AMD vs Intel Single threading?

[humbug]

Precisely. This is the problem I've generally had with HT/SMT and why I would never get a 9900K as I find Intel HT to be worse implementation than Ryzen's SMT.
Could you run the test a few times to see if there is any variation?

Intel's SMT is less efficient than AMD's, Intel gains about 25% with SMT, AMD its 35%.

I will agree for this workload the 9700K is perfect as its 8 real cores, exactly the number DX0 uses.

Sure, i'll run it 10 times.

 

[humbug]

Interesting.

First eight 23 sec this time, last two 24 sec again...

 

[TNA]

Thats my 3900X. All settings at default except the generic settings needed for 1usmus powerplan to work, on air with a Noctua D15 and the 5y old 3600C16 ram at XMP. To beat it a 9000 series Intel CPU needs to overclock at 5.2 Ghz, whole top recorded speed below is 4.6Ghz on the core that was running the benchmark.

That score is a little under 4.4GHz, as I get higher than that at 4.4GHz. If it was truly doing 4.6GHz you would be getting around 53X points. This likely means your 4.6GHz is only for light loads.

 

[Panos]

That score is a little under 4.4GHz, as I get higher than that at 4.4GHz. If it was truly doing 4.6GHz you would be getting around 53X points. This likely means your 4.6GHz is only for light loads.

Is default to what ever is doing. Saw the thread spiking to 4.6 but let it running while washing the plates. I do not have the patience to wait all that time for single core test

Also NB was at 800 instead of 1800, somehow the 2.70 bios on the taichi needs to set manually the IF now, even at XMP. That wasn't the case previously.

 

[humbug]

544 at 4.676Ghz

Score 544: AMD Ryzen R5 3600 at 4.675Ghz, TNA

Is default to what ever is doing. Saw the thread spiking to 4.6 but let it running while washing the plates. I do not have the patience to wait all that time for single core test

Also NB was at 800 instead of 1800, somehow the 2.70 bios on the taichi needs to set manually the IF now, even at XMP. That wasn't the case previously.

I have gone off ASRock somewhat, the BIOS layout is bad but they have always been solid boards, to be fair mine is but reading around it seems they are slow getting updated BIOSes out and then #### it all up.

They need to invest in some good BIOS people.

 

[CAT-THE-FIFTH]

DxO batch RAW conversion exports are reasonably well multi-threaded,as there are reviews which compare recent CPUs together and the Ryzen CPUs do very well against the current Intel crop. So you can see the higher core count Intel and AMD CPUs beating the 8 core models:
http://www.comptoir-hardware.com/ar...70-a-ryzen-7-3700x-ryzen-9-3900x.html?start=9
https://news.mynavi.jp/article/20190707-854576/3

However,once you start mucking around with filters,these tend to prefer single threaded performance.

It turned out that the 9700K is ~12% faster at 5.2Ghz and at 5.3Ghz 18% faster, though at 5.3Ghz the VRM's on this Z370 are begging for mercy so I've got a Z390 board to drop in.

How are you get another 6% more with a 100MHZ clockspeed increase? Something is not right there!

But,this really puts things in perspective for me - I have a SFF PC,so it appears at stock any of these Intel CPUs won't actually be really quicker even if you are applying filters.

 

[TNA]

Is default to what ever is doing. Saw the thread spiking to 4.6 but let it running while washing the plates. I do not have the patience to wait all that time for single core test

Also NB was at 800 instead of 1800, somehow the 2.70 bios on the taichi needs to set manually the IF now, even at XMP. That wasn't the case previously.

I just let it run while doing something else or using another machine

There is a good chance that if you put a whole CCX to 4.6GHz that score would rocket. As I have said before, setting cores one by one is not as effective, at least not for me.

As it is, for full loads I have a better single core performance than you with my 3600 as my 4.4GHz works with all loads

 

[humbug]

DxO batch RAW conversion exports are reasonably well multi-threaded,as there are reviews which compare recent CPUs together and the Ryzen CPUs do very well against the current Intel crop. So you can see the higher core count Intel and AMD CPUs beating the 8 core models:
http://www.comptoir-hardware.com/ar...70-a-ryzen-7-3700x-ryzen-9-3900x.html?start=9
https://news.mynavi.jp/article/20190707-854576/3

However,once you start mucking around with filters,these tend to prefer single threaded performance.


How are you get another 6% more with a 100MHZ clockspeed increase? Something is not right there!

But,this really puts things in perspective for me - I have a SFF PC,so it appears at stock any of these Intel CPUs won't actually be really quicker even if you are applying filters.

17 seconds with a 5.2Ghz 9700K vs 18 on whatever Amigafan is running his at doesn't seem like something worth shouting about, not that Martin is.

While this specific task is low threaded its still high enough to force mine into SMT territory making it the worst case scenario for me.

If you spend all day exporting picture with these applied filters then yes the 9700K at 5.2Ghz is your best option, but its pretty niche.

 

[Rroff]

Precisely. This is the problem I've generally had with HT/SMT and why I would never get a 9900K as I find Intel's HT to be a worse implementation than Ryzen's SMT.
Could you run the test a few times to see if there is any variation?

