What can make Tumbleweed much slower in latest Phoronix OS comparison?

English Open Chat
1

Hello everyone,
There have already been a few posts (like this one) on this forum about tumbleweed showing comparatively bad scores on benchmarks. I haven’t seen a clear explanation as to why though.
Again today I came across some multi linux distribution comparison on phoronix, on Framework Laptop 13 with the AMD Ryzen AI 9 HX 370 Zen 5 + RDNA 3.5 SoC, and tumbleweed ends up last by quite a margin.

2505-phoronix-multi-OS-comparison
2505-phoronix-multi-OS-comparison694×525 30.2 KB

I don’t think it relates to the typical Btrfs setup as the benchmarks are not about I/O. I might try and have a look at a few things in tumbleweed settings, but it could be things like compiler and compiler options, that would make investigations quite involved.

I am curious and opening the discussion here…
Thanks all

1 Like
2

It’s AMD hardware…

3

Yes, but why would this impact tumbleweed more than others?

4

@opsusemaco because there are many openSUSE users with older hardware, so many optimizations can’t be done, there have been some mitigation with hwcaps, but there could be improvements with v4 support(?).

Remember, benchmarks are only relevant in the eye of the tester. How does your hardware perform for your tasks? I have no issues here with my setups, they do what they need to do :wink:

Plus my experience with AMD graphics is abysmal, all intel/nvidia these days. I do have AMD hardware, a dual AMD gpu laptop that’s not used. My on the road laptop is AMD but getting old in the tooth as well, it does what it needs to do with Leap 15.6…

3 Likes
5

Yes like me :smiley:
Entirely agree that benchmarks have very limited usefulness, and tend to be over-interpreted. Still they provide good food for thoughts and sometimes, targeted optimisations.
Thanks

6

First, you have to pay very close attention to the benchmark details.
Some of those are measuring milliseconds, so a huge loss is less than a second. Others are measuring seconds and TW scores pretty well. The Darktable scores were 2 seconds (1.9625 I think) vs 2.6 seconds for TW. A half a second is nothing…but it is a loss.

Then TW flat lines before it gets out of the gate on GravityMark and vkpeak.

vkpeak describes their program as:

vkpeak

A synthetic benchmarking tool to measure peak capabilities of vulkan devices. It only measures the peak metrics that can be achieved using vector operations and does not represent a real-world use case.

I did not read every word but I do read Phoronix regularly.

Specs of the laptop are:

Framework Laptop 13

Ryzen AI 9 HX 370 CPU (released less than a year ago)
I looked up the CPU Mark scores:
Multithread Rating 35165
Single Thread Rating 3966

Zen 5 + RDNA 3.5 SoC
2 x 16GB DDR5-5600 memory
1TB WD_BLACK SN770 NVMe SSD

That’s not really high CPU Mark scores. Based on that and where TW failed miserably, which killed its final score, I’d say TW has a problem with RDNA 3.5 graphics (it’s a Soc), and/or a mis-configured driver or something causing the really bad scores that killed TW’s final score.

If you go read other benchmarks at Phoronix you’ll see TW is in the ballpark with most of the others, and like I said .5 seconds is not a huge win to brag about, but it’s still a win. That’s like losing a dragster race because your big dog was riding with you. :sunglasses:

Intel’s Clear Linux is optimized for Intel processors etc so it does pretty well when they’re testing Intel CPU’s. Still this isn’t Intel and it’s RDNA 3.5, so I’d say that they have it tuned up pretty good. But what does it lack that every day users want? I don’t know.

Back in the day when processors had no real horsepower, when reviewing motherboards etc, they measured 52 vs 54 seconds and with coolers it’s 57C vs 59C, but the scale of the charts made it look like you lost BIG!

They tried to push $60 coolers with a 2C advantage vs a $30 cooler that would do the same job. And I’m not going for a board with less features such as ports for a 2 second gain on some benchmark that isn’t going to affect my daily use. The missing ports do affect you every day.

Still…TW has some problems there and someone with the right hardware could test it and maybe find the problem.

3 Likes
7

Sounds right up my alley … now all I need is some nice person to donate one of them fancy laptops to me :wink: … seriously tho … going to read the article right now and see if I can download the benchmark to see what if anything “tuned” can do about it …

1 Like
8

@dart364 just zypper in phoronix-test-suite

1 Like
9

Thank you Malcolm :+1:… now about that fancy laptop donation :rofl:

10

@jsmith64 yes, so I tested darktable with openCL on a RTX4000 2.2 seconds. Seems a lot of those tests where around the cpu and not the graphics card?

