AMD vs Intel Single threading?

[Rroff]

I always have turbo boost disabled and set my clockspeeds manually, because I have the cpu to lower clockspeeds when idel anyway, so I dont see the point with turbo boost.

Depends what you can achieve manually versus boost behaviour, etc.

The huge gap between 9900K and 9900KS in SINGLE threaded tasks is weird. The main difference between these CPUs are clocks in MULTI threaded mode, while single thread clocks should be equal for both.

You get very few situations where you are truly single core boosting though - so the ratios of multi-core clocks will probably have an impact along with potentially factors like maybe the KS is boosting very highly for a short time but in doing so hits the limits so hard that it actually drops below the K briefly over a longer period before coming back up again to level out eventually (or some variant of that kind of behaviour I've not actually looked into it).

 

[Phil2008]

Depends what you can achieve manually versus boost behaviour, etc.



You get very few situations where you are truly single core boosting though - so the ratios of multi-core clocks will probably have an impact along with potentially factors like maybe the KS is boosting very highly for a short time but in doing so hits the limits so hard that it actually drops below the K briefly over a longer period before coming back up again to level out eventually (or some variant of that kind of behaviour I've not actually looked into it).

I have all of the 6 cores set at 4.9 on my 8086K cpu, so as soon as the system detects a load on the cpu, all cores shoots to 4.9ghz and they stay there untill there's not much load on the cpu or system is at idel. I'm guessing this is a much better way than using the turbo boost feature, I have done it this way with all my other recent cpu's aswell and it works flawless.

Cpu at idel


Cpu at load

 

[HangTime]

So basically the chart up the top shows that IPC differences aren't huge; even a 7600k which is essentially 3 years old now is only ~7% slower than R9-3900X at 3.5ghz. Overclocked to say 4.8ghz it's going to be even closer as most Ryzens are unlikely to clock as high.

Also interesting to see how close the IPC is between Zen and Zen+ (1600X scores the same as 2700); I had been under the impression that you'd see a bit more improvement from Zen+.

I guess what I take away from this is I need to wait for Zen3 to see if we finally get the IPC monster and high clockspeed we are waiting for.

 

[Deleted member 209350]

If I designed and made a CPU that runs at 100Hz, and it completes a task or continuously runs a task faster/in less time than a CPU that runs at 10GHz which one is faster?

The number of per-seconds (Hz) makes no difference if your design isn't as good, you compensate for a worse design with more Hz.

You guys must not have been around during the Pentium 4 days.

I think the point being made was that If for example CPU Z at 100Hz has better IPC and performs better than CPU X also at 100Hz - Z is clearly faster and superior.

However if CPU X was clocked at 500Hz, it performs better than the 100Hz processor (obviously not in all cases, but in general) then that processor is faster. I.e its made up for the worse IPC by simply just being higher clocked so it makes up the difference

 

[CuriousTomCat]

There's not much in it between AMD and Intel these days when comparing a like for like core count. I guess it comes down to whether you want to encourage stagnation by choosing Intel or progress by choosing AMD. People don't think about the consequences when buying

 

[Orange Nexus]

So basically the chart up the top shows that IPC differences aren't huge; even a 7600k which is essentially 3 years old now is only ~7% slower than R9-3900X at 3.5ghz. Overclocked to say 4.8ghz it's going to be even closer as most Ryzens are unlikely to clock as high.

Also interesting to see how close the IPC is between Zen and Zen+ (1600X scores the same as 2700); I had been under the impression that you'd see a bit more improvement from Zen+.

I guess what I take away from this is I need to wait for Zen3 to see if we finally get the IPC monster and high clockspeed we are waiting for.

Yep, and if you look at Kaby lake compared to Coffee Lake the difference is nill.

 

[HangTime]

I think the point being made was that If for example CPU Z at 100Hz has better IPC and performs better than CPU X also at 100Hz - Z is clearly faster and superior.

However if CPU X was clocked at 500Hz, it performs better than the 100Hz processor (obviously not in all cases, but in general) then that processor is faster. I.e its made up for the worse IPC by simply just being higher clocked so it makes up the difference

Or in summary you have to consider IPC and clockspeed together, there is no point having a "clearly faster and superior" processor if it runs so much slower than the competition (conversely, no point having clockspeed advantage if it performs slower e.g. P4s getting beaten by A64).

Let's be honest guys I'm sure someone could come up with a CPU that outperforms Ryzen at 100hz if they designed it on that basis, knowing they don't need to worry about thermals, just going for all out IPC, but that wouldn't make it a better buy if it couldn't scale into the ghz range.

 

[Deleted member 209350]

Or in summary you have to consider IPC and clockspeed together, there is no point having a "clearly faster and superior" processor if it runs so much slower than the competition (conversely, no point having clockspeed advantage if it performs slower e.g. P4s getting beaten by A64).

