Hardlimit test bank

[jordiqui]

In the next version, there might be the possibility of having the MPi cluster option for windows, MPHI for linux. Just out of curiosity, I think we could do a test by connecting nodes between a few.

[cobito]

@jordiqui I'll keep it in mind. Keep in mind that the program's development cycle is much slower than that of the central, mainly because I have to buy a code certificate every time I modify the program and it lasts for a year, so I accumulate change notes to apply them all at once when the time comes.

In the short term, only changes in the central are planned.

I say this so that you don't think I ignore it.

[jordiqui]

@cobito Tranquilo, creo que con estos bichos estaré unos años. Los programadores de verdad que tenéis mérito. No sabía por otra parte, esto que has comentado de las modificaciones del código. Por otra parte, no sé lo que tardarán en llegarme los e5 2697 v2, creo que de la serie e5 XXXX v2 son los más potentes. Añadiremos más variedad al banco de pruebas. Un abrazo.

[jordiqui]

Having passed the three tests with the e5 2697 v2 and easily surpassing the 2680, 2660, and 2651, I would say that this generation offers the best performance.

[cobito]

@jordiqui The closest ones are the 2697 and the 2680. Here it is curious that although the E5-2680 has a higher frequency than the 2697, the latter beats it in the single-thread test. It seems that the 5MB difference in the L3 cache compensates for all that. In any case, the single-thread performance is practically identical for both.

The appreciable difference is in multi-threading, where the 2x12 cores of the 2697 outperform the 2x10 cores of the 2680 by 14%. In this case, it seems that something is affecting the memory performance of the 2697 despite its superior cache: it may be a penalty for the 4 extra threads or that the memory configuration is different in each option.

[jordiqui]

@cobito said in Hardlimit test bench:

@jordiqui The closest ones are the 2697 and the 2680. Here it is curious that although the E5-2680 has a higher frequency than the 2697, the latter beats it in the single-thread test. It seems that the 5MB difference in the L3 cache compensates for all that. In any case, the single-thread performance is practically identical for both.

The appreciable difference is in multi-threading, where the 2x12 cores of the 2697 outperform the 2x10 cores of the 2680 by 14%. In this case, it seems that something is affecting the memory performance of the 2697 despite its superior cache: it may be a penalty for the 4 extra threads or that the memory configuration is different in each option.

The ram is identical pc8500 333 mhz, 533, no more. The only thing that has changed and I haven't commented on is the power supplies. Before they went with 250w, now 750w. I repeated the test with the new power supplies and it improved a bit compared to the previous test. Since we are here and not going to another thread, a rack cabinet arrived that must be for midgets and I gave them the exact model name, dimensions, etc., so nothing, that as long as they don't remove the front walls (glass door and the back there's no way to put anything in). And the store is known to all of us who are here. The truth is that with age I've told them that there's no problem, they'll change it for one that I can put them in and that's it, but this has been happening to me for 20 years and I was already nervous.

[cobito]

@jordiqui I was talking about the RAM because I saw that in the results with the E5-2697 v2, the memory indeed goes to 1067MHz (the PC3-8500 you mentioned), but there are several validations with the E5-2680 v2 where the memory was measured at 1600MHz, so probably that's why the memory average is better on the 2680. It seems that those at 1600MHz were all done in February.

[jordiqui]

@cobito said in Hardlimit test bench:

@jordiqui I mentioned the RAM because I saw that in the results with the E5-2697 v2, the memories indeed go to 1067MHz (the PC3-8500 you mention), but there are several validations with the E5-2680 v2 where the memories have been measured at 1600MHz, so probably that's why the memory average comes out better in the 2680. It seems that those at 1600MHz were all done in February.

True, because the ones I have of 8 gb per module are at 1600.

[Xevipiu]

I have posted new results with a new micro, the new Intel Ultra 285k
Note that it is interesting

[cobito]

@Xevipiu Gracias.

I have just added the model to the database. The Ryzen 9 9950X's gold medal didn't last long.

The results will be processed in a little while.

[Xevipiu]

@Xevipiu said in Hardlimit test bench:

What I miss would be a process LOG interface, so the wait wouldn't be so boring

I know, I don't even let it get first, I've had this micro under NDA for months

I'll pass the test again, with some significant changes

[url=https://bm.hardlimit.com/result.php?bm=5daf137507855524d732ea75e57622093488] [img]https://bm.hardlimit.com/screenshot.php?bm=5daf137507855524d732ea75e57622093488&style=1[/img] [/url]

Can you record the maximum temperature reached?

[cobito]

@Xevipiu Note taken on the log. The temperature thing is going to be complicated because it's a very hardware-dependent issue and the program is designed for minimal maintenance. Anyway, it will be looked into during the next review.

[cobito]

Well, this type of maintenance is something I've been doing so far, but I'm going to try to announce them so that it can be seen that the test bench is still receiving support.

In essence, the update policy consists of the fact that if a new processor model is received, it is added to the database immediately (on the same day, whenever possible) and is announced the next day on the front page, as long as it is a relatively recent model (a day is expected because the result has to be processed).

Memories, motherboards and computers have minimal priority and new models are added to the database every 2 or 3 weeks and the idea is to announce them in this thread from now on.

12 new RAM memory models have been added and another unidentified model has been received. In total we have 813, although some (very few) are repeated and there are also unidentified entries (even fewer than repeated ones).

11 computers (desktops and laptops), 4 motherboards and a portable console from MSI have also been added, totaling 1013 entries.

And to complete the statistics, we have 599 different processor models (here there are no repeated or unidentified ones).

[krampak]

@cobito said in Hardlimit Test Bank:

And to complete the statistics, we have 599 different processor models (there are no duplicates or unidentified ones here).

What madness!

[cobito]

In this batch, 30 RAM memory models, 12 computers, 7 motherboards, and a portable console from MSI have been added. In addition, during this month and a half, 12 CPU models have been added, bringing the total to 611.

[cobito]

With a couple more batches, you could almost get a monthly average.

This time, the following have been added:

• 32 models of RAM memory.
• 11 motherboards.
• 21 computers (almost all laptops with a couple of NUCs and a few desktops).
• A Samsung ARM tablet emulating x86 (no precise information about its hardware).
• 13 processors totaling 624 unique models.

[cobito]

We are launching the course (and season) with:

• 40 new memory models.
• 25 computers, with our first Asus NUC which, as you know, is the manufacturer now in charge of the mini PCs that Intel used to manufacture.
• 19 motherboards.
• 25 processors up to a total of 649 unique models.

[cobito]

New batch:

• 20 new models of RAM memory.
• 9 computers (1 desktop and 8 laptops).
• 12 motherboards.
• 3 processors up to a total of 652.

[Xevipiu]

Well, I think the Xeon 1680v2 is the most powerful Ivy Bridge out there, because of its speed potential with an unlocked multiplier

[cobito]

@Xevipiu From the Ivy Bridge we have, it is in 6th position in multithreading at stock frequency (the first in the list yields triple). Perhaps in single-threading; the highest frequency at which you have passed it is at about 4780MHz which, indeed, exceeds the first in the ranking of the Ivy Bridge.

The same would have to be put in the rankings also the results of record OCs apart from the stock yields. I'll make a note of it

By the way, a month without putting the pending motherboards and memories. I'll make a note of it too.