Linpack Xtreme Released

[sp00n]

I mostly use Intel lately with no access to the latest Ryzen processors. I'll send you a beta build later to test if interested.

Yes, I'd be happy to test this, though I'm off to bed now.

 

[Octopuss]

I mostly use Intel lately with no access to the latest Ryzen processors. I'll send you a beta build later to test if interested.

Can you update the bootable one too?

You should also update the TPU download links, they still have the old version.

 

[sp00n]

Do the Gflops seem unusually low? If so, it could be the same issue that I had with another Linpack test (LinX, IIRC)
(Where when I was using LinX, IIRC, it would randomly fail to fully load the CPU and I would get ridiculously low Gflops, until I restarted the test.)

I have now tried OCCT which also includes a Linpack test, and have actually noticed this behaviour there. When allocating the memory, it will load all cores 100%, but after that some cores are not fully loaded anymore, until the next round starts and it all starts over again. At least for its 2019 implementation, the 2012 one seems to work, at least for loading the cores. It uses much lower power though (probably missing AVX2 or something).

I also wanted to take a look at LinX, but it seems to have been discontinued for AMD due to the incompatibilities: https://translate.google.com/translate?sl=auto&tl=en&u=https://hwtips.tistory.com/1611

 

[RJARRRPCGP]

Well, in this case, I wasn't talking about where there is usually a pause for some seconds before going to the next round. (when the round is about to end)
It's when it fails to fully load the CPU and I have to close and relaunch it. One day, I was wondering why my core temps were lower than usual.

 

[Regeneration]

Thanks to @sp00n for beta testing, there are some improvements on Zen 2 and 3 CPUs in version 1.1.5.

Also updated bootable media, and such.

 

[CiTay]

For some reason, Microsoft's inbuilt Antivirus in Windows 10 keeps detecting LinpackXtreme_x32.exe as "Trojan:Win32/CryptInject!ml" or as "Trojan:Win32/Wacatac.DB!ml".

Anyone else have that problem?

 

[Regeneration]

For some reason, Microsoft's inbuilt Antivirus in Windows 10 keeps detecting LinpackXtreme_x32.exe as "Trojan:Win32/CryptInject!ml" or as "Trojan:Win32/Wacatac.DB!ml".

Anyone else have that problem?

Make sure to update to the latest definitions.

 

[CiTay]

Make sure to update to the latest definitions.

It seems that MSAV detected LinpackXtreme_x32.exe version 1.1.4 as a trojan, which is the version i kept, since initially it detected 1.1.5 as a trojan when it came out. Now i re-downloaded 1.1.5 and there's nothing detected anymore.

There are still some false positives though:

VirusTotal

VirusTotal

www.virustotal.com

VirusTotal

VirusTotal

www.virustotal.com

 

[Regeneration]

 

[CiTay]

Thanks, yeah it definitely seemed like a false positive from the start. Glad it's sorted now.

 

[freeagent]

I used Linpack Xtreme on my Intel systems for the last few years, before that I used IBT. When I moved to my new AMD system, Linpack Xtreme loaded my CPU running at 4400 to 60c.. I thought man these chips run cool lol. I tried 1.1.5 and yup, back in business making good heat and rockin the flops. Big difference from 60c fans on quiet to 80c fans working hard.. Nice..

 

[Regeneration]

Thanks, yeah it definitely seemed like a false positive from the start. Glad it's sorted now.

Didn't expect otherwise. After all...



The antivirus lists AutoKMS and keygen as the biggest computer threats in the world.

 

[Mussels]

Didn't expect otherwise. After all...

View attachment 184261

The antivirus lists AutoKMS and keygen as the biggest computer threats in the world.

keygens and windows cracks: the scourge of the digital seas

 

[Ferrum Master]

keygens and windows cracks: the scourge of the digital seas

It ain't a crack Mussels... activation server emulation is very far from it. A properly cracked software? I can recall some Adobe products, in my books they deserve it with that concept that basically you cannot own their product without that cloud software affinity. Even games are not that properly cracked on most parts to call it a clean job.

