Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[wierd]

Sadly, the DF running in emulation ran out of memory before reaching year 250.

However, it took 6min,30sec to land on a stable world with 11 rejections, and ran for over an hour and a half trying to generate history before running out of RAM, someplace around year 210.

I think a good deal of the sluggishness is from the phone using zram backed swap. (The emulator does not know zram swap is not normal RAM, due to android being based on Linux, and linux does not discriminate too much.) While faster than going to the SDCard, it has a large CPU penalty from the compression/decompression going on.  I know this, from running similar benchmarks on my liberated chromebook.  A fortress running at ~15fps with zram swap off, runs at ~8fps with it on, and memory being exhausted. (It is VERY apparent when zram starts getting used.)


I would suggest a lighter-weight benchmark for lower-end systems.

[Unknown72]

Sadly, the DF running in emulation ran out of memory before reaching year 250.

However, it took 6min,30sec to land on a stable world with 11 rejections, and ran for over an hour and a half trying to generate history before running out of RAM, someplace around year 210.

I think a good deal of the sluggishness is from the phone using zram backed swap. (The emulator does not know zram swap is not normal RAM, due to android being based on Linux, and linux does not discriminate too much.) While faster than going to the SDCard, it has a large CPU penalty from the compression/decompression going on.  I know this, from running similar benchmarks on my liberated chromebook.  A fortress running at ~15fps with zram swap off, runs at ~8fps with it on, and memory being exhausted. (It is VERY apparent when zram starts getting used.)


I would suggest a lighter-weight benchmark for lower-end systems.

It's interesting to note that everyone's getting 11 world rejections. It might be because everyone's using the same world seed + history seed? Or is it just that this specific world seed will always result in 11 rejections and 12 is it's magic number to get that stable one?

I could do a lighter-weight benchmark once I get home, maybe instead of large, go to medium or small world and go from there? Since it's typically size that causes things to consume more ram due to needing to load and simulate more.

[Soadreqm]

If the world generation runs the same way, as it should with the same seed, it's obviously going to generate the same number of failures every time.

[Insanegame27]

I’ll test this when I get home on a stock R7 3700X - Tomahawk Max - Radeon 5700 with 3200Mhz RAM

[Unknown72]

If the world generation runs the same way, as it should with the same seed, it's obviously going to generate the same number of failures every time.

That's fair, I never really paid much attention to anything world gen related at all. So I just assumed that the seed would give a basis, but there wouldn't be the same amount of failed worlds and the like.

[Insanegame27]

World Settings


PC hardware:
R7 3700X stock
MSI B450M Tomahawk Max
Sapphire Pulse Radeon 5700

Spoiler: 2x8GB DDR4 3200MHz Vengeance LPX (stock XMP) (click to show/hide)

Y125 = 4m 36s
Y250 = 13m 08s

Spoiler: world (click to show/hide)

One thing I would like to test is fps at year 1 vs 125 vs 250 vs 500 if we were to all to take a specific embark location with the same custom point limit embark with like 10k plump helmet spawns and 170 initial dwarves just to create lag.

[Unknown72]

One thing I would like to test is fps at year 1 vs 125 vs 250 vs 500 if we were to all to take a specific embark location with the same custom point limit embark with like 10k plump helmet spawns and 170 initial dwarves just to create lag.

Might be a bit hard to do that seeing as we all have different equipment. Unless i'm reading it wrong and you actually mean we test and compare those 4 timestamps on our end, post the results and see how it stacks to others results?

[Insanegame27]

One thing I would like to test is fps at year 1 vs 125 vs 250 vs 500 if we were to all to take a specific embark location with the same custom point limit embark with like 10k plump helmet spawns and 170 initial dwarves just to create lag.

Might be a bit hard to do that seeing as we all have different equipment. Unless i'm reading it wrong and you actually mean we test and compare those 4 timestamps on our end, post the results and see how it stacks to others results?

Exactly what I mean. I know that 40.xx started tracking the outside world while you play, moving armies and generating events and stuff, but dont know how that scales to performance.

[Larix]

Sadly, the DF running in emulation ran out of memory before reaching year 250.

However, it took 6min,30sec to land on a stable world with 11 rejections, and ran for over an hour and a half trying to generate history before running out of RAM, someplace around year 210.

How much RAM/what processor?

I tried on two systems:

System 1:
i5 - 3230M (2,6 GHz, sometimes turbos to 2,7 or 2,8)
8 GiB RAM, apparently DDR3-1600 (Thinkpad X230, seems to only use this type)
Windows 7

Time until start of history:
0:47
History generation until year 250:
13:30 - 14:00
NB: not fully sure on the histgen timing; the first, properly recorded run apparently had typos in the name seed which not only created a differently-named world, i could also see a road on my final map that didn't exist in the reference worlds, indicating different events. The second run, with the correct settings, started making history after 47 seconds as well, but i missed the exact moment when histgen ended because i was wrestling with system two at the time; it was definitely under 14 minutes.

System 2:
Atom Z3735G; 1,33 GHz (turbo up to 1,83)
RAM: 1 GiB LDDR3-1333
Windows 10 (lowest acceptable specs with the given amount of RAM and a mere 16 GB storage, but fully up to date)

Getting to the stable world:
0:10:00
History until year 250:
2:09:40
Note: "finalising" after histgen alone took almost three minutes.
I got the same world name as the screenshotted worlds in this thread, but more dead units and only ~480 000 events; maybe because this was the 32-bit version?? I'm frankly surprised the system managed to complete the task. It's by far the weakest performer in this thread.

Just for funsies, i ordered both systems to save the result. System one took just over 4 minutes to save to HDD. System two saved to an inserted micro-SD card and took 43 minutes, crawling to just over three hours for the whole ordeal.

