Hardlimit test bank

[cobito]

@krampak said in Hardlimit test bench:

@cobito I think it was on a Pentium D but I can't be sure right now, if I find it again I'll let you know.

Now I was trying to upload the ones for the i3-6100 again but there's no way, the Windows 10 antimalware service MsMpEng.exe (even with real-time scanning turned off) messes up the result.

I'll see how the issue of the digital signature of the executable goes, which apparently reduces false positives.

@krampak said in Hardlimit test bench:

@cobito the link on the download page to get the necessary Windows update on non-updated computers goes directly to the 64-bit version. I say this because I've just come across an unpatched machine running Windows 7 32-bit and I didn't understand why it wouldn't let me install the update until I noticed the bit thing. Maybe it would be good to add both links mentioning the architecture.

Noted. In principle, with just having Windows updated it works, but yes, if you want to transfer it to a PC with freshly installed Windows, I should put both architectures.

[krampak]

@cobito This afternoon I had to terminate the hlbm-detect.exe process (after finishing the last test) because it had been waiting for 10 minutes and nothing happened. I had to kill the process and the result publishing window popped up. It's the test I uploaded with an i5-3470 (I only uploaded 1 mode because I saw that it didn't reach 0.99).
Then I tried with another machine with the same processor and it didn't even let me open the task manager (also 10 minutes stuck in test4, although the test probably, as in the previous case, had already finished).

[cobito]

@krampak That process is responsible (among other things) for measuring the processor frequency. Sometimes in older models it has problems (the lock you mentioned appears) but I have never seen it in current micros. I have had it noted for a long time, to see if next week I can find some time and fix that part.

Is there something in common between the machines where that has happened to you? Windows version, background programs, updates, Windows execution mode, drivers, etc.

[Namiga]

Hello.
I'm testing it on the woman's laptop, which has already been updated to Bitdefender 2018.... and, by God, this year's antivirus is a blessing.

I don't know why it detects it as malware... and there's no way...

I had to disable everything (of the antivirus), add the folder where the benchmark is and give it to safe applications...
A good while

but it's already ok

[krampak]

I don't know what's wrong with that 4405U, I installed a Windows 7 (it had Windows 10) and without antivirus or anything, it keeps getting really bad ratios.

[cobito]

@krampak Could it be a bad cooling?

By the way, in the coming days I will start working on version 1.1.0 of the test bench. They are still just brushstrokes, but if everything goes well, the blocking problems will be solved and additional information will be added such as details about the RAM (pending timings) and some details about the motherboard. This will allow you to compare something as relevant as the chipsets.

Stay tuned to your screens.

EDITED: A moment, I am seeing that the results of your 4405U are actually fine but the page is showing it as erroneous. It is a micro with 2 cores and HT so the mmt ratio must be greater than 0.5, which it meets. I will see.

EDITED 2: It is already corrected. You have a few results that have turned green.

[Xevipiu]

Much better, still doesn't detect the real speed, but at least with the LOG you know it's not hanging
https://bm.hardlimit.com/signature.php?bm=b474fb6b6363fc62b29107ef8ad7fd3b535&style=1 real speed at 4.004mhz both under load and idle

[cobito]

@xevipiu The measured speed is always at full load of a single core.

PD: Version 1.1 is about to come out of the oven.

[cobito]

Well gentlemen, version 1.1.0 is now available. In broad terms, it now detects the brand and model of the motherboard, some details of the RAM in addition to having expanded the information of the processor. The detection of frequency in the program has also been improved although it is still measured since the one displayed does not always coincide with the real one.

The detection of physical cores is still problematic in old models but that of threads is correct (here without any problems). In addition, the frequency of the memory is not detected with total precision, so those who have a small OC will probably not see the real frequency.

In principle, the blocks in test#4 should disappear and the maximum execution time of all tests should be less than two and a half minutes.

There have also been changes in the page that shows the results where now the hardware on which the test suite has been executed appears. Here is an example.

The amount of changes is enormous (although apparently they are not too many), so the failures must be there. As always, I appreciate it when you inform me of the strange things you see.

[krampak]

I like that now the final result comes out when the test is finished. By the way, this has been happening for a while but I hadn't commented on it, the texts overlap from 4 threads when it goes from "waiting" to "test":

Mmmm it won't let me paste an image (before it was automatically uploaded when pasting) have you changed something?

[cobito]

@krampak You should be able to paste like before and also drag images. I will look into it anyway.

I haven't done much testing with more than 4 threads but I think I know what you mean. There may be some major interface changes coming up soon so I'll leave that on hold to see what I decide.

[Xevipiu]

Los modernos tambien

[krampak]

I have no way with the newly formatted HP's (Windows 10 Pro), now with an i5 7500 the ratio never reaches me 0.98 or 0.99, it stays at 0.95 or 0.96. Only Windows Defender is installed which comes with Windows but I have deactivated the real-time protection, I don't know what to do

[Xevipiu]

Disable everything from the logs

[cobito]

@xevipiu What's wrong? The frequency measurement has an error < 3% and if we trust what cpuz says, the error in your case is 1.4%.

@krampak in the task manager do you see any process consuming CPU while the benchmark is running?

[krampak]

@cobito Yes, the Defender's antimalware... even though I have real-time scanning disabled, when it gets to the fourth test, it triggers for a few seconds, I don't know why.

[cobito]

@krampak a few days ago I was reading about this and the problem is widespread among small software development companies. I will study the possibility of signing the program that they say improves the situation.

I will also see if the limit of 0.97 is too strict.

[cobito]

It seems that the program as a whole is detected as a trojan in some antivirus. But the main problem is with Windows Defender. @krampak I am going to report the false positives to Microsoft but they are asking for the "detection name" that is found in the Windows Defender history. If you have it, can you pass it on to me?

I am going to install Windows 10 so that I can do tests on my own.

[cobito]

The new Hardlimit signatures are now available, where individual results of each test have been removed and only the totals have been left. In addition, additional hardware information as well as the ranking position is shown.

If you finish first, second or third in single and/or multi-threaded, a gold, silver or bronze medal will appear next to the ranking.

[cobito]

I've been investigating the issue with antivirus software. On virustotal, sometimes there are more detections and sometimes fewer (the day before yesterday there were 8 and today 11).

I've tested Windows 10 and indeed the "Antimalware Service Executable" process belonging to Windows Defender is a headache. However, Windows Defender itself finds all the test suite components as harmless. I've read that it's normal for that process to consume a lot of CPU normally, running any program so I can't do much about it.

I imagine that this problem can be reproduced with other test suites. The issue is that in HLBM there are "too many" mechanisms to detect if the test has been performed under adequate conditions, which other programs in the field don't have and that's why they apparently don't have a problem.

I've also checked that when the test suite is run for the second time, the malware detector is much less aggressive.

The reason its activity is seen in test 4 is not because of test 4 itself, but because after running that test, "hlbm-detect" is launched, which is meant to detect the system's hardware. It's a very different process from "hlbm-core" and that's probably why it wants to analyze it. However, this wouldn't affect the results since at that point, the test suite itself has finished.

I understand that in Windows 10 it's going to be difficult to reach the 0.97 (0.5) MMT ratio so in the next revision of the program, the Windows version will be added to the rest of the information sent. In case it's Windows 10, the requirements to consider a result valid will be lowered.

Until then I can't do anything more. It's a problem of Windows 10 and its implementation of Windows defender, not of HLBM or a false positive.