Ryzen "2" ?

Martini1991 · 13 Feb 2018 at 16:19

humbug said:
Bulldozer? Bulldozer was a heap of ####, Intel's IPC was higher across the board, no ambiguity about that and its right to blatantly say with that "Intel has higher IPC" its true.

Ryzen is not Bulldozer, we are talking about Ryzen and Ryzen has SMT, you can't just make it go away.

Okay.

You say IPC is IPC.

If an application maxed out all 6 cores on the 8700K and the Ryzen 1600 (At the end of the day, this is a scenario that does happen in the real world. You can't force software to use more threads than it's capable of doing). But *didn't* to go the 7th thread onwards. At the same clock, is Ryzen performing the same as the 8700K?

humbug · 13 Feb 2018 at 16:34

Martini1991 said:
Okay.

You say IPC is IPC.

If an application maxed out all 6 cores on the 8700K and the Ryzen 1600 (At the end of the day, this is a scenario that does happen in the real world. You can't force software to use more threads than it's capable of doing). But *didn't* to go the 7th thread onwards. At the same clock, is Ryzen performing the same as the 8700K?

That's a loaded question, it doesn't, you example uses all 12 threads. You're re-framing the argument to suit your desired outcome.

The fact is Ryzen has SMT, that SMT gives it the performance to equal Coffeelake IPC, that's not a reason to disqualify SMT. Its a will on your part, a desire, but you don't get to do that.

Martini1991 · 13 Feb 2018 at 16:38

humbug said:
That's a loaded question, it doesn't, your example uses all 12 threads.

It's a loaded question I'll admit. But that doesn't make it invalid. You can't pretend those scenarios don't exist, it's a statistical fallacy to suggest every CPU intensive application is now capable of using more than 6 threads (Which is what you're doing by dismissing my question)

I'm not dismissing anything though, SMT's a good thing, I've said nothing contrary to that. But there's no "use all threads" button in software.

As far as Ryzen goes I put my money where my mouth was over half a year ago. But I'm not going to pretend it's better than it is.

gavinh87 · 13 Feb 2018 at 16:42

Coffeelake IPC, calling thestilt a shill

It gets better in here.

humbug · 13 Feb 2018 at 16:48

Martini1991 said:
It's a loaded question I'll admit. But that doesn't make it invalid. You can't pretend those scenarios don't exist, it's a statistical fallacy to suggest every CPU intensive application is now capable of using more than 6 threads (Which is what you're doing by dismissing my question)

It doesn't mater because its nothing to do with the arguments i made.

Cinedbench uses all the threads that it has available, 6 core 12 threads scoring about 1350 Points at 4Ghz, which it does, 8700K at 5Ghz scores about 1700, is a 26% difference in points vs a 25% difference in clock speed, that makes them about the same performance per threads per clock, yes if you switch SMT off on both the 8700K will score about 8% higher, that does not mean SMT on Ryzen should be disqualified, it is what it is.

gavinh87 · 13 Feb 2018 at 16:56

humbug said:
It doesn't mater because its nothing to do with the arguments i made.

Cinedbench uses all the threads that it has available, 6 core 12 threads scoring about 1350 Points at 4Ghz, which it does, 8700K at 5Ghz scores about 1700, is a 26% difference in points vs a 25% difference in clock speed, that makes them about the same performance per threads per clock, yes if you switch SMT off on both the 8700K will score about 8% higher, that does not mean SMT on Ryzen should be disqualified, it is what it is.

Score does not scale with frequency. What does your 1600 score in CB on single core?

Martini1991 · 13 Feb 2018 at 16:58

gavinh87 said:
Score does not scale with frequency. What does your 1600 score in CB on single core?

Cinebench is pretty good at scaling, with SMT's advantage, I don't exactly doubt the results to a massive degree (Although I'd still expect Coffeelake to be ahead by a few percent), albeit to talk about IPC and not actually be comparing them at the same clock speed is a bit iffy.

Single core CB, obviously the Coffeelake will be ahead.

