Hardlimit test bank

[krampak]

@cobito said in Banco de pruebas Hardlimit:

@krampak brings us back a couple of fairly recent processors, and both are very similar.

The first one is an i5-8500; a Coffe Lake from 6 months ago. It has 6 cores without HT and runs at 4.1Ghz. The most similar is another recent CPU (though not as recent): the i7-7700. Basically, they have the same TDP and go to almost the same frequency. But the Coffee Lake has 6 bare cores and the Kaby Lake has 4 with HT. In essence, the evolutionary improvement seems to be in AVX2 where the i7-7700 seems to puncture against the i5-8500 only in multithreading. Probably because that performance per extra MHz is getting it thanks to HT and this one doesn't do as well with more complex repertoires.

The other is an i5-8500T, the low-power twin brother of the previous one. Comparatively speaking, with a TDP of 30W less and a frequency reduced by 600Mhz, the performance per MHz is almost the same in both (same architecture and same cache) while the efficiency of the T series surpasses the i5-8500 by quite a bit.

I think I remember that in the 8500T it gave me a better result in SSE3 than in AVX, can that be? I repeated it several times and it came out the same.

[Xevipiu]

Hello again, I was able to make a symbolic entry of a 9900k, but it gives me an error with the entry of the 7980XE.

[cobito]

@xevipiu I'm going to have to set aside a few days to get my hands on the test bench :man_facepalming_tone1: Maybe I'll have to do something with the comparator, because this can't be.

You yourself reported that bug a while ago. The problem is that there is a limit of 32 threads. If you run those 32 threads (or 18 even, to evaluate each core separately), it should work.

By the way, thanks for continuing to feed the database.

Here is the datasheet for the i9-9900K that I see you've overclocked to 700MHz. Since it's overclocked, there are no comparisons, but doing a quick calculation, it would be number 1 of all the micros registered at stock frequency. I'm surprised that with that number of cores, that frequency and offering that performance, it has a TDP of "only" 95W.

[Xevipiu]

@cobito said in Hardlimit test bench:

@xevipiu I'm going to have to reserve a few days to tinker with the test bench :man_facepalming_tone1: Maybe I have to do something with the comparator, because this can't be right.

That bug was reported by you a long time ago. The problem is that there is a limit of 32 threads. If you run those 32 threads (or 18 even if it's to evaluate each core separately), it should work.

By the way, thanks for continuing to feed the database.

Here is the datasheet for the i9-9900K that I see you've overclocked by 700MHz. Since it's overclocked, there are no comparisons, but doing a quick calculation, it would be number 1 of all the micros registered at stock frequency. I'm surprised that with that number of cores, that frequency and offering that performance, it has a TDP of "only" 95W.

You will never get the real TDP. To start with, Intel gives 95w in its technical sheet as a reference, but in truth this amount can vary a lot depending on the base plate or the same VID of the processor.

Yes, that 9900 has delid and lappind on the same DIE

[Xevipiu]

I have already posted the results of the 7980XE at only 4.5Ghz capped at only 16 cores + HT

[cobito]

@xevipiu said in Hardlimit test bench:

I have already posted the results of the 7980XE at only 4.5Ghz capped at only 16 cores + HT

When you say you have capped at 16 cores + ht do you mean that in the test bench you have manually configured 32 threads or that you have disabled 2 cores of the processor?

In any case, @Xevipiu has brought us a beast born a year and a half ago. And with this, it breaks the million-point barrier for the first time.

It is the i9-7980XE, an 18-core, 36-thread Skylake with more than 1Mb of L1 cache, 18 megs of L2 cache and 24 megs of L3 cache. With this obscene amount of resources, at stock speed it has reached 1,200,000 points in multithread, which is almost triple that of the second in the queue.

In single-thread its performance seems ridiculous in comparison, but the reality is that it is in the top5 of all registered CPUs, very close to the first which is the i7-8700k.

The comparison is impossible except in single-thread and as I said, in this case it is one of the most powerful registered.