The fun fact is that mining trojan and that's a plague indeed on many levels no matter how we look at it.

I ran hard this this tool for a half a day... my old system is kicking the bucket slowly. OC has degraded. It didn't actually need more vcore voltage actually, but after tinkering few hours it ended up to be VCCSAA and it needed to be raised to be linpack stable, at least for a while... My Ryzen CPU is still on my way...

 

[Mussels]

Yarrrr, ye sounds like a pirate!
Time to walk the... thread? back on topic?

It's sad when OC's go bad and parts need to retire.

 

[TensorVortex]

Hi I have a question, I just built a PC with 10850k and Z490, mem 3600C16. CPU set to 1.4v mid LLC 5.0Ghz. (also tried default and other settings)

But my linpack is really low, Im getting ~270 GFLOPS, where I seen 10700k reach 470GFLOPS easy.

Im really confused what's going on, is my power limited? Im on Aorus Elite Z490. During my run of Linpack xtreme, I seen my POUT to max at 200w.

I attached an screenshot of my result and hwinfo.

 

[unclewebb]

I seen 10700k reach 470GFLOPS easy

Can you post a screenshot or a link with that information? I know LinX reports some really high GFlops.

http://imgur.com/XVWR35u

 

[Arctucas]

Hi I have a question, I just built a PC with 10850k and Z490, mem 3600C16. CPU set to 1.4v mid LLC 5.0Ghz. (also tried default and other settings)

But my linpack is really low, Im getting ~270 GFLOPS, where I seen 10700k reach 470GFLOPS easy.

Im really confused what's going on, is my power limited? Im on Aorus Elite Z490. During my run of Linpack xtreme, I seen my POUT to max at 200w.

I attached an screenshot of my result and hwinfo.

Is your tFAW at 16?

 

[unclewebb]

I found this run from a 9900K. In Linpack Xtreme it hit 558 GFlops. Looks like something has changed.

https://preview.redd.it/ud6t84fts8n51.png?width=1920&format=png&auto=webp&s=1f96c6e44219c2b60830bfb10b323f8ff2d0bc45

 

[Arctucas]

A run I did last October.

 

[unclewebb]

A run I did last October.

Can you try running the latest version, 1.15

I have a 10 core 10850K at 5.0 GHz and scores are way down compared to what people were getting with 8 core CPUs and version 1.13.

Time to check the TPU archives for an older version.

Edit - Sad times on the 10850K. The benchmark test only loads 10 of the 20 threads. Score looks terrible.

When I switched to the larger problem size, the scores got worse.
Edit - The multiplier drops down when AVX Offset is used so it must be using AVX instructions.
CPU runs at a steady 5000 MHz for the entire test so no throttling.

The numbers look a little better in a 20 Thread stress test but still well off from where they should be.

 

[CiTay]

Hi I have a question, I just built a PC with 10850k and Z490, mem 3600C16. CPU set to 1.4v mid LLC 5.0Ghz. (also tried default and other settings)

But my linpack is really low, Im getting ~270 GFLOPS, where I seen 10700k reach 470GFLOPS easy.

Im really confused what's going on, is my power limited? Im on Aorus Elite Z490. During my run of Linpack xtreme, I seen my POUT to max at 200w.

I attached an screenshot of my result and hwinfo.

In my opinion, with 1.4V, you are way past the range in which the CPU works efficiently. You are squeezing a rock at this point. Remember, the 10850K only got released because so few CPUs can reach the requirements for the 10900K during the binning. The yield rate for 5.3 GHz maximum turbo clocks (not all-core!) is supposedly very low.

The big downside you're getting is the massively increased power consumption.
And if you don't remove the PL1/PL2/Tau limits in the BIOS, and if you don't have excellent watercooling, you are hitting several limits at once with extreme AVX load.

