*** AMD "Zen" thread (inc AM4/APU discussion) ***

[AndreiD]

Yikes, apparently the guy who provided the most optimal memory timings for Ryzen CPUs and has also provided a wealth of other knowledge on the Zen architecture is an "Intel shill" now
You didn't debunk anything, The Stilt has been providing a lot of good data on Summit Ridge & Pinnacle Ridge.

 

[humbug]

Yikes, apparently the guy who provided the most optimal memory timings for Ryzen CPUs and has also provided a wealth of other knowledge on the Zen architecture is an "Intel shill" now
You didn't debunk anything, The Stilt has been providing a lot of good data on Summit Ridge & Pinnacle Ridge.

Well i can certainly see who someone who writes a lot can have an air of genius about them, it doesn't mean anything they say is actually right, his apparent findings contradict what is now common knowledge, so while you may be impressed by him i'm not.

Stop holding him up like some sort of demigod i don't see his data in the same light as you.

 

[triss]

Yikes, apparently the guy who provided the most optimal memory timings for Ryzen CPUs and has also provided a wealth of other knowledge on the Zen architecture is an "Intel shill" now
You didn't debunk anything, The Stilt has been providing a lot of good data on Summit Ridge & Pinnacle Ridge.

Some people turn off when things arent shown the way they like. I'd also of thought its odd as a "intel shill" why on earth the silts done some much work on Ryzen timings and helped many ryzen users.

 

[gavinh87]

Well i can certainly see who someone who writes a lot can have an air of genius about them, it doesn't mean anything they say is actually right, his apparent findings contradict what is now common knowledge, so while you may be impressed by him i'm not.

Stop holding him up like some sort of demigod i don't see his data in the same light as you.

He is one of the reasons ryzen could actually run ram at a half decent speed in the early days.
Far from an intel shill :facepalm:

 

[kitfit1]

He is one of the reasons ryzen could actually run ram at a half decent speed in the early days.
Far from an intel shill :facepalm:

+1
Spot on. In fact The Stilt and 1usmus between them have done more to make ram clocking accessable to Ryzen users than anybody.

 

[humbug]

Some people turn off when things arent shown the way they like. I'd also of thought its odd as a "intel shill" why on earth the silts done some much work on Ryzen timings and helped many ryzen users.

He has not done anything to help anyone, he's also running the 3200Mhz Ram at 2666Mhz despite Ryzen being rated for 2933Mhz so he's messed up there already.

 

[gavinh87]

He has not done anything to help anyone, he's also running the 3200Mhz Ram at 2666Mhz despite Ryzen being rated for 2933Mhz so he's messed up there already.

Are you saying the stilt has done nothing to help anyone? Head over to oc.net and look at what he has done over there.
Asus have even got profiles in their newer bios's literally called "thestilts profile" FFS

I urge you to find someone else that has done more for ryzen than thestilt has. Infact, go over to oc.net and tell him hes messed up

 

[humbug]

Are you saying the stilt has done nothing to help anyone? Head over to oc.net and look at what he has done over there.
Asus have even got profiles in their newer bios's literally called "thestilts profile" FFS

I urge you to find someone else that has done more for ryzen than thestilt has. Infact, go over to oc.net and tell him hes messed up

I don't see it, if he's all that you would have something to show me instead of just repeating the same crap over and over again without actually providing an ounce of facts.

Anyone can tell any noob how to set up custom ram timings, its not difficult.

I googled "Asus thestilts profile" nothing...

 

[gavinh87]

The difference is, humbug is you don't want to see it. You don't like his results therefore he is wrong.
There were many people in the early days of ryzen that certainly weren't "noobs" but he helped many people understand what is required. You were late to the party so you missed all of this.
Spend some time over there and you shall see, what he doesn't know about ryzen isn't really worth knowing. In fact he has been much more helpful than AMD themselves when it comes to overclocking.

 

[humbug]

The difference is, humbug is you don't want to see it. You don't like his results therefore he is wrong.
There were many people in the early days of ryzen that certainly weren't "noobs" but he helped many people understand what is required. You were late to the party so you missed all of this.
Spend some time over there and you shall see, what he doesn't know about ryzen isn't really worth knowing. In fact he has been much more helpful than AMD themselves when it comes to overclocking.