[Xevipiu]

Correct, I have disabled two of the physical cores that give me more problems

Today I received the 40p hole to be able to lap the DIE, as it has a convex surface of the DIE, no matter how much I do the delid, there is little improvement in temperatures, especially I have two cores that spike in temperature.

By the way, the test passed at 4.65Ghz at 1.23v with two of the cores at 102º (at 105º the thermal protector acts)

[b]Delid Direct[/b]

[i][b]x299[/b][/i]
[center][img]https://i.postimg.cc/nc6C7199/1.jpg[/img][/center]

[center][img]https://i.postimg.cc/v8KBFvz2/2.jpg[/img][/center]

[center][img]https://i.postimg.cc/W1L1tbxM/3.jpg[/img][/center]

[center][img]https://i.postimg.cc/FRR9wfJQ/4.jpg[/img][/center]

[center][img]https://i.postimg.cc/SxhY3rQF/5.jpg[/img][/center]

[center][img]https://i.postimg.cc/bYSGvMsf/6.jpg[/img][/center]

[b][i]For the 4/5/6/7/8/9 series [/i][/b]

[center][img]https://i.postimg.cc/pV40T9Vs/MSI-direct.jpg[/img][/center]

[center][img]https://i.postimg.cc/JnyNVqXM/Hielo-CPU-Opener-Protector-Delid-Die-para-LGA115X-para-Intel-CPU.jpg[/img][/center]

[url=https://es.aliexpress.com/item/Hielo-CPU-abridor-abierto-Protector-de-la-cubierta-de-Delid-morir-para-LGA115X-para-Intel-CPU/32858889426.html?cv=47843&af=358162&aff_platform=aaf&mall_affr=pr3&cpt=1542529725642&afref=&sk=VnYZvQVf&aff_trace_key=86aa289337064a34a537bd7912d68b2c-1542529725642-07505-VnYZvQVf&dp=d4c6ba8922752d89b0d23ffbe9212627&terminal_id=9e40d863efaf4ff0af0090ace35179df]Direct Direct on Aliexpress 4/5/6/7/8/9 series[/url]
[url=https://www.caseking.de/en/der8auer-skylake-x-direct-die-frame-fsd8-021.html]Skylake-X Direct Die Frame[/url]
[url=https://es.aliexpress.com/store/product/Delid-Die-Guard-Cover-Protector-IceMan-Cooler-Intel-LGA-2066-CPU-X299-CPU-Opener-Support-7820/3518027_32884067927.html?spm=a219c.search0604.3.10.68ac5e0a0x4nU2&ws_ab_test=searchweb0_0,searchweb201602_4_10065_10068_10547_319_10891_317_10548_10696_10924_453_10084_454_10083_10927_10618_10920_10921_10922_10307_10820_10301_10821_10303_537_536_10059_10884_10887_100031_321_322_10103_5727015_5727515-5727015,searchweb201603_51,ppcSwitch_0&algo_expid=43e16902-b123-49d1-936d-4bf052793d82-1&algo_pvid=43e16902-b123-49d1-936d-4bf052793d82]Aliexpress Direct Die Frame[/url]
..

[b]Lapping DIE[/b]

[img]https://i.postimg.cc/wBg4pSxH/portada.jpg[/img]

[center][img]https://i.postimg.cc/bwTWszBd/distancia-1.jpg[/img][/center]

[center][img]https://i.postimg.cc/26fZMDCF/lapping-2-mini.jpg[/img][/center]

[center][img]https://i.postimg.cc/KvnvpMRV/Laping-3.jpg[/img][/center]

[center][img]https://i.postimg.cc/T3f2z3bP/lapping-3-extender.jpg[/img][/center]

[center][img]https://i.postimg.cc/mk7Rnv2M/lapping-4.jpg[/img][/center]

[center][img]https://i.postimg.cc/m2r88QXZ/lappng-5.jpg[/img][/center]

[center][img]https://i.postimg.cc/T13qN6cN/lapping-6.jpg[/img][/center]

[center][img]https://i.postimg.cc/x13GQ8Fp/distancia-2.jpg[/img][/center]

[krampak]

Do you still have the 9900k @Xevipiu? it would be good to pass the test to the serial frequency so that it appears in the table

