TW Tweakers ... Top of the line

dart364 · May 16, 2025, 7:27am

So I got a little offended at the idea that TW lagged behind all these other OS’s and I’d like to change that … I know that it’s because OpenSuse is focused on an “overall” experience and that’s fine … I’d like to make a desktop/gamer sort of “mods” that could be easily enabled and disabled

I’ve been playing with “tuned” and find it superior to “cpupower” for the simple reason that tuned does more than just cpu

So on that note I’d love to hear about any “alterations” that would make OpenSuse the ultimate desktop machine

Start here with my Tuned thread

dart364 · May 16, 2025, 7:51am

Here’s some fun commands:

ls /proc/sys/kernel

Lotsa stuff to play with there

ls /sys/module/

So much info … so little time … you get the idea

jsmith64 · May 16, 2025, 10:42am

A year ago I built my friend a Ryzen 9 7950X3D with a 4070. He wants to get into gaming and I’m not a gamer. This looks like a good opportunity for me to learn to set up his for gaming and learn to tweak Tumbleweed at the same time.

He built it for his grandchildren hoping they’ll get off of the console and get interested in computers and maybe learn something along the way. I put Windows 11 and Tumbleweed on it. W11 for gaming and TW for all of the software and security. He’s really impressed with the software that you have without hunting for it.

He’s not hooked on all of Windows’ bad habits and is totally open to Linux, while his oldest son (about 30) doesn’t like Linux because it’s “hard”. I told him so was Windows when you started on it…and you don’t know much about it either. It’s a good time to learn the right way!

I have Steam running real well on TW, and he has Starlink so it’ll download without me taking a nap. I’m out there on the weekends so I’ll be able to test it.

I posted in the other thread about some benchmarks that I compiled and I’ll look those up…I forgot what they are. They’re small but useful . If you have a problem you’d instantly be able to see if your fix worked.

I’ll get back here later today and catch up.

dart364 · May 16, 2025, 10:59am

I installed hardinfo2 and it was very nice … but they needed (ipv3?) which I found to be a bit sketchy (gubmint) and I’m not comfortable with that … also heard that theire “test servers” are always “busy”

dart364 · May 16, 2025, 11:12am

And there was only one network test … which from reading various posts “may or may not” work …

dart364 · May 16, 2025, 11:46am

Give me actual results … changes to the profiles … real tests … etc

knurpht · May 16, 2025, 1:26pm

Bringing up this old one:

jsmith64 · May 16, 2025, 9:37pm

Here’s a benchmark that I compiled. I couldn’t remember if it wouldn’t compile to test my Intel ARC A750 or if I skipped that at the time.

But here’s the answer:
The support for the GPU will only be enabled if CUDA is found.

Just now, I was doing other things when I ran this. I’m sure that you can boot to a console and get a better score. Also, this thing floorboards it from the time you hit go! If you have stability issues, this might find them right away. Mine rebooted one time immediately after I hit enter. I changed the Curve Optimizer from -30 to -26 and that seems to have fixed it.

I think Dr-Noob might not be that much of a Noob.

Peakperf

Get it here.

~/code/peakperf-master> ./peakperf

------------------------------------------------------
    peakperf (https://github.com/Dr-Noob/peakperf)
------------------------------------------------------
        CPU: AMD Ryzen 9 5950X 16-Core Processor            
  Microarch: Zen 3
  Benchmark: Zen 3 (AVX2)
 Iterations: 1.00e+09
      GFLOP: 4096.00
    Threads: 32

   Nº  Time(s)  GFLOP/s
    1  1.75933  2328.16 *
    2  1.89295  2163.82 *
    3  1.82237  2247.62
    4  1.77950  2301.77
    5  1.77575  2306.63
    6  1.78329  2296.88
    7  1.77555  2306.89
    8  1.78178  2298.82
    9  1.78674  2292.44
   10  1.78432  2295.55
   11  1.79100  2286.99
   12  1.78201  2298.53
------------------------------------------------------
 Average performance:      2293.10 +- 16.24 GFLOP/s
------------------------------------------------------
* - warm-up, not included in average

Peakperf
Microbenchmark to achieve peak performance on x86_64 CPUs and NVIDIA GPUs.

