Discussion AMD Ryzen MegaThread! FAQ and Resources

[jaymc]

They were indeed ahead.. but there doesn't seem to be that much in it at the moment.. when production ramps up and clock speeds increase.. the gap is gonna close even more. Who know's they might even get ahead again.. wishful thinkin I know.
It's very close to the bone anyway.. an it does now look like they were hiding IPC with multitreaded benchmarks... ooohhh what a move from the underdog. Interesting times to say the least.

The server chip's an these ryzen/vega APU's are gonna be v interesting as well...

 

[8350rocks]

The information that I have is that XFR is not an all-core turbo.

There are eight frequencies defined in Zen. Four of them are for engineer's interest only like F_boot (the boot frequency) or F_min, the minimal working frequency (actually 550 MHz on desktop chips). The other four frequencies are of consumer interest: F_P0, F_TMT, F_TST, and F_MAX.

F_P0 is the base frequency. It was 3.15GHz for the engineering sample used by Canard in their preview. It is 3.6GHz for the 1800X model.

F_TMT is the all-core turbo. It was 3.3GHz for the engineering sample. I don't know its value for the 1800X. I know it is somewhat between 3.65GHz and 3.8GHz.

F_TST is the 1-core turbo. It was 3.5GHz for the engineering sample. It is 4.0GHz for the 1800X.

F_MAX is a second 1-core turbo. This frequency is not predefined by AMD. It is a variable frequency that depends on measurements of several on-the-fly parameters like the temperature. Different 1800X chips will have a different F_MAX depending on silicon lottery and the quality of the cooling you have installed. This is an auto-overclocking turbo. This is XFR.

I've had a thought about this (pure speculation here)- however do we know much about how the different frequency / turbo values are defined? I wonder if F_TMT is actually a derivative of F_MAX- whilst F_TST simply defines the default value for F_MAX unless the system determines a new (higher) value based on operating conditions? If F_TMT is defined as a proportion of F_MAX (e.g. F_TMT = 0.9 * FMAX) then as XFR increases the maximum single core turbo the max all core turbo would increase as well. Given the chip monitors power and thermals at a high rate the F_MAX would vary on loading anyway so I don't see any logical reason why it couldn't increase the all core turbo slightly if cooling allows for it (though the all core turbo would always be less than the highest single core frequency)...?

F_TMT is a fixed value pre-defined. It cannot be a percentage of a non-fixed value as F_MAX. For instance for the engineering sample used on Canard review F_TMT = 3.3GHz. I don't know the value of F_TMT for commercial chips, but if I had to guess I would say that F_TMT = 3.75GHz for the 1800X model, for instance.

X editions have no all core turbo...

The all core turbo on 1800X is 3.6 GHz.

Those two sentences contradict each other. Also 3.6GHz is the base clock, not the all-core turbo.

Exactly.

The baseclock is the all core turbo outside of XFR.

There *is* no all core turbo by conventional turbo specs. It is only XFR, and XFR is across all cores.

No. The base clock is not the all-core turbo. The all-core turbo is between the base and the 1-core turbo frequencies reported in the above table. For instance for the 1800X the base clock is 3.6GHz, the 1-core turbo is 4.0GHz and the all-core turbo is between 3.65GHz and 3.8GHz. I guess it must be about 3.75GHz.

For the ES tested by Canard the base clock was 3.15GHz, the all-core turbo was 3.3GHz, and the 1-core turbo was 3.5GHz.

How can XFR be turbo across all cores when F_TMAX is only defined as 1-core?

F_TMAX is single core turbo...it is *not* defined as XFR

 

[8350rocks]

cbr15 single thread leak, 1600x scores 146 ( @3.7ghz ? )
http://wccftech.com/amd-ryzen-5-1600x-processor-benchmark-cinebench-leak/

Man its about time all the latest leaks seem to point towards a nice product a score of 146 on single core is nice my CPU gets 172 at stock but that is with a 4.4Ghz turbo. If that is getting 146 at 3.7Ghz that is 17.8% lower with 18.9% less frequency(edit compared to my CPU) so again we are seeing Haswell level of performance.

As noted before we are seeing more and more benchmarks before launch one was from Anandtech even and its pointing towards Haswell level of IPC.

Waiting for some others to reply to these results here in this forum. Starting to think we will see a price war again. Man after 10 years of pure CPU boredom. Last time Amd was this close was back in 2005 i think.

It does appear to be haswell, or slightly better in some cases. So I think we can say between haswell and broadwell in IPC. Which I am extremely happy with...