Let's be honest guys I'm sure someone could come up with a CPU that outperforms Ryzen at 100hz if they designed it on that basis, knowing they don't need to worry about thermals, just going for all out IPC, but that wouldn't make it a better buy if it couldn't scale into the ghz range.

Yeah but when you have a single core test between AMD vs Intel latest chips, Zen2 has the better IPC but intel has the higher clock speeds which makes up the difference it loses from having worse IPC.

Once AMD can get higher clock speeds (around 5Ghz) with its already superior IPC, thats when it will crush Intel

 

[HangTime]

That's the problem though, we are still waiting for this IPC*Clockspeed behemoth to turn up. Every iteration the hype gets built up and every time we get well, not disappointed (Zen is decent, I have one; Zen2 is definitely moving things forward), but not exactly blown away either. I can't really remember the last time I went wow, this is annihilating allcomers... maybe Core 2? As for Intel getting crushed, I guess that depends on when it happens and what they have at the time at what price.

 

[humbug]

That's the problem though, we are still waiting for this IPC*Clockspeed behemoth to turn up. Every iteration the hype gets built up and every time we get well, not disappointed (Zen is decent, I have one; Zen2 is definitely moving things forward), but not exactly blown away either. I can't really remember the last time I went wow, this is annihilating allcomers... maybe Core 2? As for Intel getting crushed, I guess that depends on when it happens and what they have at the time at what price.

Irrelevant.

Score 4056: Intel Core i7 8700K at 5.2Ghz, Radox-0
Score 4041: AMD Ryzen R5 3600 at 4.5Ghz, TNA

8700K: 4056 @ 5.2Ghz
3600: 4041 @ 4.5Ghz

IPC Difference +15% to Zen 2

Cost:
8700K £350
3600 £175

"Oh but Intel GazzilionHz" doesn't matter..... Zen 2 IPC is miles behind and Intel FAR too much money....

Edit:
9, Score 547: Intel Core i7 8700K at 5.2Ghz, Radox-0
10, Score 544: AMD Ryzen R5 3600 at 4.675Ghz, TNA
(+12% IPC ST)

 

[LePhuronn]

I can't really remember the last time I went wow, this is annihilating allcomers

Um, have you not been paying attention since July? You don't consider Zen 2 to be annihilating Intel?

OK then...

 

[humbug]

Um, have you not been paying attention since July? You don't consider Zen 2 to be annihilating Intel?

OK then...

Haven't watched this yet but the Thumbnail says it all....

Edit: Look around you HangTime, Ryzen everywhere.....

 

[TNA]

Irrelevant.

Score 4056: Intel Core i7 8700K at 5.2Ghz, Radox-0
Score 4041: AMD Ryzen R5 3600 at 4.5Ghz, TNA

8700K: 4056 @ 5.2Ghz
3600: 4041 @ 4.5Ghz

IPC Difference +15% to Zen 2

Cost:
8700K £350
3600 £175

"Oh but Intel GazzilionHz" doesn't matter..... Zen 2 IPC is miles behind and Intel FAR too much money....

Edit:
9, Score 547: Intel Core i7 8700K at 5.2Ghz, Radox-0
10, Score 544: AMD Ryzen R5 3600 at 4.675Ghz, TNA
(+12% IPC ST)

Crazy isn't it, that class of performance for £175 which is what I paid. People were paying a lot more until about a year ago for this kind of performance.

 

[humbug]

Crazy isn't it, that class of performance for £175 which is what I paid. People were paying a lot more until about a year ago for this kind of performance.

Intel are still selling the 8700K, hence my pricing it at £350, it used to cost around £400. The 9700K is about the same money as current 8700K pricing, and no quicker.

So yeah....

 

[MartinPrince]

Should probably use the newer version of Maxon Cinema 4D.

Of course IPC is application dependant and Cinebench R15 is known for using a slower code path when AMD is detected, These days application vendors tend not to play this game and with that Intel have taken to creating their own benchmarking application suites and calling them "real world performance" what Intel are doing is benchmarking the CPU's themselves on their own software creations and marketing that expecting you to nood, clap and buy their crap.

I think what 8K4K is driving at is 'all things being equal' Ryzen 3000 IPC is so high it often over comes the higher clock speed of Coffeelake.

Now if he'd posted that benchmark then it would have been perfectly fine.

The simple fact is extrapolating from IPC to Single Threaded performance is just not accurate and more likely to disseminate distorted narratives.

You'll get a much better and accurate idea by running proper single threaded tests like the ones in CPU-Z.

My main usage is photo editing, which although has become more multithreaded over the years still largely benefits from raw single threaded performance, especially when not doing batch workflows where multi core is more beneficial.

I'm quite like few on here that can't have a CPU without overclocking it or optimising it in some way so I have to account for the probable overclock, for someone like me who's an average overclocker, will get out of a CPU.

I reckoned I would get 5.1Ghz/5.2Ghz out of the R0 9700k (actually goes to 5.3Ghz) and with 3900K it was mainly about getting IF and memory as close to 1900Mhz and memory tuned with the best timings.

