AMD vs Intel Single threading?

[Robert896r1]

We can do a fun test. Anyone here have a 16 thread 3700x/3800x they have tuned for daily all core? If you run say R20 on that, I can tune my 9900k to match the score and your mem frequency. Then we can run some synthetics and compare the power draw in what will be the most like for like you can get. We can also look at temps to see how much hotter the 9900k run at parity. Since I'm just using 80gbp AIO its' nothing exotic.

I find it more fun to test things and get data points.

 

[subbytna]

I'm hard pressed to find evidence for your statement about this SMT scaling advantage. To demonstrate, I just ran R20 SC and MC on my 9900k at my daily profile:



When I compare that to the R20 thread here according to your claim, the Ryzen Cpu's should be flexing their SMT scaling but that's not the case. I'm ahead in Single score and I'm ahead in multiple core. At 16 threads there's no "SMT scaling" magic or anything like that that puts me at a disadvantage or lets the Ryzen CPU's catch up.

When I look at TNA's score of 544 it's done at 4.675 and let's say that's even with me on SC. https://forums.overclockers.co.uk/posts/33152922
For MC, nghtmare's score of 5365 is at 4.45ghz. https://forums.overclockers.co.uk/posts/32985645 with R20 scaling, he'd need TNA's frequency to match me in Multi so again, I don't see any SMT advantage. If there was a scaling advantage it would show up here.

If you have synthetics that demonstrate this SMT advantage, happy to run them. I can tell you now that something like TimeSpy which also scales well with threads is a massive advantage for me also over any Ryzen 16 thread CPU.

We're on a site called overclockers so what someone does with limited knowledge or experience isn't my concern. I'm on AIO that costs the same as a D15 so we can take exotic cooling etc out of the discussion also. I'm hoping you stay on the mainstream platform for discussion and don't start jumping around randomly to other platforms.

What cooling do you use for your cpu?

 

[MartinPrince]

I remember this, this is a different benchmark variation to the slides that i posted. I'm talking about one thing and you're calling me out as if its the same thing. This was your own custom file.

You really do think i have "Learning Difficulties"

This topic is titled "AMD vs Intel Single threading?"
The simple fact is that nowadays most people will rarely use any software that can/will only use a single thread. The reality is that a lot of software can use several threads BUT can't use all threads over a certain amount.

Now I know from my use of both systems that if I have an everyday, simple overclock of both a 9700k and 3900x that when it comes to Single Threaded tasks (or software than can't take advantage of high thread counts) the 9700k is faster. Once similar tuned memory is used on both then it turns out to be ~15%.

So I post up my real world, real software results that are not fully multithreaded (DXO Photolab) and because you don't see something with Ryzen as the best you seem to start going into some kind of meltdown and want to find an Intel conspiracy.

On your Lightroom result. I don't use it so i have to concede that to you, i do think its strange that you don't show the filters used for the 9700K.

The fact that I did not mention Lightroom seems to escape you.

In what I can only see as some kind of Cognitive Dissonance you then find some Lightroom charts (bearing in mind I didn't even mention Lightroom) that show Ryzen on top, even though you did not have a clue what those charts really referred to, whether they are fully multithreaded actions or not. I know exactly what 'Passive Task Score' and 'Active Task Score' mean.

Now that you feel you've got a victory for Ryzen, without realising you've probably gone off topic, you then go on to say:

No comment on the review slides then? My Zen 2 entry level CPU is faster than your high end Coffeelake, doesn't that sting even just a little?

You compound your error further by making this baffling statement in the context of this thread:

My 3600 does a better job in Lightroom than his 9700K....!

There is simply no single threaded benchmark, program or application that your 3600 will be faster than my overclocked 9700k. In the empirical test that both you and I did you are 37% off

By making such a daft statement you turn yourself into the OcUK Forum AMD alternative to Ryan Shrout.

 

[Robert896r1]

What cooling do you use for your cpu?

https://www.overclockers.co.uk/arct...ormance-cpu-water-cooler-360mm-hs-07d-ar.html

*** competitor hint removed ***
- Armageus

 

[humbug]

Cell wasn't exactly the easiest to code for either from what I know so that wouldn't help. The Jaguar cores of the current gen though should have been ripe for multithreading and they seem to be sort of taken advantage of but I am not convinced it's fully utilised.