11

@dart364 no need for fancy, just useful :wink: This desktop has been built up over a number of years keeping an eye out from bargains…

12

Just got done looking over the Phoronix article and a lot of the Debian 13 testing marks show huge power usage … key word here is “testing” … if you were to use cpupower or tuned to change the profiles to “ondemand”’ or “performance” I’m sure pretty sure TW would fare significantly better … TW isn’t “outta the box” for blazing speed … but a couple tweaks here and there …

1 Like
13

On what may be the difference:

Screenshot_20250516_075517
Screenshot_20250516_075517747×504 29 KB

From this page

So some power tuning done differently?

1 Like
14

We need to start a “tweaker” thread … :laughing:

Check out my Tuned thread

I’m all in for that …

2 Likes
15

I read your comment about cpupower and it reminded me that I needed to fix the governor. It was not retaining what I set it to at boot. It’s working now. Note that Michael at Phoronix said the Scedutil governor is better for a lot of workloads.

The developer of Hardinfo2 (hwspeedy) has been adding features. The Synchronize now has a user and group tag that you can use to make a group with your friends and compare scores. It’ll download the scores for whoever is in the group.

You can also specify your CPU and it’ll download scores for that CPU. You used to have dogs, big dogs and old dogs :grinning:, while you were in the middle with nothing to compare to. I mentioned it to him and he said he was planning to fix that and went right to work on it. Really nice guy.

It’s very easy to compile and it’ll spit out an RPM for you to install it, so I helped him test it. We know scores vary with what’s running and even the ambient temperature. Last fall we left and the temp dropped a lot and it was kind of cool in here. I had good scores then. :roll_eyes: I told him about it and admitted that I cheated, and I may have to keep the heat off in here, but the wife wouldn’t go for it. :sunglasses:

I’ve used it a lot to get the power and temps under control on this beast. It’s really handy for a quick rough speed/temp estimate to see if your changes really matter. Once you get close they really don’t for the most part.

I ran the same machine for 12-13 years and finally upgraded when AM5 took over…got a 5950X. It’s not a dog but it’s falling behind fast. If you buy the fading generation you can save a lot of cash. I built a 7950X3D for my friend and now they’re already doing the Ryzen 9000 series. Crazy! I chose the 7950X3D for him because he wants to game (it’s better at gaming than the 7950X) and it’s only 120 watts, while the 7950X is 170w!

I invested in storage and a lot of it.
36TB in the machine and 74TB external in two 5-bay 10 Gbps enclosures (110 TB). I’m using a 12 year old EVGA 750 watt G3 and I yesterday I bought an EVGA 1300 watt G2. They’re both gold but the G2 has 140mm ball bearing fans. I’m getting ready for future upgrades, way far out there in the future…Board/Chip/RAM.

Last upgrade was the case…got a AI Raider XL.
(anidees AI Raider XL Full Tower Tempered Glass XL-ATX).
My forever case for sure! Google it.

I had a Lian Li Lancool Dragonlord and it’s still my favorite until I out grew it after 12 years. They’re just fantastic to build in.

Yeah…we need a chat, benchmarking and building thread.
I like cool simple benchmarks an have compiled a couple of those and no, I didn’t modify them. Start a chat thread and I’ll look up what they are and post them for you. They’re not complicated at all. Maybe we can tweak TW a little.

I looked up the AMD Radeon 890M “RDNA 3.5” iGPU. It’s a mobile integrated GPU that is comparable in performance to an Nvidia GeForce GTX 1650. It seems to me my A750 should do well against it. I just can’t see it beating my Tumbleweed in those same tests that TW did so poorly in.

Google Sez:
The Intel Arc A750 generally outperforms the GeForce GTX 1650 in most gaming scenarios, particularly at 1080p and 1440p resolutions. The A750 offers significantly higher clock speeds, floating-point performance, and pixel/texture rates. While the 1650 is a capable entry-level card for 1080p gaming, the A750 provides a noticeable boost in performance and features like ray tracing.

HWspeedy is making a lot of changes to Hardinfo2.

1 Like
16

The benchmark results suggest that the author never checked the detailed output. E.g. vkpeak fluctuates wildly and puts Tumbleweed at a severe disadvantage.

I don’t waste time on these kinds of errors. Users better do their own comparison based on the requirements of their use cases.

2 Likes