AMD vs Intel Single threading?

Robert896r1 · 26 Mar 2020 at 17:57

A 9900kf all core boost is 4.7ghz.

https://en.wikichip.org/wiki/intel/core_i9/i9-9900kf

humbug · 26 Mar 2020 at 18:01

Robert896r1 said:
A 9900kf all core boost is 4.7ghz.

https://en.wikichip.org/wiki/intel/core_i9/i9-9900kf

Yeah, and 5Ghz 2 core boost

Robert896r1 · 26 Mar 2020 at 18:21

humbug said:
Yeah, and 5Ghz 2 core boost

Yep and still curious how the KF managed to be that low. Here's 50/43. Even if you ignore the all core being 50 for all core, the single is more in line with what it should be at 585ish.

My data is corroborated here: https://www.guru3d.com/articles_pages/intel_core_i9_9900k_processor_review,8.html

Not sure why the built in reference in cpu z is so low.

humbug · 26 Mar 2020 at 18:30

Robert896r1 said:
Yep and still curious how the KF managed to be that low. Here's 50/43. Even if you ignore the all core being 50 for all core, the single is more in line with what it should be at 585ish.

My data is corroborated here: https://www.guru3d.com/articles_pages/intel_core_i9_9900k_processor_review,8.html

Not sure why the built in reference in cpu z is so low.

RE: Your Guru3D review Single threaded score is low on the ChipHell bench, but the multithreaded score is the same, yours is higher. Are you running all core 5Ghz lockinging in the BIOS, like MCE?

Grim5 · 28 Mar 2020 at 09:05

10900k in geekbench. Only manages the same single core score as the 9900ks, even though the 10900k runs 200 to 300mhz faster.

If my calculations are correct - it means the 10900k's IPC is 7% slower than the 9900ks due to it's security mitigations, and it's extra 300mhz clock speed is only just enough to get it on par with last gen.

Another failure from Intel incoming

https://browser.geekbench.com/v5/cpu/1584149

CAT-THE-FIFTH · 28 Mar 2020 at 10:53

MartinPrince said:
I wholeheartedly agree with pretty much everything you wrote, especially the bit I quoted above. On the subject of the German graph, I don't really put too much stock in the testing methodology, if you're running memory at different speeds then you're bound to have big differences based on memory alone. Though you wrote it has "nothing" to do with core count what you have clarified is that it is "not only" core count + IPC, which I'm well aware of as I also do audio work;

Here is the best latency I get on the 3900X (though notice the comparable read/write/copy)....(rest removed to make reply smaller)

I did check other sources,and Zen2 CPUs do seem to do well in that scenario. The TR game you tested certainly does better than Fallout 4 on Ryzen...forum tests seem to hint despite Zen2 having better IPC than the current Intel cores,its clock for clock about the same or slightly worse than CFL. Intel still wins in latency and better optimisation for ring bus CPUs,and can overclock better(if you do overclock). So Ryzen needs to "brute force" things,and in some ways the consoles using Zen2 will help developers optimise better for Zen2 based CPUs.

Its why I am interested to see what design differences come to Zen3. I expect if the CCX design is changed and it leads to lower latencies,it will lead to a better uplift in certain applications,more than any IPC or clockspeed gains.

MartinPrince · 28 Mar 2020 at 11:04

Grim5 said:
fixed for you

Thanks, but really not necessary, though you are more than welcome to make the point yourself. Future proofing to me is just a subset of when you are on a budget, which I mentioned.

If 6 months ago somebody had told you that the country would be in lock down etc then you may have wondered what drug/narcotic that person had been imbibing. When it comes to tech, things can be a little more predictable but after years in this industry, I only base my buying on what is here and now or distinctly imminent, as in just about to be released and just waiting for stock.

I have seen all too often where people predict this and that but when the CPU gets in the hands of consumers then the reality is quite far removed from that. I remember talk of the 3900X being able to do 4.7Ghz all core but the reality is very different.

AMD have said that there will be another CPU release on AM4, if that release gives the predicted 17% that I saw mentioned then that would still not be faster in non-fully threaded tasks than my 9700K, which I've shown in real word apps to be ~20% faster than a 3900X. Therefore, there is no real reason for a person to go a year+ (from last July) having a 20% slower CPU waiting for a CPU that might just bring them to parity.

When Ryzen 4000 launches I hope it surpasses all expectation and for all uses sticks it to Intel, who have too long been bleeding customers dry and in return giving back very little. Until then I will get whatever is fastest for my usage.

