Question Low compression score 7-Zip multicore 14600k ?

[diango]

Hello folks,
Problem with very low performance of multicore 7-Zip bench
TUF GAMING B760-PLUS WIFI D4
Proc: 14600k, DDR4 3600 Mhz, 64GB

Std settings, tragically low performance of around 83600 KB/s compression, 32 MB dictionary, 20 threads.

It does not matter which version of 7z the differences are marginal, checked also on latest v24.05
Here is the result more than 50% faster, is that possible that mobo gives +50% to benchmark?!
https://www.techpowerup.com/review/intel-core-i5-14600k/14.html

I don't know maybe everything is ok and this result is just so bad....
Default settings as I also tried to take off the PL1 and PL2 limits(both 181w) , but that didn't do anything. played with other settings not helped 1-3% difference only.
Temps are good max 75 celsius
Is this possible that B760 or this motherboard is so weak?

Can some of you 13600k or 14600k owner please (no matter which mainboard) post their result,?

 

[Ralston18]

Update your post to include full system hardware specs and OS information.

Include PSU and disk drive(s).

32 GB RAM - configuration = 2 x 16 GB, 4 x 8 GB?

And, to be sure that I understand the posted compression issue correctly:

Your system is performing at 83600.

However, your compression and decompression expections via 7zip bench data are in the 131000-153000 ranges per the link. Is that correct?

What other settings did you change? E.g. Dictionary size? What before and after values?

 

[TerryLaze]

Std settings, tragically low performance of around 83600 KB/s compression, 32 MB dictionary, 20 threads.

The results in the page are in MIPS and not in kb/s, your result of ~130.000 Gips is only a little below the 138.000 in the review, you basically match the stock settings of 131.000.
Although the review mistakenly calls it MIPS which is a full tier less than GIPS

 

[diango]

@Ralston18, I wrote 64GB of RAM not 32GB, probably you've missed with 32MB dictonary size of 7z...
Ram: 4x16GB - KHX3466C16D4/16GX, XMP-3466
PSU: STRAIGHT POWER 11 750W
SSD: 960 PRO SAMSUNG
OS: (Microsoft Windows [Version 10.0.19045.4412]) PRO
When comes to BIOS settings I have played various settings like Asus performance enhancement, or manualy wattage limits, but as said changes were only marginal up to 3%.

I mean compression - here is huge diff.
decompression difference is only lie 5-7% lower which is perfectly acceptable for me not worth to mention even.
Problem is compression, when comes to GIPS techpowerup have about 131 GIPS (131614 MIPS)
My result is 95 GIPS. Ok it's not 50%, but still huge difference almost 40% lower.
Would be perfect if any of you could share result, the best without any OC, just std 13600k or 14600k

ps. I'm testing with 20 threads and 32MB (default) dictonary size which I believe is default.

 

[diango]

The results in the page are in MIPS and not in kb/s, your result of ~130.000 Gips is only a little below the 138.000 in the review, you basically match the stock settings of 131.000.
Although the review mistakenly calls it MIPS which is a full tier less than GIPS

Yes I've noticed it already thanks, but you are looking at decompression, which is right, but compression (above on screenhot) is still 38% higher on review.

ps. I have tried everything including Safe Mode in Windows(results were very similar)
at same time Cinebench multicore result is comparable (up to 5% diff).
Almost 40% lower result on compression looks so fishy,( I have tried 7z 23 and latest 24.05 64 bit)

 

[Ralston18]

Going out of my comfort zone now (full disclosure) - does changing the dictionary size make any difference?

Is dictionary size controlled by the two configuration settings just below "Test wydajnosci"?

Not sure about the meanings/labling......

= = = =

What about file types and sizes? What file type and size is being used for the testing?

Can the "Test wydajnosci" window be shown in English?

 

[diango]

does changing the dictionary size make any difference?

Yes, but marginally few% tried from 16 to 64MB, but nonsense to talk about it since current and past few 7zip version dictonary default size is the same 32MB. What do you mean by file types and sizes? I guess you don't know 7-zip at all?
It's just std window from menu Tools > Benchmark. I use their official benchmark function.