PS: both systems with iGPU, unsurprisingly; intel HD4000 and the "HD" but otherwise unnamed junk Intel put on Z3700 atom chips.

[Insanegame27]

Just ran this:

11 rejected worlds
55 seconds to stable world creation
~8:40 to completion at year 250 (including stable world creation time, so 7:45 spent on history)

I had a couple of browser tabs open and ran the version bundled with the LNP so DFHack, etc. running in the background

Using all the seeds spelled out up top.

Platform:
i7-8700
SSD
16GB DDR4 16-18-18-38 @ ~2880Mhz
ASRock Z390 Phantom Gaming 4 mainboard



anewaname's results seem to suggest that, if the tests were accurate, that extra bump in RAM clock made up for a considerable difference in processing power.  What kind of RAM you got there?  Specs on latency?  Are you using standard XMP to get to 3000Mhz?

With a slower memory clock speed with similar/identical timings and a comparable CPU (mine is much newer but turbos to 4.4 instead of your 4.7) you beat my bench quite considerably. 8:40 vs 13:08. Do you think it is that small turbo edge you have that crushes my bench or do you reckon another factor is at play?

[wierd]

RE: Larix

This was a whimsical benchmark attempt from inside that silly "Windows Emulator" I have on my ARM based cellphone.

Said phone has a dual core ARM based Snapdragon 810, running at ~2ghz and 2GB of RAM (plus some swap)


The CPU emulated is whatever the hell Eltech's x86 emulator presents itself as.  I suppose I could run Sisoft Sandra on the emulator to try and find out, but why bother?

[Larix]

Re: wierd

Thanks, that's what i wanted to know, for comparison with my Atom system (it is, i kid you not, a seven-inch tablet running desktop Win10, 32bit); just to know what physical hardware backed up the emulation - it certainly does produce performance comparable to lower-end native systems.

That my tablet managed to complete the task at all with only 1G (plus swap) of RAM might have been due to it being a native system. I suspect that its very low memory bandwidth - single-channel 32bit (!) - contributed a lot to the low performance. When the processor works at full load, the system tends to heat up massively, to the point of throttling down to complete uselessness; in this benchmark, the device didn't heat up significantly, so i guess the processor didn't really have much to do and mainly waited for RAM/swap to respond.

On the other hand, i'm surprised how well my laptop did - 3rd gen mobile i5 with 1600 MHz RAM wasn't much slower than a Ryzen 3700 with 3200 MHz RAM...

[DontMineYellowSnow]

Just ran this:

11 rejected worlds
55 seconds to stable world creation
~8:40 to completion at year 250 (including stable world creation time, so 7:45 spent on history)

I had a couple of browser tabs open and ran the version bundled with the LNP so DFHack, etc. running in the background

Using all the seeds spelled out up top.

Platform:
i7-8700
SSD
16GB DDR4 16-18-18-38 @ ~2880Mhz
ASRock Z390 Phantom Gaming 4 mainboard



anewaname's results seem to suggest that, if the tests were accurate, that extra bump in RAM clock made up for a considerable difference in processing power.  What kind of RAM you got there?  Specs on latency?  Are you using standard XMP to get to 3000Mhz?

With a slower memory clock speed with similar/identical timings and a comparable CPU (mine is much newer but turbos to 4.4 instead of your 4.7) you beat my bench quite considerably. 8:40 vs 13:08. Do you think it is that small turbo edge you have that crushes my bench or do you reckon another factor is at play?

Its an Intel vs. AMD thing.  Dwarf fortress being single-threaded and so dependent on RAM speeds (once you fill the cache) is more Intel's domain the AMD's, which otherwise excel at heavily multi-threaded apps.  Things like SmartCache on the Intel chips allow that one core running DF to access the whole chip's L2/L3 cache, while AMDs CCX architecture is a L3 cache technology only, for instance.  Also, the memory (write) bandwidth and latency of AMD processors' memory controllers are both weak points, though not quite as bad as it used to be.  These are the things that DF, and a lot of games, need any why AMD suffers in gaming vs Intels which they should, on paper, blow away.  Great write-up though using an i9 which is a bit faster than mine at a single core workload:

https://www.techspot.com/article/1876-4ghz-ryzen-3rd-gen-vs-core-i9/

[carewolf]

Linux Debian 64-bit
CPU: Intel Core i7-5820K @3.40GHz 6C/12T (Haswell EE)
GPU: GeForce GTX 1070 8GB
RAM: 32GB (4x8GB) DDR4 2600MHz (4x channel)

Y0: 0:45
Y125: 1.46
Total: 11:39

I thought it would be more interesting. My processor due to being an EE model with 4 memory channels generally performs very well in Dwarf Fortress. If I take out two of the memory sticks a late game fortress will halve the FPS, but it appears world gen really is CPU only, and didn't get boosted by having a server-style memory configuration.

[Larix]

Today i stumbled over an old Acer Aspire someone had left by their door, complete with power brick but without battery. Took it home to see if it was a freebie or just someone outsourcing their garbage disposal and yes, it's alive and has a clean Win10 installed. So before i re-donate it, i ran The Benchmark (tm):

CPU: Intel Core2 Duo P8400 2,26 GHz (pub. Q3 2008)
RAM: 4 GiB 1066 MHz
Stable World: 1 min 54 s
History until year 250: 29 min 12 s (3:31 until 125, 25:41 for the rest)
Total: 31 min 5 s

Barely beaten by methylatedspirit's AMD A9, still massively faster than an Apollo Lake Celeron (but then that's just an upmarket Atom).