Let's first look at the 10850K clock frequencies for three different types of load: https://www.techpowerup.com/review/intel-core-i9-10850k/20.html

Although advertised at 5.2 GHz max. boost, it will never actually hit the 5.2 GHz. Maybe the turbo budget doesn't allow it, or the TVB implementation is too conservative in the BIOS. I heard that ASUS released a new BIOS with TVB modifications to make the 10900K actually hit the 5.3 GHz, because when it initially was given to reviewers, they said they never observed 5.3 GHz. And since they wanted to avoid a Ryzen 3000 situation, which at the start also never hit the maximum boost, they made some modifications.

Anyway, this is only about single-core turbo, there's the only chance to hit 5.2 GHz. All-core 5.2 GHz would require massively more voltage.

Now, on to why overclocking the 10850K/10900K might not be that wise.

For calculating efficiency, we have to consider 1) performance, and 2) power consumption (which results in heat production). Ideally, you'd want to look at "energy spent per time spent per calculating task", but for a rough estimation, it is enough to compare benchmark results and the momentary power consumption during that load.

1) Performance:

So let's compare the 10900K at all-core 5.2 GHz (VCore 1.385V, first bar) vs. the 10900K with no PL1/PL2/Tau limits in the BIOS (second bar) vs. stock performance with limits in place (third bar).
We have 6616 vs. 6375 points in CineBench R20 multicore (which almost exclusively responds to CPU performance). This is a 3.7% increase in performance.


2) Power consumption:

Here is the big downside. The CPU gobbles up 384.7W at all-core 5.2 GHz. This is a 70.45% increase (!) over stock with all power limits removed.

For each single percent of performance improvement over stock, you basically have to take a 20% increase in power consumption. This is almost unheard of in previous CPU generations, and it demonstrates that Intel put this CPU right at the limit of efficiency by default. Going above it simply makes no sense. Like i said, Intel released the 10850K, supposedly because they don't even have enough yield from their wafers to bin enough dies as 10900K. But the 10850K is equally close to its limit, it's just a slightly lower binning (less handpicked CPU quality), so the limit is slightly lower. But Intel got 99% out of the CPU already.

By the way, these are the voltages they needed during OC:

Their cooling was a custom Alphacool water cooler with 360mm radiator.

Today, RAM OC is way more useful than CPU OC. And it won't lead to huge increases in power consumption and heat production. Overclocking the CPU isn't what it used to be, certainly not with the top-of-the-line CPUs. With the RAM, you can still squeeze a lot out of it. Even above 3600 MHz you still can get quite some additional bandwidth.


All that being said, i won't put so much focus on the GFLOPS value of Linpack Xtreme. It is not a benchmark. If you want to benchmark CPU speed, use something like Cinebench, it gives comparable numbers and almost only looks at CPU speed. If you want to see what happens during Linpack Xtreme that may cause your performance to be low, look closely at the CPU package power consumption, at the start and during a run, for example with AIDA64 which produces nice graphs. It will also corellate with temperature and even fan speeds. There you can see a power limit in effect.

Here is an example with a 9600K with the 95W power limit enabled in the BIOS. You can observe that it hits over 100W power consumption, i don't remember exactly, maybe 115W, then after PL1 hits, it goes down to the 95W limit. Visible from the temps...



...as well as the fans:





The Intel defaults, meaning, setting PL1, PL2 and Tau limits according to spec, are actually not hampering real-world performance much. It may show an impact in certain benchmarks, but they are fully loading every core to the maximum. This is sort of artificial. In normal use, for example games, even if they heavily use multithreading, you will see a much lower power consumption (package power) of the CPU, hence the limits don't really "limit" much there. And with other applications, you won't have extremely long full load on all cores.

Maximum multi-core AVX load is the most artificial you can get, that's why it doesn't make much sense to try to solve this. The actual problem is that Intel made the fastest CPU they could, by any means necessary. With the 10900K they released a CPU that, if you remove the limits, behaves like a factory-overclocked CPU, in power consumption as well as heat production. The 10850K is only a slightly toned down version of it. If you OC them, you're overclocking on top of what would've been an overclock with any previous CPU models. Without the Intel limits applied in the BIOS, they go way beyond where any of their previous CPU models have gone, except maybe the 9900K in a more tame form. That's why it can make sense to actually set those limits.