With any decent system nowadays most things are pretty snappy and hard to measure but for my workflow the longest thing that takes time is an export from DXO Photolab to Lightroom using Prime Noise Reduction. Luckily DX0 has an inbuilt timer so you can see how long this takes.

Here is a test using the same photo and settings on my 9700K at 5.2Ghz and 3900X using 1usmus power plan and IF 1:1 at 1866 Mhz with tuned C14 timings. The 3900X can boost as high at 4650Mhz but normally 2 cores will do 4625Mhz. I ran the test 5 times for each.

3900X

9700K

As you can see the 9700K is 10% faster for this usage.
If you're not an overclocker and just a plug and player then you go Ryzen all day but if you are, then you might want to take your particularly usage into account.

 

[humbug]

Now if he'd posted that benchmark then it would have been perfectly fine.

The simple fact is extrapolating from IPC to Single Threaded performance is just not accurate and more likely to disseminate distorted narratives.

You'll get a much better and accurate idea by running proper single threaded tests like the ones in CPU-Z.

My main usage is photo editing, which although has become more multithreaded over the years still largely benefits from raw single threaded performance, especially when not doing batch workflows where multi core is more beneficial.

I'm quite like few on here that can't have a CPU without overclocking it or optimising it in some way so I have to account for the probable overclock, for someone like me who's an average overclocker, will get out of a CPU.

I reckoned I would get 5.1Ghz/5.2Ghz out of the R0 9700k (actually goes to 5.3Ghz) and with 3900K it was mainly about getting IF and memory as close to 1900Mhz and memory tuned with the best timings.

With any decent system nowadays most things are pretty snappy and hard to measure but for my workflow the longest thing that takes time is an export from DXO Photolab to Lightroom using Prime Noise Reduction. Luckily DX0 has an inbuilt timer so you can see how long this takes.

Here is a test using the same photo and settings on my 9700K at 5.2Ghz and 3900X using 1usmus power plan and IF 1:1 at 1866 Mhz with tuned C14 timings. The 3900X can boost as high at 4650Mhz but normally 2 cores will do 4625Mhz. I ran the test 5 times for each.

3900X

9700K

As you can see the 9700K is 12% faster for this usage.
If you're not an overclocker and just a plug and player then you go Ryzen all day but if you are then you might want to take your particularly usage into account.

You don't think Maxon Cinema 4D is a "Proper" workload but CPU-Z is? I don't think you know what you're talking about.

19 seconds vs 17 seconds. Why no filters on the 9700K export? Genuine question, awaits answer....

 

[MartinPrince]

You don't think Maxon Cinema 4D is a "Proper" workload but CPU-Z is? I don't think you know what you're talking about.

Cinebench R20 gives me a similar result as CPU-Z so they are both better than using IPC to extrapolate single thread performance. In both Cinebench R20 and CPU-Z my 9700K scores more than my 3900X in single threaded.

19 seconds vs 17 seconds. Why no filters on the 9700K export?

DXO users will know the 'filters' are always there, they are just minimised for the 9700K whereas the 3900K was expanded. They both used exactly the same preset as it was just exported from one machine to the next. The photo was also the same in case you are questioning why they look different - just a different part of the photo.

Are there any other tests you might like me to run?

 

[humbug]

Cinebench R20 gives me a similar result as CPU-Z so they are both better than using IPC to extrapolate single thread performance. In both Cinebench R20 and CPU-Z my 9700K scores more than my 3900X in single threaded.

DXO users will know the 'filters' are always there, they are just minimised for the 9700K whereas the 3900K was expanded. They both used exactly the same preset as it was just exported from one machine to the next. The photo was also the same in case you are questioning why they look different - just a different part of the photo.

Are there any other tests you might like me to run?

The highest ST R20 Ryzen score is 547, this is at 4.7Ghz, the highest Coffeelake score is 580, this is at 5.45Ghz, a score difference of 6% with a clock speed difference of 16%, so yes if you can get past the IPC difference between Zen 2 and Coffeelake then you can beat Zen 2 score, but that IPC difference is very obviously there, if you cannot see that you don't know what IPC is.

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.

However, i can use google and the first review i came across has Zen 2 winning.







https://www.pugetsystems.com/labs/a...adripper-2-Intel-9th-Gen-Intel-X-series-1592/

 

[MartinPrince]

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.

It is not strange at all as the only 'filter' that you can see in 3900K shot is the Prime NR set to 40. All the others are minimised.

Obviously you are totally ignorant and clueless as to how DXO works - which I was at one point, so just for you I'll run the test again showing the exact 'filter' window you can see in the 3900X.

 

[humbug]

It is not strange at all as the only 'filter' that you can see in 3900K shot is the Prime NR set to 40. All the others are minimised.

Obviously you are totally ignorant and clueless as to how DXO works - which I was at one point, so just for you I'll run the test again showing the exact 'filter' window you can see in the 3900X.

Right.

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?