Got any decent benchmarks for that? I've never found proper testing online. (You can find results for individual CPUs but not good comparisons side by side in equal circumstances over a broad range of application).

I've not tested it in awhile but when I did I found that AMD's SMT typically had a bigger penalty for enabling it which makes it look like the gains are bigger but really aren't. AMD's implementation tended to do better in synthetic tests and situations that were highly multi-threaded (probably due to the way they utilise the integer units) but worse in general applications and situations with more mixed workload demands - overall results were largely about the same.

 

[CAT-THE-FIFTH]

17 seconds with a 5.2Ghz 9700K vs 18 on whatever Amigafan is running his at doesn't seem like something worth shouting about, not that Martin is.

While this specific task is low threaded its still high enough to force mine into SMT territory making it the worst case scenario for me.

If you spend all day exporting picture with these applied filters then yes the 9700K at 5.2Ghz is your best option, but its pretty niche.

I use DxO myself BTW,and I tested an earlier version with my Ryzen 5 2600 using Prime NR. During batch export I had all the threads on my Xeon E3 1230 V2 and Ryzen 5 2600 pegged at nearly 100% or thereabouts. I was using 24MP files from my D600.

I don't overclock as I use a mini-ITX system,so doing some rough calculations for stock performance,Ryzen looks fine for me,its just Intel can push ahead in single threaded situations due to clockspeed,and some filters require it more than others.

 

[humbug]

I use DxO myself BTW,and I tested an earlier version with my Ryzen 5 2600 using Prime NR. During batch export I had all the threads on my Xeon E3 1230 V2 and Ryzen 5 2600 pegged at nearly 100% or thereabouts.

I don't overclock as I use a mini-ITX system,so doing some rough calculations for stock performance,Ryzen looks fine for me,its just Intel can push ahead in single threaded situations due to clockspeed,and some filters require it more than others.

And that's with Zen+, given the IPC difference and clock speed i'm seeing a solid 20% ST performance uplift vs my old 1600.

On a broader note, given Intel and AMD CPU architectures are very different one can always present them in whatever light one choses, often not deliberately, a good example of this, RoTR, take 4 reviewers Hardware Unboxed have Zen 2 about 10% slower vs Coffeeake in that game, as do GN, Jayz2sents and Der8ouer very recently have Zen 2 the same performance if not a few % faster.

They are all right, but their testing methodology is different. look at everything, you get worst and best case comparisons for the same game.

Got any decent benchmarks for that? I've never found proper testing online. (You can find results for individual CPUs but not good comparisons side by side in equal circumstances over a broad range of application).

I've not tested it in awhile but when I did I found that AMD's SMT typically had a bigger penalty for enabling it which makes it look like the gains are bigger but really aren't. AMD's implementation tended to do better in synthetic tests and situations that were highly multi-threaded (probably due to the way they utilise the integer units) but worse in general applications and situations with more mixed workload demands - overall results were largely about the same.

That would be difficult to quantity 'full stop' Given Zen 2 CPU's all have SMT, so i don't even know where you would get that conclusion from. I know this much, an 8 core Zen 2 has a better performance difference vs the 9900K in MT Cinebench scores than it does ST, which suggests Zen 2 using SMT is more efficient than Intel using SMT.

 

[Deleted member 66701]

17 seconds with a 5.2Ghz 9700K vs 18 on whatever Amigafan is running his at doesn't seem like something worth shouting about, not that Martin is.

Stock - not even PBO.

 

[humbug]

Stock - not even PBO.

Right. so what's that 4.5Ghz?

This vs 5.2Ghz.

 

[Orange Nexus]

Right. so what's that 4.5Ghz?

This vs 5.2Ghz.

According to Martin his 9700K system is not tuned.

 

[humbug]

According to Martin his 9700K system is not tuned.

Oh ok

He did talk somewhat about his CPU being overclocked to 5.2Ghz, but guess if he later clarified that its actually not, or at least not when he'd using DX0 then i guess that's how it is.

My 3900X runs slightly warmer than my overclocked 9700K when I tested them with the same cooler.
The 3900X is definitely more efficient and AMD had done a brilliant job in this regard.

This same CPU scored 528 in a different motherboard which would only boost to 4575Mhz but in this one that normally boosts to 4625Mhz it scored sightly less, which was not what I expected. On the plus side this motherboard scores 200 points higher in multicore at stock, which is higher than most I've seen. As this is mainly used for multicore work I'm happy with that.

When I was assessing which system I wanted to use for my photo editing, I just wanted to get the fastest one, be that just 10% faster, it didn't really matter, I just wanted to get the fastest for my usage within reason.

I couldn't really find any reviews on Puget etc that fitted my particularly workflow or with the systems that I would likely be running. I even had chats and help from @CAT-THE-FIFTH so thanks to him for his assistance.

I decided to get both systems so I could test for myself exactly which one would be the fastest. My 9700K uses ~147w at full load and the Ryzen 3900X ~142w so power usage wasn't really a concern. Though the 3900X is brilliantly efficient.