 

[jwcrellin]

These AM4 socket AsRock boards look really nice. I used to write them off as a cut price brand but I may need to reconsider that assumption now.
http://hothardware.com/news/asrock-x370-fatal1ity-gaming-k4-killer-sli-taichi-ryzen-motherboards

 

[8350rocks]

Was looking to go gigabyte again for Ryzen, but that Taichi Asrock x370 looks pretty good...just wish I could find more info about the upcoming gigabyte boards to compare.

 

[$hawn]

Not sure if this has been mentioned before, but new Ryzen 8 core leaks - This is very promising if true!
Note: The source is wccftect

http://wccftech.com/amd-ryzen-7-1800x-8-core-benchmarks/

 

[Crumpet 1]

I think the AsRock K4 x370 is looking quite tasty.

 

[juanrga]

cbr15 single thread leak, 1600x scores 146 ( @3.7ghz ? )
http://wccftech.com/amd-ryzen-5-1600x-processor-benchmark-cinebench-leak/

Man its about time all the latest leaks seem to point towards a nice product a score of 146 on single core is nice my CPU gets 172 at stock but that is with a 4.4Ghz turbo. If that is getting 146 at 3.7Ghz that is 17.8% lower with 18.9% less frequency(edit compared to my CPU) so again we are seeing Haswell level of performance.

As noted before we are seeing more and more benchmarks before launch one was from Anandtech even and its pointing towards Haswell level of IPC.

Waiting for some others to reply to these results here in this forum. Starting to think we will see a price war again. Man after 10 years of pure CPU boredom. Last time Amd was this close was back in 2005 i think.

The overclocker from which WCFTECCHSWERdgdgZ stole content claims that Ryzen does 134 on ST @3.4GHz. Correcting for clocks Ryzen does 118 @3GHz This puts Ryzen just behind Haswell and about 9% above Sandy

It is an excellent score. I guess my ancient prediction

Zen IPC ~ Sandy Bridge IPC

has to be finally written like

RyZen IPC = 1.09 Sandy Bridge IPC

 

[juanrga]

Yeah it's looking good- just a side point- back in the 2003 - 2005 time frame *AMD were ahead* in IPC... Also fun fact, Athlon 64 is the oldest cpu that is capable of running Windows 10 Not bad for a chip that's over 12 years old.

Let me put that in a graph

Disclaimer: I didn't make the graph. The estimated CB score for Zen is too high.

 

[juanrga]

The information that I have is that XFR is not an all-core turbo.

There are eight frequencies defined in Zen. Four of them are for engineer's interest only like F_boot (the boot frequency) or F_min, the minimal working frequency (actually 550 MHz on desktop chips). The other four frequencies are of consumer interest: F_P0, F_TMT, F_TST, and F_MAX.

F_P0 is the base frequency. It was 3.15GHz for the engineering sample used by Canard in their preview. It is 3.6GHz for the 1800X model.

F_TMT is the all-core turbo. It was 3.3GHz for the engineering sample. I don't know its value for the 1800X. I know it is somewhat between 3.65GHz and 3.8GHz.

F_TST is the 1-core turbo. It was 3.5GHz for the engineering sample. It is 4.0GHz for the 1800X.

F_MAX is a second 1-core turbo. This frequency is not predefined by AMD. It is a variable frequency that depends on measurements of several on-the-fly parameters like the temperature. Different 1800X chips will have a different F_MAX depending on silicon lottery and the quality of the cooling you have installed. This is an auto-overclocking turbo. This is XFR.

I've had a thought about this (pure speculation here)- however do we know much about how the different frequency / turbo values are defined? I wonder if F_TMT is actually a derivative of F_MAX- whilst F_TST simply defines the default value for F_MAX unless the system determines a new (higher) value based on operating conditions? If F_TMT is defined as a proportion of F_MAX (e.g. F_TMT = 0.9 * FMAX) then as XFR increases the maximum single core turbo the max all core turbo would increase as well. Given the chip monitors power and thermals at a high rate the F_MAX would vary on loading anyway so I don't see any logical reason why it couldn't increase the all core turbo slightly if cooling allows for it (though the all core turbo would always be less than the highest single core frequency)...?

F_TMT is a fixed value pre-defined. It cannot be a percentage of a non-fixed value as F_MAX. For instance for the engineering sample used on Canard review F_TMT = 3.3GHz. I don't know the value of F_TMT for commercial chips, but if I had to guess I would say that F_TMT = 3.75GHz for the 1800X model, for instance.

X editions have no all core turbo...

