View Issue Details
| ID | Project | Category | View Status | Date Submitted | Last Update |
|---|---|---|---|---|---|
| 0008741 | Dwarf Fortress | Technical -- General | public | 2015-01-15 03:56 | 2015-01-16 07:27 |
| Reporter | vomov | Assigned To | |||
| Priority | low | Severity | minor | Reproducibility | always |
| Status | resolved | Resolution | no change required | ||
| Platform | PC | OS | Windows | OS Version | 7, 8.1 |
| Product Version | 0.40.24 | ||||
| Summary | 0008741: DF runs quite slow on a powerfull AMD, and blazingly fast on a crappy Intel | ||||
| Description | I wasn't sure this could be classified as a bug, but it is something odd that needs to be written down somewhere, at least for reference. Since I'm frequently waiting for Matlab to complete operations for the next few weeks, I've been running DF on my work PC (from my dropbox). The problem is, that it runs awesomely fast. For comparison: PC 1 (home PC): x64 windows 8.1 AMD A8-6500 APU @ 3.8GHz, quad core (and a Radeon 5800 as the videocard, I think) 8GB RAM Running DF of an OCZ Agility FPS: 30 PC 2 (work PC): windows 7 Intel Pentium 4 @ 2.4GHz, single core 2GB RAM Running DF of an generic hard drive Note: also running matlab, which is happily churning away at some large datasets. DF while waiting. FPS: 110 I've found earlier threads about the AMD/ATI openGL options, and fiddled with the drawing options on both sides. The FPS on my home PC is the highest I've gotten, using 'VBO' (+2fps), and the 110 is the minimum FPS I've gotten on the work PC, using 'standard'. In general, I don't think these options make that much difference. So, this issue seems to be related to the type of processor; AMD just doesn't like DF. Can anyone confirm? | ||||
| Steps To Reproduce | Get two PC's, one Intel, one AMD. Run on both, compare FPS. | ||||
| Tags | No tags attached. | ||||
|
|
How fast is your RAM, and capacity and speed of your cache? IIRC, the problem normally lies in the memory bus. Also, what other programs might you have running? Are you using the same save? Any differences in init.txt and d_init.txt? |
|
|
Exact same save and DF installation (running out of my dropbox, so I can shut down one PC and resume on the other). The max FSB (memory) on the home PC is 2400MHz; also got the memory to go with it, even though the cpu officially goes to 1866 (I did overclock the FSB, but am unsure how much). the work PC seems to run at 800MHz. What do you mean with cache speed? Are you talking about the pagefile, or the cache in the processors itself? The latter is 512kb in the work-pc, and I think the home-pc has four 1024kb caches. The home PC is a fresh install of Windows 8.1, and Windows Defender and Chrome are running in the background. The work PC is an old install, and runs F-secure (which eats up roughly 400MB of RAM, and 20-40% CPU constantly), which I can't disable. Chrome is active, as is Matlab, and a dozen weird IT-utilities that are required on our stupid corporate network. As a reference; one full pass through my data using Matlab on my home PC takes about two hours, while the work PC takes more than a weekend. In terms of sheer power, the home PC wins hands down. |
|
|
Those are interesting results, but I'm not sure what qualifies as a bug. It's not surprising that DF has different requirements than Matlab, and whatever's making the performance difference is not guaranteed to be unique to DF (it could even be a compiler issue). For a technical report like this, we'd need much more specific information about the cause of the discrepancy. |
|
|
I've got no explanation on why this exists, but have seen similar discrepancies between Intel and AMD, but usually favouring AMD, especially in graphics-heavy games, but sometimes also in brute data processing. It's quite rare to see such a large difference, especially one favouring Intel. After checking other Intel-based PC's in the department, and get similar results, and I do seem to remember hovering around an fps of 40 with my previous home PC (also AMD). I'd love to see someone confirm this discrepancy between AMD and Intel, and then we could keep it here for reference. It's not really something that requires fixing, in my opinion. My boss might disagree, but he can't have everything. |
