AMD Zen FX - information, benchmarks, experience of users

[chew*]

Nope. Answers the cache theory to.

[vario]

Well, sheduler working properly , doesnt mean theres no hit when thread bounce between CCX`s ...

[BenchZowner]

Well, sheduler working properly , doesnt mean theres no hit when thread bounce between CCX`s ...

Sure, but that can't be fixed.
So here comes nothing.

Clearly, the lower core count parts are still unable to clock any higher than the 8 core parts, thus AMD is taking its time before releasing those hoping for the process to mature ( I can't believe the "Ryzen+" in September rumors ).

[vario]

Sure, but that can't be fixed.
So here comes nothing.

Clearly, the lower core count parts are still unable to clock any higher than the 8 core parts, thus AMD is taking its time before releasing those hoping for the process to mature ( I can't believe the "Ryzen+" in September rumors ).

Im 99,99% positive this can be fixed. Cores are numbered, and CCX`s , system knows to which CCX`s which cores are assigned, system then bounces threads only on cores that are in the same CCX that the thread was started. Boom. fixed.
And it has been shown that performing a task all in the same CCX is faster, there were benchmarks showing 2+2 o4 0+4 bios setting.0+4 was faster.
But hey, its AMD, they clearly know how to shoot themselves in the foot. Pretty much every launch for them is some kind of self hurt.

[BenchZowner]

When the number of threads utilized by the app surpasses the core count of a CCX even by a single core, then it'll get penalized again.
That's what I mean it can't be fixed.
You can keep an app ( I would love to see them making this though, because it is easier said than done, especially if we're talking about everything, in games, in productivity apps, etc ).

They kinda went with the design they were laughing about when Intel was using it ( Core 2 Quad )

[vario]

When the number of threads utilized by the app surpasses the core count of a CCX even by a single core, then it'll get penalized again.
That's what I mean it can't be fixed.
You can keep an app ( I would love to see them making this though, because it is easier said than done, especially if we're talking about everything, in games, in productivity apps, etc ).

They kinda went with the design they were laughing about when Intel was using it ( Core 2 Quad )

Well, there are cores and also SMT threads, and add some kidn of balancing to it, should be 90% efficient. Still better than nonstop random L3 hits.
At lest some kinda "Booster" app, that would had priviliges to do it for certain apps/drivers.Would help with games.But it seems AMD solution is to bury its head in the sand .
I mean its probably doable setting affinity or using process lasso, but not automated way is gonna be hard and not always useful (when thread is short lived etc.)
I dont know. Seems like a wasted opportunity.

[BenchZowner]

I've seen games ignoring affinity set via user/software before.
That's why I'm saying it's not going to be easy ( or done at all :p )

[vario]

I've seen games ignoring affinity set via user/software before.
That's why I'm saying it's not going to be easy ( or done at all :p )

Thats why i say not automated way is hard ;-) . AMD driver should be able to override anything else.Anyhow, on reddit i read that AMD IS working on some kind of driver fo power states, as it is now, balanced is slow, and performance disables turbo (thats what i read). So there is hope for a bit better stock results at least.

[BenchZowner]

Can't see how High Performance preset could disable turbo.
Anyway, if all they do is jack up a batch file mock up to configure the advanced power settings in control panel, it will be hilarious.

I've been customizing the power plan in Windows since XP, can't help but wonder how people even some so called reviewers never fiddled with those

[MattiasNYC]

AMD concludes there's no Windows 10 scheduler issue at all

https://community.amd.com/community/...mmunity-update

Not great news for those expecting a performance increase

Right, but I think there's a bit of nuance missing in the commentary in general.

People think that people are saying that as long as the scheduler is "fixed" we'd gain a lot of performance, and that then means the scheduler isn't working properly. So now, people are saying "ah-hah! No big gains so people were wrong!". But, I think the real issue is still there and it takes two to tango; it's still the design of the CPU with two CCX - AND - the scheduler moving threads across those CCX.

So, I still think that if threads can be kept in one complex when it makes sense, that would be an improvement. And that's the thing; it's not a "scheduler issue", it's a matter of what the likelihood is of optimizing either W10 scheduling or software doing the job.

I'm still looking at audio and the huge discrepancy for us pro-audio guys:

- For those that need a very low latency the penalty is large compared to Intel CPUs. We count the amount of load we can put on the CPU before audio breaks up, all at low latency, and there's a big difference between the brands.