The all core turbo on 1800X is 3.6 GHz.

Those two sentences contradict each other. Also 3.6GHz is the base clock, not the all-core turbo.

Exactly.

The baseclock is the all core turbo outside of XFR.

There *is* no all core turbo by conventional turbo specs. It is only XFR, and XFR is across all cores.

No. The base clock is not the all-core turbo. The all-core turbo is between the base and the 1-core turbo frequencies reported in the above table. For instance for the 1800X the base clock is 3.6GHz, the 1-core turbo is 4.0GHz and the all-core turbo is between 3.65GHz and 3.8GHz. I guess it must be about 3.75GHz.

For the ES tested by Canard the base clock was 3.15GHz, the all-core turbo was 3.3GHz, and the 1-core turbo was 3.5GHz.

How can XFR be turbo across all cores when F_TMAX is only defined as 1-core?

F_TMAX is single core turbo...it is *not* defined as XFR

I know it is single-core turbo. I wrote it first, as can be seen above in the quotations. I will rephrase my question:

How can XFR boost all cores at once if the only temperature-dependent frequency in Ryzen is the 1-core F_TMAX?

 

[juanrga]

Not sure if this has been mentioned before, but new Ryzen 8 core leaks - This is very promising if true!
Note: The source is wccftect

http://wccftech.com/amd-ryzen-7-1800x-8-core-benchmarks/

Another classic nonsensical article from WCFCFTHREFWRF. They confound a 1800X with an engineering sample. They claim that turbo is disabled, but don't prove it. They mix scores from different versions of the benchmark suite,....

 

[Crumpet 1]

You know Juanrga, i'm really unsure if you're just trying to keep the hype train on the rails, or if you're trying to derail it and cause a fire with burning disappointment...

I'm thankful for all the information in this thread, I mean ALL of it, including the fake leaked stuff as it keeps you aware of the situation.

I personally think we're looking at just under Haswell IPC with great multi-thread scaling, from all the current leaked benchmarks, and these next 9 days cannot pass quickly enough.

 

[juanrga]

Well, I wrote "This puts Ryzen just behind Haswell" and "It is an excellent score." No sure where is the problem here. On reporting that WCCFTECH doesn't have any idea about what they are doing? That they publish nonsense most of time? It is not new, and people in other forums is making similar claims about their supposed 1800X benches...

You need to take a closer look at what they (wccwhatever) used for results. Notice (PT8) next to some of the scores? Yes , that means it is taken from an older(different) version of Passmark and they are not directly comparable. Bad journalism unfortunately.

https://forums.anandtech.com/threads/summit-ridge-zen-benchmarks.2482739/page-239#post-38746221
https://forums.anandtech.com/threads/summit-ridge-zen-benchmarks.2482739/page-239#post-38746225
https://forums.anandtech.com/threads/summit-ridge-zen-benchmarks.2482739/page-240#post-38746250

I wrote a similar criticism directly on comments section of the WCCFTECHgwrs article and I got six upvotes

http://wccftech.com/amd-ryzen-7-1800x-8-core-benchmarks/#comment-3162558568

 

[Filipe_6]

dont ryzen get some APU?

 

[truegenius]

here lets do some maths ( surprised no one is doing walls of calculations about that cb leak ), if we assume that cb score to be actual cb score at 3.3/3.7ghz then
1136 @3.3ghz 6c/12t
so 189@3.3ghz 1c/2t
= 212 @3.7ghz
and we have 146 1c @3.7ghz
so 1c+smt = 212
so 1c+smt / 1c = 212/146 = 1.45
means smt is providing 45% performance

and people if you can keep those walls of quotes small or in spoiler then it will be better for thise with mobile devices ( as quoted text is now appearing as 1 alphabet per line )

 

[juanrga]

here lets do some maths ( surprised no one is doing walls of calculations about that cb leak ), if we assume that cb score to be actual cb score at 3.3/3.7ghz then
1136 @3.3ghz 6c/12t
so 189@3.3ghz 1c/2t
= 212 @3.7ghz
and we have 146 1c @3.7ghz
so 1c+smt = 212
so 1c+smt / 1c = 212/146 = 1.45
means smt is providing 45% performance

and people if you can keep those walls of quotes small or in spoiler then it will be better for thise with mobile devices ( as quoted text is now appearing as 1 alphabet per line )

The 1136cb score is for 3.4GHz. The SMT yield is 41%. CPU-Z also showed good SMT yields for Ryzen.

 

[Crumpet 1]

dont ryzen get some APU?