 

[uWebb429]

@diango
B760 boards automatically enable Undervolt Protection and IA-CEP current protection. Check the BIOS to see if you can change to the early 0x104 microcode which came out before those protection mechanisms existed. Many Asus motherboards have this option.

Windows 10 does not officially support Intel 12th Gen and newer CPUs that have P and E cores. That can cause reduced performance in some benchmarks.

 

[Ralston18]

I meant what file types and sizes are being compressed?

Larger files and some file types (e.g., video, text) will compress more). Not sure about end results on performance....

In any case, as I understand your post, you have tested performance with different values than the default 32 MB Dictionary.

Basically looking or otherwise trying to find ways to narrow down the problem. Directly or by elimination.

I was hoping to see the window with your results presented in English.

 

[diango]

I meant what file types and sizes are being compressed?

Larger files and some file types (e.g., video, text) will compress more). Not sure about end results on performance....

In any case, as I understand your post, you have tested performance with different values than the default 32 MB Dictionary.

Basically looking or otherwise trying to find ways to narrow down the problem. Directly or by elimination.

I was hoping to see the window with your results presented in English.

I know what are you talking about and the same I don't know I'm talking about Benchmark function of 7z, so it compress some either random data or have some internal algo, it's not real files compression test.
Results can be in even in Japanse too, just install 7z and open this windows to see in eaxctly the same positions translations, to see it, we are talking about pure numbers so lang is not important, anyway thanks for notice next time I will remember to post with eng.

 

[diango]

@diango
B760 boards automatically enable Undervolt Protection and IA-CEP current protection. Check the BIOS to see if you can change to the early 0x104 microcode which came out before those protection mechanisms existed. Many Asus motherboards have this option.

Windows 10 does not officially support Intel 12th Gen and newer CPUs that have P and E cores. That can cause reduced performance in some benchmarks.

Thanks, that was my first suspicion to install Win11 (at techpowerup test 11 is used) first and check as I remember I have already checked those settings including 0x104 microcode instead of "current" (applied the latest).
Maybe I will be forced finally to install Win11(hate it)

My biggest concern is that my mainboard is piece of ... and z790 or z690 chipset are needed (however in theory nonsense)

 

[uWebb429]

@diango
Try running ThrottleStop 9.6

ThrottleStop 9.6 Download

ThrottleStop is a small application designed to monitor for and correct the three main types of CPU throttling that are being used on many lapto

www.techpowerup.com

Watch the Limit Reasons window while you are testing. Are any throttling reasons lighting up red? Some motherboards limit the cache speed equal to the base frequency of the CPU which can reduce benchmark performance. Post some ThrottleStop screenshots including the FIVR and TPL windows. The FIVR monitoring table will show the cache ratio while testing.

Can your CPU maintain full speed during a Cinebench run? There should be zero throttling of any kind during Cinebench.

 

[diango]

@uWebb429
7z from Windows 11(English) only like 2 % better(margin level error) so W11 did not helped

TPL:

Limit Reasons: (EDP OTHER@RING - BLINKING FEW TIMES WHILE TESTING 7z)

While testing Cinebench R23 multi core VR CURRENT@ CORE blinking and EDP OTHER@RING)
Despite than over 24000 points which is perfectly good for this CPU comparing to other results.
Only 7 ZIP Compression is very low.

FIVR:

 

[35below0]

Bench results:

https://imgur.com/GagETUm

View: https://i.imgur.com/GagETUm.png


13600K - z790 UD - 2x32Gb 4800/CL40 RAM


Results are comparable in decompression. Something it definetly off with your compression results. Very strange as all other tests are fine.


Look at what HWMonitor reports, in addition to the Throttlestop suggestion above.

 

[diango]

So your 13600k is 25% faster at comperssion, I think that my new mainboard(or chipset) is just bad.
I have tried probably all settings and bios combinations including updating bios today to latest (not helped too)
Wonder if anyone with B760 chipset and 13600k or 14600k could post his result.

