instructions:
- Download GZDoom 1.5.3 and extract it into an empty directory.
- Start GZDoom and set it up to your preferences. Then set the following GL options:
* shader options: all on (if available in the menu)
* fog mode: Radial
* Light mode: Doom
* Vertical sync: off
* Rendering quality: Quality
- Download the savegames and extract them into the same folder.
- Bind a key to the 'bench' console command to a key. You can bind another key to 'exit' so that you can exit quicker.
- download all mods to be tested and copy them into the same folder. Here's the links:
Hellcore 2.0
Phobos: Anomaly Reborn
Knee-Deep in ZDoom
Phobia: The Age
ZPack
Ultimate TNT
The Darkest Hour
Super Sonic Doom
BGPA Missions Liberation
Operation Overlord
EPIC
Cheogsh
Sunder
- when everything is set up, start 'bench.bat'. This will load all test savegames sequentially. In each level press the key you bound to 'bench' and wait until a message appears in the top left corner. After that quit, wait for the next level to load and repeat. Please make sure that the console is closed at all times! If you call the 'bench' command from the console the results are useless.
Post the numbers here, along with the output you get from starting 'gzdoom -logfile log'.
Note: This test only works on Windows because the low level timers are not implemented for the Linux build.
Last edited by Graf Zahl on Mon Aug 12, 2013 8:40, edited 1 time in total.
-
- Posts: 15
- Joined: Fri Jan 01, 2010 18:00
Re: Another round of benchmarks
Burghead wouldn't run.
Core i7 930 and Radeon 5870 using Cat 10.9
1920x1080 16xAF, Cat AI Off
2x is 2x multisample AA
8x is 8x supersample AA
Core i7 930 and Radeon 5870 using Cat 10.9
1920x1080 16xAF, Cat AI Off
2x is 2x multisample AA
8x is 8x supersample AA
Last edited by DaMan on Sun Sep 26, 2010 13:28, edited 1 time in total.
- Graf Zahl
- GZDoom Developer
- Posts: 7148
- Joined: Wed Jul 20, 2005 9:48
- Location: Germany
- Contact:
Re: Another round of benchmarks
Ok, since you used HQ4x I'll better add this requirement: No hires texture filters, please. They tend to distort the results.
What was your problem with Burghead?
What was your problem with Burghead?
- Gez
- Developer
- Posts: 1399
- Joined: Mon Oct 22, 2007 16:47
Re: Another round of benchmarks
Might be a good idea to provide a custom benchmark config as well. That way, even resolution wouldn't be a factor, you could go with a baseline (like 800x600 windowed) which should work on any computer.
- Enjay
- Developer
- Posts: 4720
- Joined: Tue Aug 30, 2005 23:19
- Location: Scotland
- Contact:
Re: Another round of benchmarks
I haven't had a chance to do the benchmarks yet (not sure when I will manage) but Burghead certainly runs for me.
It does spam a load of error messages to the console on startup These are genuine errors in as much as I just included my doomdefs lump as it was an it contains a bunch of references to actors that are not in the game. When it was released, these were still errors, of course, but GZdoom did not complain about them so I had no quick way of weeding them all out, so I didn't. All that being said though, GZdoom handles them correctly: it reports them and then just carries on loading and playing as it should.
I note that the main thrust of these benchmarks are to get info for the Geforce GTX400 series. I have a GTX285. Is my information still of interest?
It does spam a load of error messages to the console on startup These are genuine errors in as much as I just included my doomdefs lump as it was an it contains a bunch of references to actors that are not in the game. When it was released, these were still errors, of course, but GZdoom did not complain about them so I had no quick way of weeding them all out, so I didn't. All that being said though, GZdoom handles them correctly: it reports them and then just carries on loading and playing as it should.
I note that the main thrust of these benchmarks are to get info for the Geforce GTX400 series. I have a GTX285. Is my information still of interest?
- Graf Zahl
- GZDoom Developer
- Posts: 7148
- Joined: Wed Jul 20, 2005 9:48
- Location: Germany
- Contact:
Re: Another round of benchmarks
Yes, of course. Without any reference the GTX400 values are not worth much. The last round of benchmarks was done before some major optimizations, after all. The more data I have to compare the better.
That's pointless. At that screen size all that gets measured is the CPU. Even on my old Geforce 8600 at that screen size it's clearly CPU bound - even on large and complex levels. I'd rather have some high-resolution values, too.Gez wrote:(like 800x600 windowed)
- NeoHippo
- Posts: 402
- Joined: Tue Sep 13, 2005 0:47
- Location: British Columbia Canada
Re: Another round of benchmarks
old post deleted
see new post below
see new post below
Last edited by NeoHippo on Sat Sep 25, 2010 23:21, edited 1 time in total.