Hardware support

Supported microarchitectures are:

CPU (x86_64):
- Intel: Sandy Bridge and newer.
- AMD: Zen and newer.
GPU:
- NVIDIA: Compute Capabitliy >= 2.0.

For a complete list of supported microarchitectures, see section [5]

NOTES:

Only GPUs that support to read the freqeuncy in real time (using freq.sh) can be actually evaluated.

Other microarchitectures not mentioned here may also work.

jsmith64 · May 16, 2025, 11:31pm

CPUmark. I didn’t compile this one. Get it here.

I remember running CPUmark on an AMD K6 II 500 MHz back in the day. I scored almost 16 (15.xx something). I still have the benchmark. The board was a Tyan Trinity 1590S. My friend was only scoring 9 with the same CPU and Stepping. He asked me why his scored so low. I told him that it was his cheap PC Chips board. He always called me for advice but he didn’t before he bought that board.

If you were around in those days then you know what they were called. They were also called Plastic Cache Chips Inc.

I brought this up because a good stable board is a must. You don’t have to buy the very best, but study what’s available and make sure what you buy is a good stable board because it’s the foundation of your entire system. Also, a Ryzen 5600G (with Radeon Graphics) uses PCIe Gen 3 and a Ryzen 5700X (no graphics) uses PCIe Gen 4. These are the small details that inexperienced users miss and they make a huge difference in performance.

The average scores for a Ryzen 9 5950X CPU are:
Multithread Rating is 45456
Single Thread Rating 3469

As you can see I’m beating those scores considerably. Again, keep in mind that the guys that don’t know how to set up their computer properly drive the average scores down, just like the guys pushing their machine with dangerous voltages and high temps drive the averages up.

I just ran it with the browser running and I have a lot of other stuff running that you don’t need for basic computing, but I’m not a basic computer user. I have CoolerControl and two years worth of customization running, so I’m not sure what else. It’s not worth my time to check. But that will drive the average scores down a little because my score will be a little lower than it could or should be (but I didn’t submit my scores).

PassMark PerformanceTest Linux


AMD Ryzen 9 5950X 16-Core Processor (x86_64)
16 cores @ 5086 MHz  |  62.7 GiB RAM
Number of Processes: 32  |  Test Iterations: 1  |  Test Duration: Medium
--------------------------------------------------------------------------
CPU Mark:                          48667
  Integer Math                     192432 Million Operations/s
  Floating Point Math              108280 Million Operations/s
  Prime Numbers                    236 Million Primes/s
  Sorting                          63217 Thousand Strings/s
  Encryption                       47600 MB/s
  Compression                      682305 KB/s
  CPU Single Threaded              3676 Million Operations/s
  Physics                          2198 Frames/s
  Extended Instructions (SSE)      37937 Million Matrices/s

Memory Mark:                       Incomplete
  Database Operations              0.0 Thousand Operations/s
  Memory Read Cached               0.0 MB/s
  Memory Read Uncached             0.0 MB/s
  Memory Write                     0.0 MB/s
  Available RAM                    0 Megabytes
  Memory Latency                   0 Nanoseconds
  Memory Threaded                  0.0 MB/s
--------------------------------------------------------------------------
Results not submitted

One thing I do is put my browser temp files in a RAM Disk. This really helps the speed if you’re using a standard hard spinner drive because fetching data from RAM is instant compared to a drive. With a SSD of any kind, you’ll get a good boost in your browser speed, but it’s a lot of wear on a SSD. Using RAM speeds it up, but you won’t notice like you would with a spinner. Still, you need to stop the constant writes to your SSD by using a RAM Disk. It adds up over the years.

As for tweaking Tumbleweed itself, I’ve probably did a few, but I can’t remember what they all are. When I fix or customize something I make a text file that says fix whether it was a fix or a customization because both are the same to me. If you aren’t getting full performance, then it needs fixed. I guess I need to add tweak to the file name to make them easier to find.

The simpler benchmarks are great for finding out if you have a real problem like instability. They’re also good to get a rough estimate if you machine is performing as it should. If you run them once in a while and see a huge drop, then you need to find the problem. This happened to me when my Sysbench Multi-Thread score in Hardinfo2 dropped from 105,000 to 88,000 (17,000 lower).