Ok, so why does he not use his own memory profiles when performance testing?

 

[gavinh87]

Ok, so why does he not use his own memory profiles when performance testing?

Ask him, you are the one doubting him. All CPUs are tested at the same memory frequency so what is the issue? Would you still be arguing with the results if they were run at 3200? Of course you would, because you do not like them.
IMO without him, elmor from asus and 1usmus. Ryzen would have stayed in that **** storm of a launch state for a LOT longer than it was. But again, you missed all this.

 

[humbug]

Ask him, you are the one doubting him. All CPUs are tested at the same memory frequency so what is the issue? Would you still be arguing with the results if they were run at 3200? Of course you would, because you do not like them.
IMO without him, elmor from asus and 1usmus. Ryzen would have stayed in that **** storm of a launch state for a LOT longer than it was. But again, you missed all this.

That's an easy shrug-off.
The point is his reviews look like a noob did them, for a start the memory he used is Samsung B-Die 3200Mhz and yet he ran them at 2666Mhz despite Pinnacle Ridge being rated for 2933Mhz, why would you use lower than rated clocked ram?
Some of his reviews are IO bound and nothing to do with CPU performance and yet he presents the results as CPU performance. Compression runs reading from disk as CPU performance, school boy error.
He didn't do any multithread testing, like someone with no Ryzen experience just assumed one singular core performance is the same at multicore performance if you just multiply the performance, nothing could be further from the truth.....

The review looks like a noob did it, its full of incorrect assumptions and flawed testing methods.

He might be good at setting up memory, but he didn't bother here and he's no reviewer,

 

[Killer7xx]

Ask him, you are the one doubting him. All CPUs are tested at the same memory frequency so what is the issue? Would you still be arguing with the results if they were run at 3200? Of course you would, because you do not like them.
IMO without him, elmor from asus and 1usmus. Ryzen would have stayed in that **** storm of a launch state for a LOT longer than it was. But again, you missed all this.

If I tested a 1080Ti using a Pentium 4 CPU and said that it only performs as well as a 560Ti, that would be a bit underhanded and dishonest wouldn't it? Even if the bottleneck would mean that indeed the 1080Ti performs the same as a 560Ti in that scenario, we all know that the 1080Ti is capable of more without the bottleneck.

2666 MHz RAM is a bottleneck for Ryzen and also for Intel to a much lesser extent, it's not really a fair review if you test with a bottleneck and then present findings where Ryzen is seemingly 20% deficient in IPC. We all know that is not the case, otherwise Ryzen must be using some black magic to be so close to Intel in CPU intensive tasks.

 

[humbug]

I'll give you an example..

7zip compression.

Spoiler: 7zip

in doing that you're reading data off the disk to stick in a Zip folder, you can only do that as fast as your drive can read it.

If you want to do 7Zip testing do Decompression, that extracts the folder in memory and then writes it to disk, its the performance of the in memory extraction that 7Zip benchmarking measures and its whats important to CPU performance.

He didn't do any testing like that, not even with 7Zip.

Honestly one of the most amateurist mistakes one can make when doing reviews.

 

[gavinh87]

I'll give you an example..

7zip compression.

Spoiler: 7zip

in doing that you're reading data off the disk to stick in a Zip folder, you can only do that as fast as your drive can read it.

If you want to do 7Zip testing do Decompression, that extracts the folder in memory and then writes it to disk, its the performance of the in memory extraction that 7Zip benchmarking measures and its whats important to CPU performance.

He didn't do any testing like that, not even with 7Zip.

Honestly one of the most amateurist mistakes one can make when doing reviews.

There are plenty of multicore tests on there. Take the blinkers off. Nobody has done as much IPC testing as he has.
Nor does he claim to be a reviewer. Nobody here has said that.
That 7zip test is the only IO test on there. He did another compression test in winrar and if you actually look at the tests, the results are vastly different to each other. Not what I'd call an IO bottleneck.
Again, if you actually look at the tests you will see he went all the way to 3466mhz on all platforms.
Here is an extract

Despite the extremely welcome latency improvement in Pinnacle Ridge, the memory latency is unfortunately still < 38% higher on average (2133-3466MHz) than on its closest rival from Intel (Coffee Lake).