8 Jaguar threads were equivalent to about half of one Zen 2 thread.

I'm just guessing there but its probably pretty accurate.

Jaguar were low power cores based on Bulldozer, we know Zen has 72% higher IPC than Bulldozer, Zen + was 4% higher than Zen, Zen 2 about 15% higher than Zen +, so add 20% to 72% brings it to 92% higher IPC than Bulldozer / Jaguar, Jaguar was a much smaller 20 Watt 2Ghz core, so 30 Watts 8 thread Jaguar vs 120 Watts 8 thread Bulldozer.... just a single Zen 2 core is probably twice as fast as the whole Jaguar CPU.

So it didn't really matter that Jaguar had 8 threads, a single Ivy Bridge core could do the same work as the whole Jaguar CPU.

The new consoles are going to be quite different with 16 Zen 2 threads running at probably 3.6Ghz, which would be a 40 Watt CPU, about 40 times as fast as the old Jaguar CPU, that will cause a difference in how games will use CPU's as there is vastly more performance available in those consoles, most games these days are made for consol and ported to PC, made for Zen and Navi architectures.....

 

[humbug]

In fact you only need to look at AMD's new Mobile APU's, they are 16 Zen 2 threads running at between 3 and 4Ghz on a 15 / 25 / 35 Watts TDP. in MT performance they are faster than Intel brand new 10nm Icelake with ST more or less on par.

The high end of those is what will be in the new consoles.

 

[BigEars]

I'm hard pressed to find evidence for your statement about this SMT scaling advantage.

SMT scaling is a ratio between a single thread and multiple threads on the same core and to measure it all threads must run at the same frequency. Also IPC is being banded around without any context to what instructions are being used such as Integer, FPU, SIMD, encryption and so on. For example with Cinebench R15 you run ST and MT at the same clocks then divide the MP ratio by the number of cores to get the SMT scaling for that bench. CPU-z also offers something similar with its inbuilt benches but once again ratio's will be dependent on instructions used and optimization. With well optimized code it is even possible for degradation to occur with SMT meaning 2 or more threads on the same physical core running with less performance than just running with one thread on that core.

 

[MartinPrince]

.... no 5.3Ghz 300 watt 8 core Coffeelake is going to keep up with that.

Reading some of the nonsense you can come out with is reminiscent of those fanatical cult leaders; 90% of what they are saying might be correct and plausible but every now and then they throw in some BS belief, that if you didn't know better you would take for gospel.

Anybody reading that would indeed think that all 8 core CoffeeLakes use 300 Watts and seeing mine is the only 5.3Ghz 8 CoffeeLake in the conversation let's see how much I actually use.

Here is my HWInfo64 during AIDA64 Stability Stress Test:

As you can see in the CPU Package Power my 5.3Ghz 8 Core CoffeeLake uses ~135 Watts at full load.
Yet you feel it's OK to more than double that amount in one of your off topic "Ryzen is better" posts

Can't you see why I would question if you might have some cognitive impairment? That, or you are indeed the OcUK Forum AMD alternative to Ryan Shrout.

 

[sideways14a]

135w at 5.3? is that only 8 thread as its a 9700?

SHT ect will add a bit onto that as the pipelines are kept more busy.

 

[Grim5]

voltages are very low = lucky chip lotto winner
Hyper threading is disabled, add it in and its a different story - you can cut AMD consumption also in half by disabling SMT

79c at just 135w is hot though

No need to argue though, most reviewers support the data that Intel is power hungry

 

[MartinPrince]

135w at 5.3? is that only 8 thread as its a 9700?

SHT ect will add a bit onto that as the pipelines are kept more busy.

Definitely, a 9900K will use more but when Humbug's nonsense statement is made it generalises ALL 8 core Coffeelake and then distorts the view that they ALL run at 300W

 

[Grim5]

Definitely, a 9900K will use more but when Humbug's nonsense statement is made it generalises ALL 8 core Coffeelake and then distorts the view that they ALL run at 300W

He was probably exaggerating a bit. Depending on bin quality, clocks and voltages, Intel 16 thread CPU's run between 200w and 250w. 300w is more likely where the 10 core model will end up based on a 1:1 scale of power draw to core count - we saw it with the move from 4 to 6, then 6 to 8 - Intel needs an extra 50w of power draw for every 2 hyper threaded cores it adds onto Coffeelake, then also needs a little extra if clock speeds are increased.