|
|
That's not really a practical request for feedback. We can't try other compilers, intelligently profile the code, or really investigate this problem at all. The game running poorly on a specific architecture is almost certainly an unintended defect, and Toady should probably take a look at the issue at some point. |
|
|
I'm not asking anyone to recompile the game. However, broad claims require broad evidence. The current evidence is so narrow that it's impossible to tell whether the conclusion ("DF runs much better on Intel than on AMD") is generally true, or just a quirk of this test (one save, one pair of computers, one build of DF, no comparison of similar applications). Unless compelling evidence exists, it's not worth Toady's time, same as any other report. This evidence is more difficult to gather than most reports, but that's the nature of the claim. |
|
|
My guess would be Toady set a compiler flag to optimize for Intel chips. Said optimized code may be running in "compatibility mode" on AMD, significantly degrading performance. Checking Toady's compiler options would take Toady minutes, while I simply cannot do it at all. |
|
|
Offered info can beside claims like "DF runs much better on Intel than on AMD" also make for claims like "DF runs much worse on Windows 8.1 then on older Windows version" which (maybe also with "lower-layer-ness" of ascii graphics) seems (_to me_) more probable then the previous claim. But I would rather suggest to start a thread on DF forums with something like "try current version of DF on various platforms and write here your results" than making bugreports about this. If few dozen reports then manage to find some platforms where DF works really poorly, then it may be better time to get those results to Toady so those found issues may be somehow dealt with. |
|
|
You're right, Detros, this might not have been the proper place for question. I've made a thread on the forums (http://www.bay12forums.com/smf/index.php?topic=147618), copy-pasted some of the details in here, and am asking for others to share their experiences. Perhaps we'll find a consistent discrepancy, and maybe Toady traces this back to something, like the potential 'optimized for Intel'-flag (but more likely something more complicated). EDIT: I don't know how to make it a hyperlink, sorry. |
|
|
That seems like a good start. Please reopen this if a trend emerges. |
| Date Modified | Username | Field | Change |
|---|---|---|---|
| 2015-01-15 03:56 | vomov | New Issue | |
| 2015-01-15 05:29 | RoaryStar | Note Added: 0031933 | |
| 2015-01-15 05:56 | vomov | Note Added: 0031934 | |
| 2015-01-15 07:02 |
|
Note Added: 0031935 | |
| 2015-01-15 07:02 |
|
Status | new => resolved |
| 2015-01-15 07:02 |
|
Resolution | open => no change required |
| 2015-01-15 07:02 |
|
Assigned To | => user6 |
| 2015-01-15 07:03 |
|
Status | resolved => feedback |
| 2015-01-15 07:03 |
|
Resolution | no change required => reopened |
| 2015-01-15 07:03 |
|
Resolution | reopened => open |
| 2015-01-15 07:17 | vomov | Note Added: 0031937 | |
| 2015-01-15 07:17 | vomov | Status | feedback => assigned |
| 2015-01-15 07:54 | Loci | Note Added: 0031939 | |
| 2015-01-15 08:59 |
|
Note Added: 0031940 | |
| 2015-01-15 08:59 |
|
Status | assigned => feedback |
| 2015-01-15 12:58 | Loci | Note Added: 0031945 | |
| 2015-01-15 12:59 |
|
Note Added: 0031946 | |
| 2015-01-16 01:38 | vomov | Note Added: 0031951 | |
| 2015-01-16 01:38 | vomov | Status | feedback => assigned |
| 2015-01-16 01:38 | vomov | Note Edited: 0031951 | |
| 2015-01-16 01:38 | vomov | Note Edited: 0031951 | |
| 2015-01-16 07:27 |
|
Note Added: 0031954 | |
| 2015-01-16 07:27 |
|
Status | assigned => resolved |
| 2015-01-16 07:27 |
|
Resolution | open => no change required |
| 2015-01-19 13:18 | lethosor | Note Edited: 0031951 |