They upgraded Sysbench from 1.0.19 to 1.0.20 and that caused the score to drop. You can run Sysbench without Hardinfo2. This is the exact command that’s run from hardinfo2 but change the number of threads to 16 if you have an 8 core CPU (8 if you have a quad core etc). You must install Sysbench then run this.

sysbench --test=cpu --time=7 --num-threads=32 --cpu-max-prime=10000 run

They made changes to Sysbench and going back to the older version confirmed it. I reported it to the Hardinfo2 developer so he’d know what was going on if users started reporting that his program was the cause. A lot of this is old news to you more experienced users, but the new guys need to know this stuff.

A lot of them are young gamers who haven’t had much real computer experience. Some have ran Windows for years but now they are working with Tumbleweed and need some tips. I run into info from the older timers, the guys who administered Unix systems for years, all of the time. They make me look like a beginner and I am, in comparison to them. I haven’t had any training like they did and you just can’t replace experience with anything else.

jsmith64 · June 3, 2025, 9:53am

I’ve been playing with tuned and have came to the conclusion that people do not know how to tune their BIOS to get the best performance out of it. Now Michael at Phoronix knows his stuff well. If scores are low there, it’s the OS I would think.

I tried all kinds of things and can’t get enough gain to call it a win. But my machine is running right up with the best using these same parts and air cooling. I can’t even match my CPUmark score that I posted here in this thread. Wait…I just checked. I posted 48,667 and got it back up to 48726.

I had it all messed up…45K and worse. I’ve been building this machine since January 2023. I’m upgrading and replacing parts and working on the cooling. Then the zypper downgrade bug hit.

I have more details but got side tracked on zypper dup. Anyway, I just don’t see how TW is that far behind. Now I know Michael was running special benchmarks that may favor Intel or AMD.

But I couldn’t get any impressive improvement at all. I upgraded the 12 year old 750w power supply to a good 1300w. I was having a random reboot once a week, twice or none for 3 weeks etc. Random.

I replaced the RAM with good G.skill, didn’t help. Then there was a BIOS update. Couldn’t be that easy but I haven’t had a reboot since May 12th. I replaced the PSU after that because it was already planned. So I may never know…but I’m going with the BIOS.

The BIOS was released on 2025/04/07 and I knew about it, but didn’t think that was it. It said to support newer processors. Well…duh! That’s what I get for thinking. A lot of guys sent their CPU’s in as defective…new CPU was the same.

So I figured that BIOS update hurt my scores. No, it was crap I changed and didn’t make good notes. I’m back up to speed and did some more testing with Tuned and Cockpit’s options.

I did another unusual upgrade to the dual tower CPU cooler.
It’s a Thermalright Phantom Spirit 120 EVO CPU Cooler.

See that gap? There’s 30mm of space between the towers. The guys crammed a Phanteks T30 120 x 30mm thick fan in there. Some said it fit, others said it was too tight.

The T30 cost $40+ and nope, I ain’t paying that. I found the Thermalright TL-H12-X28 ARGB 120 x 28mm thick fan for $11. It fits great, has more airflow than the originals, and I’m only using one in between the towers. It cools just as good and is quieter than the factory dual K12 fans and I now have 2 spare K12’s if I need them. And it doesn’t block the RAM at all.

So again, I thought the cooling was killing my score.
Nope. BIOS settings.

Michael has all kinds of video cards and things and I think he had something off causing those low scores. That’s 100% based on me reading tons of articles on performance and benchmarks etc over the last 30 years.

I also built my very first Kernel, version 6.15. I didn’t see any performance gains to brag about. I’m not an expert at setting kernel build options and they can make a huge difference.

They need to get the zypper bug fixed and I’ll see about building Kernel 6.16. You’ll be able to optimize it for your particular CPU! Now you might see significant, repeatable, measurable gains. It still wouldn’t make a noticeable difference for casual users, but it may boost benchmark scores by a lot.

I really don’t care unless I’m way below normal because that would mean I have a problem. But SUSE was behind in Michael’s benchmarks. And remember, there’s a database of what your machine should do whether it’s a 500 MHz Pentium, or a 128-core AMD EPYC Processor.

I was waiting to see if factory TW 20250602 was going to be published but it’s looking like it’ll go to 20250603. I’m looking forward to tweaking the Kernel and 6.16 won’t be released for a while. I’ll probably play with 6.15 a little more if I quit wasting time on bugs.