Hardlimit test bank

[krampak]

@kynes said in Hardlimit test bench:

@krampak I don't know if this will happen, on some micros that have a higher turbo for 1-threaded applications it was impossible for me to validate it, it always stayed below the minimum.

It could, but it is still curious that it is only using those 2 modes, what you describe should happen in all modes.

[kynes]

@krampak right, I hadn't read you properly. I thought it was in all modes.

[cobito]

@krampak Well, that seems very strange indeed. I'll have to review the mmt ratio restrictions soon because as @kynes says, there are cases where it's normal for it to come out below 0.97/0.5. The only reason that comes to mind (a bit far-fetched) is that in those modes it gets hotter than usual in test#4 and stops the car a bit on the first peak of test#1. But looking at the performance graph, that doesn't seem to be the case.

If we look at the coincidences, it could be that some program in the background coincidentally did something during that execution. By the way, in a few weeks there will be a live-usb of the test bench to avoid interference from other programs and make it agnostic to the installed operating system.

If you repeat it and it goes wrong again, put the link to the result here.

[krampak]

Yes I have tried it again, same thing: https://bm.hardlimit.com/result.php?bm=70620578ee58b11e46bcbfbfbec8104e588

I will try with the live USB when it is available.

[kynes]

@krampak Have you tried with 4 threads instead of 8?

Mesplico, let's see if there is a program problem with the hyperthreading implementation of those micros, and with four physical cores it's not a problem. The ideal thing would be to be able to disable it in the bios, to see if it gives a coherent result without those threads.

[krampak]

@cobito I'm testing it at the Lan Party and I'm not sure if it's the PC or the server, but this is what happens when I click on validate results:

https://imgur.com/a/Q6vEI

[cobito]

@krampak ok, that's a bug from the change to the new version. It comes out because you're using an old version of bm. Download 1.1 and it will work.

[krampak]

I don't know how I didn't notice, it worked later @cobito.
By the way, how would you see creating a test bench page that showed all the CPUs sorted by performance (taking the best result of each CPU)?

PD: In About you have a mistake, it says: About Hardlimit Benchmack

[cobito]

@krampak said in Hardlimit Benchmack Test Bank:

P.D.: In About you have a mistake, it says: About Hardlimit Benchmack

Corrected.

I don't know how I didn't notice it, it worked later @cobito.
By the way, how would you feel about creating a test bank page that shows all the CPUs sorted by performance (taking the best result of each CPU)?

To be honest, it's a good idea. I've been thinking for a while about how to do it with bar graphs. At the moment, many results are being lost because in the tops only logged results appear and I have sent a good number of models anonymously that don't appear anywhere. This week I'll get to work on it.

[cobito]

@krampak, is already on the front page of the site a top50 of processors. For now they only appear in mode 0. I have to do some optimizations in the internal engine because it's starting to get a bit slow. When it's lighter, other tops will appear.

[Yorus]

Well, I have already tested my equipment and I have updated the signature, now I will see if they are more or less normal results.

Edit: I have tested mode 0 and my processor comes out in the top 50 with error:

15. error op t3= 0. Please, report

It is the one with the score 24024 in mono and 78797 in multi.

[cobito]

@yorus Ok, that error appears when processor identification data is missing and I have to enter it manually (it will always appear when a model is submitted for the first time).

It already works fine.

[Yorus]

Perfect, I will go ahead and test some more, I don't spend money on very normal micros, hehe.

[cobito]

We have just launched the new internal engine that accelerates the loading time of the heaviest pages by an order of magnitude. For the moment it has only been applied to the front page processor rankings in order to see how it behaves. Over time it will be extended to the rest of the site.

And as now it seems that the front page rankings are going at a reasonable speed, two general tops are also shown for the maximum modes of each model (you will see a much more interesting ranking with more models).

[QVENGADOR]

Wonderful.

[cobito]

Now the front-page CPU rankings have a different aesthetic that remains the same regardless of the browser (as long as it is recent).

The new back-end has also been applied to the fiches that now load much faster.

Taking advantage of the speed improvement, a new section has been implemented where the 10 most similar CPUs in every way are shown: those that are most similar in both single and multi-core performance in both mode 0 and the highest mode available. The similarity is ordered in descending order. It is possible that some models in lower positions do not look too similar simply because the available repertoire is still limited.

Example

[Xevipiu]

It still recognizes 80mhz more, but well, now the application is starting to have a better color

[cobito]

@xevipiu Thanks. The thing is that each architecture is a bit different when it comes to measuring frequency, but a correction per model is planned (for which I need that model to have gone through the test bench beforehand). Anyway, an error of 80MHz in 3Ghz? still complies with the <3% rule. But yes, it is something that needs to be improved.

[cobito]

Guys, I'm pushing version 1.2. I need someone with a Sandy Bridge (2000 series any i3, 5 or 7) to do me a favor by running the benchmark in mode 6.

Thanks in advance.

[Sylver]

@cobito I just went through it and it crashes my hlbm-core, in the end it says that something went wrong and that I should try a lower mode.