- Graf Zahl
- GZDoom Developer
- Posts: 7148
- Joined: Wed Jul 20, 2005 9:48
- Location: Germany
- Contact:
Re: Another round of benchmarks
Concerning the load screen, it only shows the saves for Doom2 mods. Which is normal as it omits anything using other IWADs. And apparently you don't have Doom1. All Doom1 using benchmarks failed.
Also, what's your specs? You don't mention anything.
Also, what's your specs? You don't mention anything.
- NeoHippo
- Posts: 402
- Joined: Tue Sep 13, 2005 0:47
- Location: British Columbia Canada
Re: Another round of benchmarks
OK, I reran the benchmark, this time on two different machines.
See attached.
The results are quite interesting.
It appears, that upgrading to a GTX400 series video card is not worth the effort.
At least not as far as GZDoom is concerned.
See attached.
The results are quite interesting.
It appears, that upgrading to a GTX400 series video card is not worth the effort.
At least not as far as GZDoom is concerned.
- Attachments
-
- GTX 460_benchmarks_Sat-09-25-2010_0312.txt
- (19.15 KiB) Downloaded 182 times
-
- GT240_benchmarks_Sat-09-25-2010_0238.txt
- (19.33 KiB) Downloaded 153 times
- Graf Zahl
- GZDoom Developer
- Posts: 7148
- Joined: Wed Jul 20, 2005 9:48
- Location: Germany
- Contact:
Re: Another round of benchmarks
What I find more interesting is that the AMD Phenom looks to be quite inferior even to older Intel CPUs. I wonder how much that influences the results.
But aside from that, yes, it seems that these 2 cards are almost an even match concerning GZDoom - most likely because the engine can't send the data fast enough. This looks like it's completely CPU-bound on both.
But aside from that, yes, it seems that these 2 cards are almost an even match concerning GZDoom - most likely because the engine can't send the data fast enough. This looks like it's completely CPU-bound on both.
- TIHan
- Posts: 4
- Joined: Fri Sep 24, 2010 20:02
Re: Another round of benchmarks
If you look at some of the benchmarks of the Phenom II, they are right up there with the core 2 quads(extremes as well).
http://www.tomshardware.com/reviews/phe ... 14-18.html
http://www.tomshardware.com/reviews/phe ... 14-18.html
- Graf Zahl
- GZDoom Developer
- Posts: 7148
- Joined: Wed Jul 20, 2005 9:48
- Location: Germany
- Contact:
Re: Another round of benchmarks
Well, it's definitely slower than the one NeoHippo had in his GT240 system.
Anyway, I can see one area where the GTX460 is definitely lagging behind - and that's when a lot of dynamic lights have to be processed. It seems it has problems uploading large amounts of data to the GPU - something other people already noticed on the OpenGL forums. I guess we'll have to wait and see if it's a driver issue or a hardware problem. I hope it's the former.
Anyway, I can see one area where the GTX460 is definitely lagging behind - and that's when a lot of dynamic lights have to be processed. It seems it has problems uploading large amounts of data to the GPU - something other people already noticed on the OpenGL forums. I guess we'll have to wait and see if it's a driver issue or a hardware problem. I hope it's the former.
-
- Posts: 15
- Joined: Fri Jan 01, 2010 18:00
Re: Another round of benchmarks
Burghead says unknow object code (0) in archive when loading savegame. Newgame works fine.
Reran the benchmarks at both 3.6GHZ(3GHZ uncore) and 2.8GHZ(2.6GHZ Uncore) if you want to check CPU limited ness.
Sonic Doom is the only GPU limited bench. Its the only that that got much slower increasing AA on mine and the only that got faster on NeoHippo's 460 system. But I still got a 17-32% when overclocking like the rest so it still wants CPU.
Reran the benchmarks at both 3.6GHZ(3GHZ uncore) and 2.8GHZ(2.6GHZ Uncore) if you want to check CPU limited ness.
Sonic Doom is the only GPU limited bench. Its the only that that got much slower increasing AA on mine and the only that got faster on NeoHippo's 460 system. But I still got a 17-32% when overclocking like the rest so it still wants CPU.
- Graf Zahl
- GZDoom Developer
- Posts: 7148
- Joined: Wed Jul 20, 2005 9:48
- Location: Germany
- Contact:
Re: Another round of benchmarks
Sonic does not surprise me. That one's using skyboxes in a very weird way that requires lots ans lots of stencil buffers and occlusion queries. That will inevitably cause a GPU bottleneck. It's precisely why I added these levels to the list. On my GF 8600 they run all below 30 fps, the second one below 10, btw, because they completely stall the GPU.
- BouncyTEM
- Posts: 4
- Joined: Sun Sep 26, 2010 22:02
Re: Another round of benchmarks
Here's mine. i7 950 + 2 Nvidia GTX 295s on 258.96 drivers.
- Attachments
-
- benchmarks_Sun-09-26-2010_0459.txt
- also on 1920x1200.
- (12.29 KiB) Downloaded 134 times