 

[Robert896r1]

For 5hgz all core my 9900k pulls 143w in r20 and temps are mid 60’s. That’s a 10 run average and matches an oc’d 3700/3800x oc’d to a 4.4+ all core.

I’d be curious to see thermal, volts and power on a 3700x/3800x at 4.4+

Anyone do actual testing here?

 

[Grim5]

For 5hgz all core my 9900k pulls 143w in r20 and temps are mid 60’s. That’s a 10 run average and matches an oc’d 3700/3800x oc’d to a 4.4+ all core.

I’d be curious to see thermal, volts and power on a 3700x/3800x at 4.4+

Anyone do actual testing here?

Cinebench is not considered a stress test, and your 9900k probably has a big AVX offset which is very commonly done when overclocking it - so you are reducing performance in AVX load to reduce power draw, which is not a fair comparison

and for reference my 1.37v 5ghz 8700k draws more than that running cinebench r20 and that's because I did not use AVX offsets - thats the true power draw.

I will continue to trust Reviewer's data rather than some random person and not a single reviewer has reproduced a 9900k using 143w under a stress test

 

[sideways14a]

Definitely, a 9900K will use more but when Humbug's nonsense statement is made it generalises ALL 8 core Coffeelake and then distorts the view that they ALL run at 300W

Its not very good though is it? Only 8 threads and still gulping down a fair few watts.

 

[Robert896r1]

Cinebench is not considered a stress test, and your 9900k probably has a big AVX offset which is very commonly done when overclocking it - so you are reducing performance in AVX load to reduce power draw, which is not a fair comparison

and for reference my 1.37v 5ghz 8700k draws more than that running cinebench r20 and that's because I did not use AVX offsets - thats the true power draw.

I will continue to trust Reviewer's data rather than some random person and not a single reviewer has reproduced a 9900k using 143w under a stress test

Where did I say it's a stress test? It's a synthetic that's heavily used. Infact, it's hallowed ground around here so all of a sudden it's not relevant?

You're right. It's not 143, it's 140w at peak. Sorry that you had a bad chip. I can't help that.

Here's the run. I can't be bothered to lower voltage below 1.18v (load = 1.13v). I don't do AVX offsets. If anything, they cause transient spikes which leads to instability.

 

[MartinPrince]

Its not very good though is it? Only 8 threads and still gulping down a fair few watts.

But it is not 300W as Humbug alluded to! It is 134W. That is the only point here! You just can't go around making up crap and passing it off as fact.

 

[Robert896r1]

But it is not 300W as Humbug alluded to! It is 134W. That is the only point here! You just can't go around making up crap and passing it off as fact.

You can probably get 300w out of a 10core. It'll depend on the voltage needed + current under load. If I wasn't lazy, I'd check the per core amp load and 10x it. That would be a base formula to then apply different load voltage to see what's needed to hit 300w.

You'd want a consistent load like p95 small that stays more or less constant to get the measure.

To be fair half the time people aren't reading the right power draw figure so it's hard to know what they're referencing unless they state the value they're using.

 

[humbug]

Software power monitoring......

voltages are very low = lucky chip lotto winner
Hyper threading is disabled, add it in and its a different story - you can cut AMD consumption also in half by disabling SMT

79c at just 135w is hot though

No need to argue though, most reviewers support the data that Intel is power hungry

Yeah its really not that much of a stretch given that OC 5.2Ghz 9900KS chips run at around 250 watts in Cinebench.

 

[FoxEye]

@MartinPrince Just wanted to say thanks for the effort you put into this thread.

I would suggest tho you're wasting your time arguing with the usual AMD die-hards. They are emotionally invested in AMD and will not agree under any circumstances that AMD chips are simply not the best at every task; single-threaded, multi-threaded, any workload - their answer is always that AMD is best.

You won't have any joy no matter how much data and facts you can present. The "truth" is whatever they want it to be.

It's sad because we all know AMD are making good chips these days - they don't need the campaign of disinformation from the AMD super-fans to sell themselves.

But they do still have some (limited) weaknesses against Intel chips and the unbiased among us enjoy reading these threads and seeing it laid bare.

So thanks to you again Was a good read.