Intel Core Family

[humbug]

AMD need to sort out needing specific RAM to work properly, as well as higher clock speeds. Btw has anyone guessed where clocks should be for the next chips? Probably 4.0 - 4.2 base?

The RAM i'm using isn't even on the Motherbords vendor list, it works properly, even at above its ratings, and its cheap stuff.

Intel can run higher clocked memory, it can also overclock lower rated memory higher, but thats very different from "needing specific RAM to work properly" Ryzen works properly with any memory.

Above 3200 / 3400 doesn't make any difference to Ryzen or Coffeelake, so its all just academic.

 

[MagicBoy]

With 24% higher clocks the 8700K scores 13% higher.

Not an exact science because of scaling but there is 12% IPC missing in that.

Your science is flawed - IPC is measured on a single core. I'll take your point as regards multicore scaling in certain workloads being better on Ryzen, but that's not IPC.

 

[humbug]

Your science is flawed - IPC is measured on a single core. I'll take your point as regards multicore scaling in certain workloads being better on Ryzen, but that's not IPC.

Can you even explain that? why would IPC measurement have to be restricted to single core, if a CPU with 12 compute threads is faster than another CPU also with 12 compute threads then does it also not have higher IPC? Why should that be excluded?

CPU architectures are the sum of all their parts, not just a selective bit irrelevant to most use cases.

 

[Martini1991]

In another episode of Ryzen getting faster everytime Humbug talks about it, Ryzen now has higher IPC than Coffelake.

 

[Martini1991]

The RAM i'm using isn't even on the Motherbords vendor list, it works properly, even at above its ratings, and its cheap stuff.

Intel can run higher clocked memory, it can also overclock lower rated memory higher, but thats very different from "needing specific RAM to work properly" Ryzen works properly with any memory.

Above 3200 / 3400 doesn't make any difference to Ryzen or Coffeelake, so its all just academic.

People have had loads of issues with Ryzen and memory, to say otherwise is ignorant.

I've personally not had any issues myself with the few Ryzen builds I've done but I'm obviously aware there's been and are issues.

 

[Martini1991]

Can you even explain that? why would IPC measurement have to be restricted to single core, if a CPU with 12 compute threads is faster than another CPU also with 12 compute threads then does it also not have higher IPC? Why should that be excluded?

CPU architectures are the sum of all their parts, not just a selective bit irrelevant to most use cases.

Wouldn't it be SMT/HT irrelevant to "most use cases"?

 

[humbug]

Wouldn't it be SMT/HT irrelevant to "most use cases"?

For that to be true one would have to assume that A: SMT is the reason for AMD's better multicore scaling and B that most use cases don't make use of it.

To answer A: who knows and i don't think it matters if its SMT scaling, Multicore scaling or a bit of both.
To answer B: no, SMT is so ingrained into applications now its used in most cases.

What is never used is one singular thread, so IPC restricted to one singular thread is not relevant.
Maxcon Cinema 4D never uses just one thread, it uses all, ST is just an academic benchmark.

 

[Martini1991]

For that to be true one would have to assume that A: SMT is the reason for AMD's better multicore scaling and B that most use cases don't make use of it.

To answer A: who knows and i don't think it matters if its SMT scaling, Multicore scaling or a bit of both.
To answer B: no, SMT is so ingrained into applications now its used in most cases.

We know SMT scales better than HT, in that the percentage gain from SMT is higher than HT. We know this as fact.
And since when has this magical massively parallel software roll out happened?

Either way, you're once again hyping Ryzen to heights that have never been reached before. My 2700 gets faster every day.

EDIT ; 6 Intel cores of a Coffeelake beat 6 Cores of a Ryzen in performance per clock, that's multithreaded not single threaded. It only changes when SMT/HT come into play.

Surely it'd be wrong to say that Ryzen has higher IPC when it'll be losing out until X amount of HT/SMT threads are being used.

 

[humbug]

We know SMT scales better than HT, in that the percentage gain from SMT is higher than HT. We know this as fact.
And since when has this magical massively parallel software roll out happened?

Either way, you're once again hyping Ryzen to heights that have never been reached before. My 2700 gets faster every day.

Do we? i haven't seen anything to be so convinced if its SMT, Multicore scaling or a combination of both, frankly i don't care beyond academic knowledge because the result is what it is, with the same number of threads Ryzen 2 has at least as higher IPC as Coffeelake.

Do you deny the results which confirm that?

 

[Martini1991]

Do we? i haven't seen anything to be so convinced if its SMT, Multicore scaling or a combination of both, frankly i don't care beyond academic knowledge because the result is what it is, with the same number of threads Ryzen 2 has at least as higher IPC as Coffeelake.

It's simple isn't it?
Run an i7 8700 against a Ryzen 1600 with SMT and HT turned off and see what the results are. I expect the Intel to be ahead at the same clocks. Run with HT/SMT and I expect the Ryzen to pull ahead because SMT nets higher gain than HT (In fact I'm sure benchmarks have already proved this previously).

EDIT : What results am I denying? My Ryzen doesn't have better than or equal to coffeelake IPC. I know it doesn't. The world knows it doesn't. However because of SMT netting higher than HT, I do have a higher performing chip when my chip's going 100% (But I've got a 2700, so I know that at the same clocks going 100% I kill a 6 core Intel).

Obviously, Ryzens problems are low clocks.

 

[humbug]

It's simple isn't it?
Run an i7 8700 against a Ryzen 1600 with SMT and HT turned off and see what the results are. I expect the Intel to be ahead at the same clocks. Run with HT/SMT and I expect the Ryzen to pull ahead because SMT nets higher gain than HT (In fact I'm sure benchmarks have already proved this previously).