We have Summit Ridge, which (correc tme if i'm wrong) is essentially excavator APU on AM4 platform, i've seen some convincing gameplay on A12-9800 for console level gaming;

https://www.youtube.com/watch?v=JGftIcp2SC0
(hopefully not fake)

And we should be getting Raven Ridge some time in 2017, which should have 4 Ryzen cores in it.

 

[juanrga]

dont ryzen get some APU?

We have Summit Ridge, which (correc tme if i'm wrong) is essentially excavator APU on AM4 platform, i've seen some convincing gameplay on A12-9800 for console level gaming;

https://www.youtube.com/watch?v=JGftIcp2SC0
(hopefully not fake)

And we should be getting Raven Ridge some time in 2017, which should have 4 Ryzen cores in it.

Bristol Ridge --> Excavator APU on AM4
Summit Ridge --> Zen CPU on AM4
Raven Ridge --> Zen APU on AM4

 

[Crumpet 1]

Well I did say to correct me

 

[cdrkf]

Yeah it's looking good- just a side point- back in the 2003 - 2005 time frame *AMD were ahead* in IPC... Also fun fact, Athlon 64 is the oldest cpu that is capable of running Windows 10 Not bad for a chip that's over 12 years old.

Let me put that in a graph

Disclaimer: I didn't make the graph. The estimated CB score for Zen is too high.

Nice! Note assuming the 146 @ 3.6ghz is correct, the score at 4ghz would be about 160 (so a tad lower than suggested). Still that's an impressive jump.

It's also interesting how steady Intel's gains have been until a more significant jump more recently (I'm guessing that last jump was Haswell?). It really highlights just how much a jump Conroe was, and also that Nahlem and Sandy weren't actually as large of a gain in IPC as people would think (although I'm guessing the multi thread scores are a somewhat different story).

@jdwii, I think the thing that impresses me with Ryzen is I honestly expected Ryzen to position AMD much like they were with Phenom II vs Nahlem- where they were a bit behind in raw IPC but made up for it with additional physical cores (Phenom II X6 was very competitive with Nahlem in multi thread and sufficient in single thread at the time). What we actually have is a much more even fight core for core, clock for clock than I ever expected (although Intel do look to hold the clock speed advantage, especially at lower core counts). Still, interesting times ahead I think!

 

[juanrga]

Yeah it's looking good- just a side point- back in the 2003 - 2005 time frame *AMD were ahead* in IPC... Also fun fact, Athlon 64 is the oldest cpu that is capable of running Windows 10 Not bad for a chip that's over 12 years old.

Let me put that in a graph

Disclaimer: I didn't make the graph. The estimated CB score for Zen is too high.

Nice! Note assuming the 146 @ 3.6ghz is correct, the score at 4ghz would be about 160 (so a tad lower than suggested). Still that's an impressive jump.

It's also interesting how steady Intel's gains have been until a more significant jump more recently (I'm guessing that last jump was Haswell?). It really highlights just how much a jump Conroe was, and also that Nahlem and Sandy weren't actually as large of a gain in IPC as people would think (although I'm guessing the multi thread scores are a somewhat different story).

@jdwii, I think the thing that impresses me with Ryzen is I honestly expected Ryzen to position AMD much like they were with Phenom II vs Nahlem- where they were a bit behind in raw IPC but made up for it with additional physical cores (Phenom II X6 was very competitive with Nahlem in multi thread and sufficient in single thread at the time). What we actually have is a much more even fight core for core, clock for clock than I ever expected (although Intel do look to hold the clock speed advantage, especially at lower core counts). Still, interesting times ahead I think!

146 is for 3.7GHz, the turbo of the 1600X model they tested. Yes, it is an impressive ~60% jump.

 

[8350rocks]

That chart also shows just how much of a monster K8, especially thunderbird, was...

 

[cdrkf]

That chart also shows just how much of a monster K8, especially thunderbird, was...

Thunderbird was the first chip to 1ghz- but it was pre K8- must have been K7 surely?

 

[jdwii]

Was looking to go gigabyte again for Ryzen, but that Taichi Asrock x370 looks pretty good...just wish I could find more info about the upcoming gigabyte boards to compare.

I'm liking what MSI and Asrock have which are typically what i go with used to be a Asus guy until i noticed their performance per dollar was highly unfavorable to me.

 

[illegaloperation]

I'm liking what MSI and Asrock have which are typically what i go with used to be a Asus guy until i noticed their performance per dollar was highly unfavorable to me.

I definitely won't go with MSI again. The MSI X58 Platinum is the worst motherboard I've ever used.