[Xevipiu]

@krampak said in Hardlimit Test Bench:

Do you still have the 9900k @Xevipiu? it would be good to run the test at the stock frequency so it appears in the table

Yes, it's the one I'm posting now, then I'll post it at stock

Tomorrow I'll do the lapping on the 7980XE

Now I'm going to deal with the paperwork for my new GT Turbo Phase1 xD

[Xevipiu]

I have already posted the results with the processor fully in stock

[cobito]

@xevipiu The truth is that they are interesting results. Comparatively speaking, the closest processor found is the Ryzen 7 2700X. As usual, AMD scores in AVX2 but at the lowest mode, Zen has a slight advantage over Coffee Lake.

The base frequency of both is similar as well as the cache, the number of cores and the HT/SMT (although, the Ryzen has double the L2 cache than the i9). The 5GHz boost frequency doesn't seem to help much for Intel's micro compared to AMD's 4.3Ghz since the single-thread performance differences are not very noticeable. In fact, in AVX2 those extra 700MHz are doing absolutely nothing.

For optimized software, the Ryzen gets 25% less performance so in games and other, the i9-9900K is clearly superior to the 2700X. Here it is important to know what the micro will be used for. If it's for a server, for example, Zen is probably the best option. If it's for gaming, video editing, rendering... Coffee Lake is superior without a doubt.

Considering that the i9 costs double the Ryzen 7, it seems that Intel hasn't come out with a very competitive CPU this time.

[Xevipiu]

The differences are in the architecture, the AMDs go up to 4.3/4.35Ghz is their physical barrier, while the Intels have no limit :women_wrestling:

Today I have raised the 7980XE again after Lapping on its DIE, I have lost some small transistor, but it works perfectly


[Xevipiu]

It will surely go up even more

[cobito]

@krampak brings us a couple of new CPUs to our database tested in all modes: an i3 Skylake and an i5 from 6 years ago.

The first one is a Core i3-6100T released in 2015. It has two cores with HT, 3 Mb of cache and 35W TDP. In multicore it has a performance similar to an i5-7200U, which although it is of a "higher" category, is actually a U series (castrated in all ways to contain consumption). In single core it approaches a portable Haswell although this time from a higher range (for real): an i7-4720HQ.

The second one is an i5-3317U. Considering that this is starting to be a mature CPU, not much can be expected from it. With its 2 cores without HT, in the end it is comparable to CPUs of the same branch: portable processors of more or less a couple of generations of difference.

[Xevipiu]

Hello, well I have almost 200 micros of all classes, I am already uploading little by little to fill the table Hello holidays!

[cobito]

Perfect @xevipiu, that will motivate me a lot. Thanks for the contributions!

[Xevipiu]

We have uploaded the first results of a 3570k pata negra, at only 5Ghz at 1.33v

[Xevipiu]

The joot has not had enough! It heats up like crap, which doesn't even have lapping, and even less delid

[Xevipiu]

Another one, a very decent 3770k for watercooling

[cobito]

A couple of days ago, @krampak brought us a micro from AMD that will be 8 years old next summer. It's a portable Bobcat APU (intended more for netbooks).

The processor is not great. The closest thing I've found is a Core Duo (without the 2) that I haven't even found a release date for. In short, it has two cores at 1.6GHz with a relatively high TDP of 18W.

And I'll stop there because the truth is I'm starting to write this with a certain animosity towards that architecture. I have a AMD C-60 (the little brother of the E-450) that are basically identical only that the C series runs a few MHz slower. Speaking quickly and clearly, it has been the worst processor I have ever tested. The netbook that comes with it was supposed to replace an Atom N450 (considerably older and with only one core) and although in certain places it was said that it increased performance by 50% compared to the Atom while reducing consumption, the reality is that its performance is tedious. And that's even though this PC has been used exclusively to handle a fairly simple Calc sheet and the web browser on Linux.

The curious thing (and it's something that I find quite annoying) is that on Windows 7 it works reasonably (within what is possible if compared to Linux). But on Linux it is completely horrible. Now that's true, the battery lasted more than double.