 

[diango]

Bench results:

https://imgur.com/GagETUm

View: https://i.imgur.com/GagETUm.png


13600K - z790 UD - 2x32Gb 4800/CL40 RAM


Results are comparable in decompression. Something it definetly off with your compression results. Very strange as all other tests are fine.


Look at what HWMonitor reports, in addition to the Throttlestop suggestion above.

Thanks.,so your 13600k is 25% faster at comperssion, I think that my new mainboard(or chipset) is just piece of shit.
Man..it's extremely strange, first time in my computer experience I see such stuff - nonsense.
Tred v 22, 23 and 24 of 7zip, two OSes 10 and 11. More nonsense when you add 24000 points Cinebench r23 result.
I have tried probably all settings and bios combinations including updating bios today to latest (not helped too)
Wonder if anyone with B760 chipset and 13600k or 14600k could post his result.
Looks like a failed comeback to Intel, or something with these cheap chipsets is wrong.
ps. I'm cooling all with NH-D15.
I bought 2 days ago both MB+CPU in shop brand new.

 

[35below0]

So your 13600k is 25% faster at comperssion, I think that my new mainboard(or chipset) is just bad.
I have tried probably all settings and bios combinations including updating bios today to latest (not helped too)
Wonder if anyone with B760 chipset and 13600k or 14600k could post his result.

You could ask 7zip forum.

If it is your mobo, then why are other test results good?
Also, are there any signs of throttling?
My CPU temp shoots up to ~81-88C

 

[diango]

You could ask 7zip forum.

If it is your mobo, then why are other test results good?
Also, are there any signs of throttling?
My CPU temp shoots up to ~81-88C

I don't think it's good idea to buzz at 7z forum, looks like strictly my HW problem(especially when checked few 7z editions and two Windows 10 and 11)

After 10 seconds run - HWMonitor shot

 

[35below0]

I don't think it's good idea to buzz at 7z forum, looks like strictly my HW problem(especially when checked few 7z editions and two Windows 10 and 11)

After 10 seconds run - HWMonitor shot

Almost identical to mine. 10W higher Package max in your case.

I have no idea why the compression result would be lower, but it doesn't seem like the CPU is doing anything different.
Similarly, i have no idea if a z motherboard would give you back those 40% back. How?


The only other difference is the VID (Max) max. Yours reads 1.467 V. For me it doesn't exceed 1.249 V.
Not sure what to make of that.

 

[diango]

VID max because I took of all limits, Now I have 1,232 as max while testing.
One more thing maybe not only 7z is lower, here latest other app:
Geekbench 6.3.0 for Windows AVX2
2950 and 15100 multi-core score,
Again multi sounds like lowered a bit? For now all indications are that the b760 is just plain crap
Anyway Geekbench is not fully realiable to me I've noticed it even do not raise temps too much, looks like most of their tests do not utilize CPU max

Looks like throttling becasue of bad voltage regulation in 7z, it starts from 118k MIPS and then in 1 second lowering to 100 and below.
Probably the only way is to buy any Z690 or Z790 and check....or just get back to AMD and forget about Intel ovens.

 

[35below0]

VID max because I took of all limits, Now I have 1,232 as max while testing.
One more thing maybe not only 7z is lower, here latest other app:
Geekbench 6.3.0 for Windows AVX2
2950 and 15100 multi-core score,
Again multi sounds like lowered a bit? For now all indications are that the b760 is just plain crap
Anyway Geekbench is not fully realiable to me I've noticed it even do not raise temps too much, looks like most of their tests do not utilize CPU max

Looks like throttling becasue of bad voltage regulation in 7z, it starts from 118k MIPS and then in 1 second lowering to 100 and below.
Probably the only way is to buy any Z690 or Z790 and check....or just get back to AMD and forget about Intel ovens.

Geekbench is not a perfect benchmark.