Do you get any kind of limiting in normal benchmarks which stress the CPU? And i mean long benchmarks like Cinebench R23 which you can set for more than a minute or so. Because TAU, the Turbo Time Parameter, is 56 seconds with those CPUs. Even with the limit set to 4096W in the BIOS, the EMWA (Exponentially Weighted Moving Average) might still cause a 250W limit to be in effect.

 

[Arctucas]

Can you try running the latest version, 1.15

I have a 10 core 10850K at 5.0 GHz and scores are way down compared to what people were getting with 8 core CPUs and version 1.13.

Time to check the TPU archives for an older version.

Edit - Sad times on the 10850K. The benchmark test only loads 10 of the 20 threads. Score looks terrible.

View attachment 188513

When I switched to the larger problem size, the scores got worse.
Edit - The multiplier drops down when AVX Offset is used so it must be using AVX instructions.
CPU runs at a steady 5000 MHz for the entire test so no throttling.

View attachment 188518

The numbers look a little better in a 20 Thread stress test but still well off from where they should be.

View attachment 188522

 

[CiTay]

As i said, it's not a benchmark, don't focus on the GFLOPS. It is not meant to produce a number that's comparable among different systems or settings.

If you have wildly lower results than what other people get with your CPU in actual benchmarks, that's a reason to worry. Then we can do a deep-dive as to what causes it. A too ambitious overclock for example can cause many unwanted things, yes, even lower performance. It's counter-intuitive, but it's been shown on the latest high-end CPUs of both Intel and AMD. They each have perfected their turbo algorithms to where you can hardly do better manually, and they each tend to run so closely at their limit that you can easily cause reason for throttling or limiting during OC.

 

[TensorVortex]

UPDATE:

I put 4x8gb samsung b die 3600 16-16-16-36 ram in the 10850k pc, and now linpack xtreme reports around 490Gflops.

So it's the ram but no idea why. anyway that was bothering me a lot yesterday, glad it's not my pc issue.

Now I'm going to try oc my ram see what it can do.

Thanks for all the help.

Is your tFAW at 16?

Hi sorry I been busy yesterday, my ram is 3600 16-19-19-39 Hynix.

In my opinion, with 1.4V, you are way past the range in which the CPU works efficiently. You are squeezing a rock at this point. Remember, the 10850K only got released because so few CPUs can reach the requirements for the 10900K during the binning. The yield rate for 5.3 GHz maximum turbo clocks (not all-core!) is supposedly very low.

The big downside you're getting is the massively increased power consumption.
And if you don't remove the PL1/PL2/Tau limits in the BIOS, and if you don't have excellent watercooling, you are hitting several limits at once with extreme AVX load.

View attachment 188521

Let's first look at the 10850K clock frequencies for three different types of load: https://www.techpowerup.com/review/intel-core-i9-10850k/20.html

Although advertised at 5.2 GHz max. boost, it will never actually hit the 5.2 GHz. Maybe the turbo budget doesn't allow it, or the TVB implementation is too conservative in the BIOS. I heard that ASUS released a new BIOS with TVB modifications to make the 10900K actually hit the 5.3 GHz, because when it initially was given to reviewers, they said they never observed 5.3 GHz. And since they wanted to avoid a Ryzen 3000 situation, which at the start also never hit the maximum boost, they made some modifications.

Anyway, this is only about single-core turbo, there's the only chance to hit 5.2 GHz. All-core 5.2 GHz would require massively more voltage.

Now, on to why overclocking the 10850K/10900K might not be that wise.

For calculating efficiency, we have to consider 1) performance, and 2) power consumption (which results in heat production). Ideally, you'd want to look at "energy spent per time spent per calculating task", but for a rough estimation, it is enough to compare benchmark results and the momentary power consumption during that load.