- Yet for those of us (me included) who don't need low latency the performance-per-dollar all of a sudden looks great with the 1700.

IF some developers on the audio software side were able to lock threads down in the respective CCX, and that's a big "if" btw, it'd be a no-brainer to get the Ryzen. Now it depends.... but I digress...

[techtrancer]

Some Information why the coretemp of 1700 is 20C lower than 1700X & 1800X

https://community.amd.com/community/...mmunity-update

[vario]

https://www.youtube.com/watch?v=U9DE83lMVio
https://www.youtube.com/watch?v=XAXS8rYwGzg
http://www.hardware.fr/articles/956-...ire-suite.html

AMD-> "There is no problem!"

[chew*]

Im not completely buying the temperature thing. The 1700x,1800x still makes my heatsink warmer.

[techtrancer]

How did you test this?

[chew*]

How did you test this?

temp sensor provided in the motherboard box tossed in heatsink.

Even if gigas temps sensors are not accurate if i'm using the same board same sensor and get a higher reading vs a lower it is a reliable enough method to at least tell if one part is indeed putting out more heat.

[StAndrew]

I wonder how the scheduling will work with 6 core processors if the CCX's are not uniform...

I'm assuming all the 4 cores will just be a single CCX or do we expect them to also be cut down from an 8 core die?

[techtrancer]

temp sensor provided in the motherboard box tossed in heatsink.

Even if gigas temps sensors are not accurate if i'm using the same board same sensor and get a higher reading vs a lower it is a reliable enough method to at least tell if one part is indeed putting out more heat.

And what did you think why?

[chew*]

Leakage

[chew*]

I wonder how the scheduling will work with 6 core processors if the CCX's are not uniform...

I'm assuming all the 4 cores will just be a single CCX or do we expect them to also be cut down from an 8 core die?

That is the question of all questions concerning 4 core parts.

If one cpu is 2+2 another 3+1 and or 1+3 another 4+0 or 0+4 will adopters of 4c/8t parts all have varying results due to which cores are disabled.

I would bet odds wise the 0+4 or 4+0 would be a rarity however it is certainly possible.

[techtrancer]

Leakage

But should not have normaly the not so good samples more?

[Nelly]

Holy poo! I've just read the Ryzen 5 CPUs will be available for the retail channel within the next couple of weeks.

No messing about from AMD lol

Hi there


AMD have authorised OcUK to drop prices on all older generation CPU to make sure it is firmly below Ryzen 7 and also the upcoming Ryzen 5 which shall be available in next couple of weeks.

Source

[bowman]

Would be cool, and good for AMD if true, but I doubt it. Probably just nonsense from OCUK.

[vario]

Would be cool, and good for AMD if true, but I doubt it. Probably just nonsense from OCUK.

Thats totally inline with what AMD said about R5 launch, its april one way or another, may be 2 weeks, may be 5.
As for 4 core parts, with 99% probability they will use cut down R7 , in what way they are gonna do it however, no idea. One would assume they should have one functional CCX intact and cut the rest, but depends on what they are working with, could be any diff configuration.I just hope its one solid CCX.Although if it would be 2+2, these cores would have more L3 cache for themselves.
I wonder however, AMD allows unlocking from time to time if they need some positive press. Maybe they will be unlockable :-D

[KiSUAN]

Holy poo! I've just read the Ryzen 5 CPUs will be available for the retail channel within the next couple of weeks.

I can confirm this (I have no interest in spreading FUD), a distributor in Miami already offer me and I'm in South America so we are the last to get anything (Ryzen 7 is just arriving next week in my country )...



This are the prices they send me.

Ryzen 5 1600X - $229.59
Ryzen 5 1600 - $198.98
Ryzen 5 1500x - $173.47
Ryzen 5 1400 - $153.06

[vario]

I can confirm this (I have no interest in spreading FUD), a distributor in Miami already offer me and I'm in South America so we are the last to get anything (Ryzen 7 is just arriving next week in my country )...



This are the prices they send me.

Ryzen 5 1600X - $229.59
Ryzen 5 1600 - $198.98
Ryzen 5 1500x - $173.47
Ryzen 5 1400 - $153.06

So, this kinda confirms 1500X will be 2 CCX`s, look at the cache size.