It turned out that the 9700K is ~12% faster at 5.2Ghz and at 5.3Ghz 18% faster, though at 5.3Ghz the VRM's on this Z370 are begging for mercy so I've got a Z390 board to drop in.

When you're getting through 1000's of photo every percentage or second is a help in the right direction.

No, the 3900X was tuned for all tests. The 9700K has not been tuned at all, and is still not tuned. Just running Corsair Vengeance 2666Mhz at stock XMP. You can see this in the CPU-Z shot.

The 9700K? what Screenshot? the only one you posted was for the 3900X.

 

[humbug]

This one?

Yeah, Okay. Back to Single Threaded performance.

Here is my tuned 3900X Cinebench R20 and CPU-Z single threaded

and here is my 9700K @ 5.2Ghz

This is on a Windows 10 install that was cloned from one machine to another, so OS wise and re background apps running, everything is exactly the same apart from the respective drivers

Oh and here is the screenshot of my DXO 9700K with the "filters" showing just so it doesn't appear "strange" or suspicious to you.

I presume there are no other tests you want me to run?

 

[CAT-THE-FIFTH]

@humbug @MartinPrince I decided to test the raw image with my Ryzen 5 2600. Its running at stock clockspeeds under a Wraith Spire stock cooler,and with this RAM kit running at its XMP settings:

https://www.overclockers.co.uk/team...-3200mhz-dual-channel-kit-blac-my-0a2-tg.html

I am running the files off a NVME WD Black SN750 SSD.

So you can say my system isn't really tuned in anyway apart from the RAM being run at XMP - its even using a stock cooler!!

I tested with both SMT off and on.

These are the output settings:



With SMT off the CPU stayed at a constant 3.65GHZ to 3.66GHZ across all cores.



Each picture took 28~29 seconds to process,I would say its mostly on the cusp of 28~29 seconds. The cores were at 100% until the transcode was finished.

Now I enabled SMT.



Clockspeeds hovered closer to 3.6GHZ and you can see that the cores were not at 100% or they were but very briefly before the load was moved to the next thread by the scheduler.



However,consistantly the transcode finished in 27 seconds,so it was definitely slightly faster with SMT enabled.

So for a CPU just running at 3.6GHZ in a mini-ITX system using the stock cooler I think I am doing OK TBH!!

I am intending to buy a Ryzen 7 3700X at some point,so another 1/3 more cores,and another 20% clockspeed(plus the IPC increase),should give me a nice decrease in transcoding times.

 

[MartinPrince]

Oh ok

He did talk somewhat about his CPU being overclocked to 5.2Ghz, but guess if he later clarified that its actually not, or at least not when he'd using DX0 then i guess that's how it is.

The 9700K is just overclocked to 5.2Ghz the 2666Mhz memory was not tuned at all just run at XMP. The 3900X is running the latest bios and AMD drivers, with memory tuned with DRam Calc to run 3666 Mhz C14. This is all shown in the CPU-Z screenshot I posted.

 

[humbug]

Edit: ok thank you for the clarification ^^^^

@humbug @MartinPrince I decided to test the raw image with my Ryzen 5 2600. Its running at stock clockspeeds under a Wraith Spire stock cooler,and with this RAM kit running at its XMP settings:

https://www.overclockers.co.uk/team...-3200mhz-dual-channel-kit-blac-my-0a2-tg.html

I am running the files off a NVME WD Black SN750 SSD.

So you can say my system isn't really tuned in anyway apart from the RAM being run at XMP - its even using a stock cooler!!

I tested with both SMT off and on.

These are the output settings:



With SMT off the CPU stayed at a constant 3.65GHZ to 3.66GHZ across all cores.



Each picture took 28~29 seconds to process,I would say its mostly on the cusp of 28~29 seconds. The cores were at 100% until the transcode was finished.

Now I enabled SMT.



Clockspeeds hovered closer to 3.6GHZ and you can see that the cores were not at 100% or they were but very briefly before the load was moved to the next thread by the scheduler.



However,consistantly the transcode finished in 27 seconds,so it was definitely slightly faster with SMT enabled.

So for a CPU just running at 3.6GHZ in a mini-ITX system using the stock cooler I think I am doing OK TBH!!

I am intending to buy a Ryzen 7 3700X at some point,so another 1/3 more cores,and another 20% clockspeed(plus the IPC increase),should give me a nice decrease in transcoding times.

That's a lot of work, thanks for doing that.

Its very interesting, you're using the same image so they should be the same, what is interesting about this is you appear to be loading up all 12 threads BUT not to 100%.

If you look at mine i'm only loading up 8 thread's but all at 100%, could that be a difference in Thread scheduling in the architecture? facinating....

 

[CAT-THE-FIFTH]

@humbug the increase in performance of your over 4.0GHZ Ryzen 5 3600 over my 3.6GHZ Ryzen 5 2600 is not as big as I expected - so I am at 27 seconds,and you are at 23~24 seconds,so its almost like its the clockspeed difference accounting for the performance jump in your case.

Edit!!

I will badger my mate with an overclocked Ryzen 7 2700 to see if he can run the test too,although his RAM is only running at 3000MHZ IIRC.