The b760 is a budget motherboard. Why don't you look for a z690 motherboard with features you like? It will cost less than a z790 motherboard.
Flash it before installing the 14600K!

Why is benchmark performance important to you? It seems two benchmarks are a little bit off but most other results are what they should be. Do you have a problem running the PC normaly?
If not, enjoy it.
I would like to help you figure out what is screwing up this compression test but i can't. I am curious. At the same time i would not want to get hung up over it. Seems it's 95% exactly what it should be.

 

[diango]

Geekbench is not a perfect benchmark.

The b760 is a budget motherboard. Why don't you look for a z690 motherboard with features you like? It will cost less than a z790 motherboard.
Flash it before installing the 14600K!

Why is benchmark performance important to you? It seems two benchmarks are a little bit off but most other results are what they should be. Do you have a problem running the PC normaly?
If not, enjoy it.
I would like to help you figure out what is screwing up this compression test but i can't. I am curious. At the same time i would not want to get hung up over it. Seems it's 95% exactly what it should be.

That's why even started benchmarking I like geeking sometimes with results etc. but not this time, compressing often with 7z and I've noticed on 1GB txt file I saw like hey WTF, it's only about 25% faster than my old 5600x 6 cores(also non OC)
I saw it with MB/s on same disk, then started to look closer and do benchmarking with 7z benchmark feature.
So however I'm fixated now on 7z compress benchmark, initially real world scenario brought me there (and it's confirmed with benchmark feature only)

I have another suspicion that maybe DD4 vs DDR5 make so big difference there?
If not chipset is so bad....too bad I need to buy MB to check it

 

[uWebb429]

While testing Cinebench R23 multi core VR CURRENT@ CORE blinking

That happens when a motherboard manufacturer uses voltage regulators that are not capable of fully powering a 14600K during a full load stress test. Boards with the B760 chipset are often times under designed to save a few pennies.

Turn on the ThrottleStop Log File option before running Cinebench. ThrottleStop can accurately track how much throttling this is causing. When done testing, exit ThrottleStop so it can finalize the log file. Copy and paste the log file data to www.pastebin.com

ThrottleStop also confirms that Undervolt Protection is enabled. With a Z series motherboard you should be able to turn this off. With the B760 you have to use the early microcode, 0x104. If you removed this safety feature, you could undervolt your CPU and drop full load CPU temperatures up to 15°C. Less power consumption would also help your CPU stay away from VR CURRENT throttling.

(EDP OTHER@RING - BLINKING FEW TIMES WHILE TESTING 7z

That shows the ring / cache speed is being throttled. This drops the cache speed down to the base frequency of the processor which is 3.5 GHz for a 14600K. A benchmark like 7z is highly dependent on the cache speed. This is likely the main reason why you are seeing a performance hit during the 7z benchmark.

Did you try watching the FIVR monitoring window while 7z is running? Watch the Cache Ratio. It might show rapid drops from 45 to 35 depending on the duration of the throttling. Some throttling can be happening hundreds of times per second so some monitoring software might not be able to catch what is going on internally. The throttling flags within the CPU that cause Limit Reasons to light up red are an accurate indication of throttling in progress.

 

[35below0]

I have another suspicion that maybe DD4 vs DDR5 make so big difference there?

I doubt it very highly. For example my RAM is 4800 Mhz CL40 as i wrote above. That is not much faster.

And yeah, i didn't mean to say "oh don't fixate on benchmarks", because something really is off. Weird how specific the problem is.
If all your benhcmark results sucked, at least we could blame something. But this...

Take a look at an AsRock Riptide or AsRock Extreme z690 boards.
~$125 + BIOS flash and you're off to the races. Keep your DDR4 as well.

 

[35below0]

That shows the ring / cache speed is being throttled. This drops the cache speed down to the base frequency of the processor which is 3.5 GHz for a 14600K. A benchmark like 7z is highly dependent on the cache speed. This is likely the main reason why you are seeing a performance hit during the 7z benchmark.

Thank you. That helps a lot!