Robert896r1 · 28 Mar 2020 at 11:12

humbug said:
RE: Your Guru3D review Single threaded score is low on the ChipHell bench, but the multithreaded score is the same, yours is higher. Are you running all core 5Ghz lockinging in the BIOS, like MCE?

I'm only focusing on the single core score. Both mine and Guru 3D are showing a similar number which is more in line with what it should be.

The "reference" in CPU-Z is quite far below that. I'm not sure why but again it shows the value of doing your own testing.

Robert896r1 · 28 Mar 2020 at 12:07

Grim5 said:
10900k in geekbench. Only manages the same single core score as the 9900ks, even though the 10900k runs 200 to 300mhz faster.

If my calculations are correct - it means the 10900k's IPC is 7% slower than the 9900ks due to it's security mitigations, and it's extra 300mhz clock speed is only just enough to get it on par with last gen.

Another failure from Intel incoming

https://browser.geekbench.com/v5/cpu/1584149

https://browser.geekbench.com/v5/cpu/1604638

That's my chip at 50/45 with highly tuned RAM which does come into play. Single score seems fine to me.

The boost clock on the 10900k is 5.1ghz, not 5.3ghz. TVB is what gets it to 5.3 but we don't know how that's implemented or what use cases there are for it. The normal boost is 5.1ghz

CAT-THE-FIFTH · 28 Mar 2020 at 12:18

humbug said:
Ryzen 7 4800 Beats the Intel Core i9-9900KF in CPU-Z Benchmark while Drawing just 45W

I score about 500 @ 4.2Ghz boost. The Ryzen R7 4800H is a Zen 2 mobile part, its probably boosting to about 4.6Ghz single core, this isn't even the highest end Zen 2 mobile part, that would be the Ryzen R9 4900HS.

The single and multicore performance for a 45 Watt part is impressive, that's down to 7nm and a very good architecture. Its basically full fat 9900K performance in a laptop.

From what I read about Renoir,it is an SOC,as it means IF can be run at a lower speed,and needs less active links than having chiplets. This reduces power consumption and heat quite a bit apparently,meaning Renoir can boost higher for a given TDP than Zen2 CPUs.

humbug · 28 Mar 2020 at 14:53

Robert896r1 said:
I'm only focusing on the single core score. Both mine and Guru 3D are showing a similar number which is more in line with what it should be.

The "reference" in CPU-Z is quite far below that. I'm not sure why but again it shows the value of doing your own testing.

"Far" is a strong word. Tho it is significant in the context we are talking about yes, anywho, there's 40 points between your 5Ghz 9900K and this 45 Watt Laptop chip in ST, given the ChipHell result are the same as Guru3D results in MT again this thing scores the same as an out of box 9900K, yes you can overclock the 9900K 300Mhz, it would be pulling near 200 Watts by that point and we are talking about a 45 Watt AMD Laptop chip here, overclock the 9900K to make it faster than an 8 core Upper Mid Range Ryzen Laptop chip..... saying that would be a joke.

humbug · 28 Mar 2020 at 15:09

CAT-THE-FIFTH said:
From what I read about Renoir,it is an SOC,as it means IF can be run at a lower speed,and needs less active links than having chiplets. This reduces power consumption and heat quite a bit apparently,meaning Renoir can boost higher for a given TDP than Zen2 CPUs.

Also better silicon, the 3700X runs at about 80 Watts stress testing. Its not hard to imagine that with some tweaks, yes like the IF given that its a monolithic chip, some fine-tuning to the voltage and higher quality silicon its quite possible to shave 35 watts off it and you have a mobile chip that performs like a top of the range 8 core Desktop chip.

CAT-THE-FIFTH · 28 Mar 2020 at 15:12

humbug said:
Also better silicon, the 3700X runs at about 80 Watts stress testing. Its not hard to imagine that with some tweaks, yes like the IF given that its a monolithic chip, some fine-tuning to the voltage and higher quality silicon its quite possible to shave 35 watts off it and you have a mobile chip that performs like a top of the range 8 core Desktop chip.

It will be interesting to see how much the L3 cache reduction will have on client application performance - I wonder if Renoir could also have lower latency too,ie,memory controller is nearer to the cores.