The 1600 is Ryzen one, i'm talking about Ryzen 2.

Lets try this another way

Spoiler: Coffeelake vs Ryzen 2

The 2600 at 4.2Ghz with 12 threads scores 1482
The 8700 at 5.2Ghz with 12 threads scores 1664

5.2Ghz is 24% higher than 4.2Ghz, 1664 is 12% higher than 1482, so with 24% higher clock speed the 8700K is 12% faster, clock for clock the 8700K is in this instance 12% slower than the Ryzen 2600.

I'm not going to put that down to IPC because of clock speed scaling, i'm not suggesting that at 5.2Ghz Rzyen 2 would scale 1:1 and score 1482 + 24% = 1837, but i think we can agree the core clock scaling on Coffeelake is higher than 0.5:1 and with that is performing lower per clock rate than the 2600.

 

[kitfit1]

Obviously, Ryzens problems are low clocks.

Spot on and 100% agree. As a 2700x owner myself, all the benchies i have run since buying it prove that conclusively. It was the same with 1700 before as well. Once AMD get the clocks closer to or above 5Ghz on Ryzen, then Intel will have nothing to shout about. Until that happens though we are at the status quo.

 

[sideways14a]

News just out that Intels CEO has resigned because he was fiddling with another employee and intel have rules about that.. (bit like football?)

 

[Martini1991]

The 1600 is Ryzen one, i'm talking about Ryzen 2.

Lets try this another way

Spoiler: Coffeelake vs Ryzen 2

The 2600 at 4.2Ghz with 12 threads scores 1482
The 8700 at 5.2Ghz with 12 threads scores 1664

5.2Ghz is 24% higher than 4.2Ghz, 1664 is 12% higher than 1482, so with 24% higher clock speed the 8700K is 12% faster, clock for clock the 8700K is in this instance 12% slower than the Ryzen 2600.

I'm not going to put that down to IPC because of clock speed scaling, i'm not suggesting that at 5.2Ghz Rzyen 2 would scale 1:1 and score 1482 + 24% = 1837, but i think we can agree the core clock scaling on Coffeelake is higher than 0.5:1 and with that is performing lower per clock rate than the 2600.

Didn't we come to the conclusion that Ryzen 1 and "2" are nigh on the same IPC? ~3% gain to Ryzen "2".
Either way, Ryzen "2" is still not Coffeelake IPC.

Also, the way I suggested is simple and just works. Looking at graphs like yours and extrapolating figures just isn't the same.

SMT nets higher gains than HT.

 

[humbug]

Didn't we come to the conclusion that Ryzen 1 and "2" are nigh on the same IPC? ~3% gain to Ryzen "2".
Either way, Ryzen "2" is still not Coffeelake IPC.

Also, the way I suggested is simple and just works. Looking at graphs like yours and extrapolating figures just isn't the same.

SMT nets higher gains than HT.

I think we did establish Ryzen 2 does have a small IPC pump.

We can analyse it now.

I scored 1324 in Cinebench at 3,975Ghz, that's as high as i could get it, my sparing partner beat me with his 1600, he managed to get it to a flat 4Ghz and scored 1340, its all in the Cinebench thread, i'll dig it up if needs be.

So we will go from 4Ghz 1340, nice and easy, 4Ghz +5% = 4.2Ghz (Ryzen 2600 clock speed where it scored 1482) Ryzen 1600 score 1340 + 5% = 1407, the Ryzen 2600 score at the same clock speed was 1482, thats 6% higher, with scaling Ryzen 2 has probably gained about 7%

To double down on it, on that slide the Ryzen 1600 scored 1326 at 4Ghz.

Edited the IPC number, i was scaling as negative, should be positive.

 

[Martini1991]

When the IPC was compared in a review, it was 3%.

Because the gain is 3%. Why does every Ryzen get faster when you talk about it

 

[humbug]

When the IPC was compared in a review, it was 3%.

Because the gain is 3%. Why does every Ryzen get faster when you talk about it

In a review? here we have a review with tangible figures, if we are to take your reasoning as fact over this one we first need to know where you are coming from, so what review?

 

[Martini1991]

In a review? here we have a review with tangible figures, if we are to take your reasoning as fact over this one we first need to know where you are coming from, so what review?

We are retreading old ground again.
Whatever, my 2700 is 7% faster than my 1700 was. Never had a Cpu get faster because of a post on a forum, but miracles do exist.

 

[MagicBoy]

Can you even explain that? why would IPC measurement have to be restricted to single core, if a CPU with 12 compute threads is faster than another CPU also with 12 compute threads then does it also not have higher IPC? Why should that be excluded?

CPU architectures are the sum of all their parts, not just a selective bit irrelevant to most use cases.

By definition it's instructions per clock cycle. It's single core only - you can't measure the effects of configurations of cache, pipelines, ALU units etc across architectures or models within an architecture with different core counts. On your definition of IPC a hex core would be ~ 50% faster than a quad core of the same architecture. Which is ludicrous.

Go ahead and compare multithreaded benchmarks, it adds real world context as to how well an architecture scales. It's still a benchmark, not IPC.

 

[humbug]

By definition it's instructions per clock cycle. It's single core only - you can't measure the effects of configurations of cache, pipelines, ALU units etc across architectures or models within an architecture with different core counts. On your definition of IPC a hex core would be ~ 50% faster than a quad core of the same architecture. Which is ludicrous.

Go ahead and compare multithreaded benchmarks, it adds real world context as to how well an architecture scales. It's still a benchmark, not IPC.

We are not, we are measuring a 6 core with SMT vs a 6 core with SMT.

Cinebench is not just a benchmark, its Maxcon Cinema 4D which is a real world application, its the same sort of thing as Blender and some Autodesk and Adobe software.