Same clock 12 threads versus 12 threads. Why don't you both run that? To me it doesn't change things. Coffeelake's still the better gamer of the two.

gavinh87 · 13 Feb 2018 at 17:00

Martini1991 said:
Cinebench is pretty good at scaling, with SMT's advantage, I don't exactly doubt the results to a massive degree, albeit to talk about IPC and not actually be comparing them at the same clock speed is a bit iffy.
Single core CB, obviously the Coffeelake will be ahead.

Same clock 12 threads versus 12 threads. Why don't you both run that?

This is what I'm trying to achieve, remove intel's clockspeed advantage and lets see how it fares at 3.9ghz or whatever. Toms hardware shown that above 4.8ghz coffeelake no longer scales linearly in CB.

humbug · 13 Feb 2018 at 17:01

gavinh87 said:
This is what I'm trying to achieve, remove intel's clockspeed advantage and lets see how it fares at 3.9ghz or whatever. Toms hardware shown that above 4.8ghz coffeelake no longer scales linearly in CB.

My CB score at 3.9Ghz was 1329, something like that, i'm trying to find it now...

gavinh87 · 13 Feb 2018 at 17:02

humbug said:
My CB score at 3.9Ghz was 1329, something like that, i'm trying to find it now...

And the single?

Martini1991 · 13 Feb 2018 at 17:03

gavinh87 said:
And the single?

Ha.

I don't see the point. You know it'll be ahead.

humbug · 13 Feb 2018 at 17:04

gavinh87 said:
And the single?

Don't know, 160 something? didn't do a run of that in the end, its not what we are after anyway..

Here: 1324. @ 3.9Ghz with 2933Mhz Ram.

https://forums.overclockers.co.uk/t...nchmark-thread.18546142/page-50#post-31571186

gavinh87 · 13 Feb 2018 at 17:05

humbug said:
Don't know, 160 something, didn't do a run of that in the end, its not what we are after anyway..

Here: 1324. @ 3.9Ghz with 2933Mhz Ram.

https://forums.overclockers.co.uk/t...nchmark-thread.18546142/page-50#post-31571186

Well for IPC, it kind of is.

Martini1991 · 13 Feb 2018 at 17:06

To be honest, Cinebench single core for Ryzen's closer to Skylake than it is Haswell. It's actually not a bad showing for it (Hence I had no issue buying a 1700 with the intention to replace it to AMD's successor)

humbug · 13 Feb 2018 at 17:07

gavinh87 said:
Well for IPC, it kind of is.

Are you going to give me a run at the same speed or not?

gavinh87 · 13 Feb 2018 at 17:08

humbug said:
Are you going to give me a run at the same speed or not?

I'm asking for your single and multi core scores lol. Give me a sec i need to reboot to downgrade my CPU speeds.
Any chance of a unigene heaven bench at 720p?

humbug · 13 Feb 2018 at 17:11

gavinh87 said:
I'm asking for your single and multi core scores lol. Give me a sec i need to reboot to downgrade my CPU speeds.
Any chance of a unigene heaven bench at 720p?

Given how Pascal GPU boost 3 behaves that's tricky, its impossible to nail it down to a set and constant speed so you can't get apples for apples comparisons.
Single threaded performance Coffeelake is faster, we know that.

gavinh87 · 13 Feb 2018 at 17:14

humbug said:
Given how Pascal GPU boost 3 behaves that's tricky, its impossible to nail it down to a set and constant speed so you can't get apples for apples comparisons.

A minor difference in GPU clockspeed from the same chip wont make much of a difference in all fairness.

Done

So no, ryzen does not have coffeelake IPC. Even with SMT being better than HT
Its close in multi but behind in single.

humbug · 13 Feb 2018 at 17:18

gavinh87 said:
A minor difference in GPU clockspeed from the same chip wont make much of a difference in all fairness.

Done

So no, ryzen does not have coffeelake IPC. Even with SMT being better than HT
Its close in multi but behind in single.

1324 vs 1340

I make that a 1.2% difference, agreed?

gavinh87 · 13 Feb 2018 at 17:20

humbug said:
1324 vs 1340

I make that a 1.2% difference, agreed?

Agreed, now look at the single. Lets try with SMT/HT off