Robert896r1 · 28 Mar 2020 at 15:16

humbug said:
"Far" is a strong word. Tho it is significant in the context we are talking about yes, anywho, there's 40 points between your 5Ghz 9900K and this 45 Watt Laptop chip in ST, given the ChipHell result are the same as Guru3D results in MT again this thing scores the same as an out of box 9900K, yes you can overclock the 9900K 300Mhz, it would be pulling near 200 Watts by that point and we are talking about a 45 Watt AMD Laptop chip here, overclock the 9900K to make it faster than an 8 core Upper Mid Range Ryzen Laptop chip..... saying that would be a joke.

It is far. Without the quotes. 543 is a 4.6ghz single core rating. Are you going to downplay 400mhz all of a sudden now?

Here's a cpu validation also: https://valid.x86.fr/3t0cxp

humbug · 28 Mar 2020 at 15:17

Robert896r1 said:
It is far. Without the quotes. 543 is a 4.6ghz single core rating. Are you going to downplay 400mhz all of a sudden now?

Here's a cpu validation also: https://valid.x86.fr/3t0cxp

It about 8%, 45 Watt Laptop chip and its not even the highest end 8 core Laptop chip, that will be the 4900HS, vs 150 Watt top of the range Desktop part, what point are you trying to make?

CAT-THE-FIFTH said:
It will be interesting to see how much the L3 cache reduction will have on client application performance - I wonder if Renoir could also have lower latency too,ie,memory controller is nearer to the cores.

Tim from Hardware Unboxed will no doubt get one and put all this to the test, should be interesting to see how it does when the package is 'more like' Coffeelake.

CAT-THE-FIFTH · 28 Mar 2020 at 15:23

humbug said:
Tim from Hardware Unboxed will no doubt get one and put all this to the test, should be interesting to see how it does when the package is 'more like' Coffeelake.

It will be interesting to see how it performance against the Ryzen 7 3700X. If Renoir can have most of the application performance of desktop Zen2 CPUs despite the lower L3 cache,it's seriously impressive.

Robert896r1 · 28 Mar 2020 at 15:24

humbug said:
It about 8%, 45 Watt Laptop chip and its not even the highest end 8 core Laptop chip, that will be the 4900HS, vs 150 Watt top of the range Desktop part, what point are you trying to make?
.

The point is you repeatedly make false claims and don't cross reference anything for validation. Then run around spamming it as some fact.

You're simply wrong here. Again.

The bold part is junk you add your posts so have a narrative to shift to when called out on your underlying facts being wrong.

humbug · 28 Mar 2020 at 15:30

Robert896r1 said:
The point is you repeatedly make false claims and don't cross reference anything for validation. Then run around spamming it as some fact.

You're simply wrong here. Again.

The bold part is junk you add your posts so have a narrative to shift to when called out on your underlying facts being wrong.

I took the data from the ChipHell screenshot, that was 543 Points, you put up your own screenshot of your own 9900K and a Guru3D link to back it up stating that the score should be 583. The first thing i did was accept the score in the ChipHell screenshot was too low and conceded yours was accurate.

I can't do more than that. I mean what more do you want?

Robert896r1 · 28 Mar 2020 at 15:35

humbug said:
I took the data from the ChipHell screenshot, that was 543 Points, you put up your own screenshot of your own 9900K and a Guru3D link to back it up stating that the score should be 583. The first thing i did was accept the score in the ChipHell screenshot was too low and conceded yours was accurate.

I can't do more than that. I mean what more do you want?

It'd just be ideal if we can get people to cross reference data points, do some of their own analysis/fact finding before posting them. It'd save a lot of noise that engulfs forums. Worse yet, gives clicks to these guys who then go do it gain because the controversial scores spread across the internet faster.

If I overreacted, I apologize.

humbug · 28 Mar 2020 at 15:47

Robert896r1 said:
It'd just be ideal if we can get people to cross reference data points, do some of their own analysis/fact finding before posting them. It'd save a lot of noise that engulfs forums. Worse yet, gives clicks to these guys who then go do it gain because the controversial scores spread across the internet faster.

If I overreacted, I apologize.

You didn't over react, i appreciate your providing this data to portray an accurate picture of what we are talking about here, I agree accuracy is key. I didn't feel the difference was to such an extent that it destroyed the spirit of what i was saying, this is still a 45 Watt Laptop chip performing quite similar to a 9900K. So i didn't get and still don't get your continuation about arguing about the 45 points difference in the two screenshots, i already accepted yours are the ones that are accurate.