1) Performance:
View attachment 188516

So let's compare the 10900K at all-core 5.2 GHz (VCore 1.385V, first bar) vs. the 10900K with no PL1/PL2/Tau limits in the BIOS (second bar) vs. stock performance with limits in place (third bar).
We have 6616 vs. 6375 points in CineBench R20 multicore (which almost exclusively responds to CPU performance). This is a 3.7% increase in performance.


2) Power consumption:
View attachment 188515

Here is the big downside. The CPU gobbles up 384.7W at all-core 5.2 GHz. This is a 70.45% increase (!) over stock with all power limits removed.

For each single percent of performance improvement over stock, you basically have to take a 20% increase in power consumption. This is almost unheard of in previous CPU generations, and it demonstrates that Intel put this CPU right at the limit of efficiency by default. Going above it simply makes no sense. Like i said, Intel released the 10850K, supposedly because they don't even have enough yield from their wafers to bin enough dies as 10900K. But the 10850K is equally close to its limit, it's just a slightly lower binning (less handpicked CPU quality), so the limit is slightly lower. But Intel got 99% out of the CPU already.

By the way, these are the voltages they needed during OC:

View attachment 188514

Their cooling was a custom Alphacool water cooler with 360mm radiator.

Today, RAM OC is way more useful than CPU OC. And it won't lead to huge increases in power consumption and heat production. Overclocking the CPU isn't what it used to be, certainly not with the top-of-the-line CPUs. With the RAM, you can still squeeze a lot out of it. Even above 3600 MHz you still can get quite some additional bandwidth.


All that being said, i won't put so much focus on the GFLOPS value of Linpack Xtreme. It is not a benchmark. If you want to benchmark CPU speed, use something like Cinebench, it gives comparable numbers and almost only looks at CPU speed. If you want to see what happens during Linpack Xtreme that may cause your performance to be low, look closely at the CPU package power consumption, at the start and during a run, for example with AIDA64 which produces nice graphs. It will also corellate with temperature and even fan speeds. There you can see a power limit in effect.

Here is an example with a 9600K with the 95W power limit enabled in the BIOS. You can observe that it hits over 100W power consumption, i don't remember exactly, maybe 115W, then after PL1 hits, it goes down to the 95W limit. Visible from the temps...

View attachment 188519

...as well as the fans:

View attachment 188520



The Intel defaults, meaning, setting PL1, PL2 and Tau limits according to spec, are actually not hampering real-world performance much. It may show an impact in certain benchmarks, but they are fully loading every core to the maximum. This is sort of artificial. In normal use, for example games, even if they heavily use multithreading, you will see a much lower power consumption (package power) of the CPU, hence the limits don't really "limit" much there. And with other applications, you won't have extremely long full load on all cores.

Maximum multi-core AVX load is the most artificial you can get, that's why it doesn't make much sense to try to solve this. The actual problem is that Intel made the fastest CPU they could, by any means necessary. With the 10900K they released a CPU that, if you remove the limits, behaves like a factory-overclocked CPU, in power consumption as well as heat production. The 10850K is only a slightly toned down version of it. If you OC them, you're overclocking on top of what would've been an overclock with any previous CPU models. Without the Intel limits applied in the BIOS, they go way beyond where any of their previous CPU models have gone, except maybe the 9900K in a more tame form. That's why it can make sense to actually set those limits.

Do you get any kind of limiting in normal benchmarks which stress the CPU? And i mean long benchmarks like Cinebench R23 which you can set for more than a minute or so. Because TAU, the Turbo Time Parameter, is 56 seconds with those CPUs. Even with the limit set to 4096W in the BIOS, the EMWA (Exponentially Weighted Moving Average) might still cause a 250W limit to be in effect.

Hi Thanks for the info. Im at 1.4 mid LLC, so full load AVX2 is around 250w with 1.3v core. Still have no idea why my score is that low tho.