Whether you believe his testing or not, I couldn't care less. But calling him a shill is just outright stupid. Take some time to see the work he has done for AMD to see that he is far from it.

 

[humbug]

WinRar is also IO bound, more so than 7Zip because even in decompression it writes straight to disk ^^^^ .

If I tested a 1080Ti using a Pentium 4 CPU and said that it only performs as well as a 560Ti, that would be a bit underhanded and dishonest wouldn't it? Even if the bottleneck would mean that indeed the 1080Ti performs the same as a 560Ti in that scenario, we all know that the 1080Ti is capable of more without the bottleneck.

2666 MHz RAM is a bottleneck for Ryzen and also for Intel to a much lesser extent, it's not really a fair review if you test with a bottleneck and then present findings where Ryzen is seemingly 20% deficient in IPC. We all know that is not the case, otherwise Ryzen must be using some black magic to be so close to Intel in CPU intensive tasks.

IPC is very dependant on context, IPC is only ever relevant to what you use the CPU for at any given time,

So adding everything up into a single slide chart is a good way to round everything off nicely and thats fine but he should also be clear that its not the be all and end all, with that he is very absolute with his findings and fails to mention that Ryzen scales much better than Intel in multicore workloads, i suspect he doesn't even know.

With that he claims SkyLake-X has the highest IPC, in very very specific workloads where higher FP cache helps it is, it seems that's almost entirely what he based the application choices on, if i wanted to show that actually Skylake-X has higher IPC how he did it is exactly how i would do it.

With that he didn't do any Game testing at all, had he done he could have shown that overall Skylake-X has significantly lower IPC than Ryzen.

 

[gavinh87]

He actually does talk about SMT and how it is superior to HT, but again you must have missed it.
Winrar test cannot be IO bound as the results are vastly different.
So ryzen beats skylake x by a small % and its a win.
Coffeelake beats ryzen by a small % and its a draw
I'm ******* done with this forum lol, ill stay around for the MM but seriously, its getting worse around here with all the BS that gets spouted.

 

[humbug]

There are plenty of multicore tests on there. Take the blinkers off. Nobody has done as much IPC testing as he has.
Nor does he claim to be a reviewer. Nobody here has said that.
That 7zip test is the only IO test on there. He did another compression test in winrar and if you actually look at the tests, the results are vastly different to each other. Not what I'd call an IO bottleneck.
Again, if you actually look at the tests you will see he went all the way to 3466mhz on all platforms.
Here is an extract

Whether you believe his testing or not, I couldn't care less. But calling him a shill is just outright stupid. Take some time to see the work he has done for AMD to see that he is far from it.

He said as you quoted...

Despite the extremely welcome latency improvement in Pinnacle Ridge, the memory latency is unfortunately still < 38% higher on average (2133-3466MHz) than on its closest rival from Intel (Coffee Lake).

Irrelevant white-noise.

Intercore latency only matters to the CPU its self, its only relevant to its self, comparing it to coffeelake or anyother CPU is just academic at best, at worst its trying to use an irrelevant comparison in a misleading way to drive home an agenda.

Intercore Latency matters most with things like gaming, Ryzen's Intercore latency is also higher than SkyLake-X and yet Ryzen has overall higher gaming performance, Higher gaming IPC.

Its a meaningless sentence.

So ryzen beats skylake x by a small % and its a win

Ryzen in that test is also 400Mhz lower clocked.

2600 @ 4.2Ghz
7800 @ 4.6Ghz.

This IPC difference is 10 to 15%, or somewhere in the middle of that.

 

[subbytna]

 

[gavinh87]

You are contradicting yourself.

IPC is very dependant on context, IPC is only ever relevant to what you use the CPU for at any given time

And then....

With that he didn't do any Game testing at all, had he done he could have shown that overall Skylake-X has significantly lower IPC than Ryzen

CB15 single thread results show when CPU's are clocked at 3.5ghz the 2600 is 4.1% behind coffeelake.
CB15 single thread @ stock speeds show the 2600 is 16.5% behind.


You have taken one gaming test and come to the conclusion that skylake x is 10-15% behind pinnacle ridge.......
The stilt has run many, single and multicore tests and his results show a 15.6% difference in IPC.


Meh, cba with people with blinkers on. Believe what you want. The more open minded people on this forum know and respect thestilts findings over yours.