Speculation on Ryzen Overclocking

[unseenmorbidity]

I'm expecting ~4.5Ghz on air with decent examples...

Those clock speeds are kind of rare for Broadwell-e aren't they!? That is a high expectation imo.

if you like to gamble for cheap performance the 6c/12t is the place to put your money. You might get a Golden Sample chip that had a faulty core or you might get a dud since they are all cut down chips. With the 8c/16t I really think you will get what you pay for, Cheapest model = one that has 8 functioning cores and that's about it, especially this early in the run.

But the 6 cores have different models as well. So, it's not like any chip with faulty cores is thrown a big random bag. They are likely binned as well.

 

[CatMerc]

Those clock speeds are kind of rare for Broadwell-e aren't they!? That is a high expectation imo.

But this isn't Broadwell.

 

[bjt2]

Which is not based on real information or data, but on assumptions based on technologies used in various different products

It might be accurate for AVFS and clock stretching in general, but not for XFR.

Bristol ridge shadow p-states, according to the papers, works almost like XFR, except that is locked to programmed p-states. It's very easy to break these chains...

It is an accurate summary of the papers...

 

[Atari2600]

XFR stands for Extended Frequency Range and that's exactly what it is, literally.

So literal interpretations of that could be (probably missing some):

(1)Extended Frequency values, which could be:
(i) bigger differentiation between lowest and highest clocked cores on part
(ii) higher upper limit on clocks than indicated
(iii) bigger differentiation in clocks between core and uncore parts within each logical unit to optimise power

(2)Extended duration values at turbo frequencies before throttling by optimising thermals of CPU via per core throttling


Given, 1(i) and 2 are already largely done in existing chips, you'd imagine its between 1(ii) and 1(iii).

 

[Rifter]

Yeah im writing off XFR, for the simple reason AMD had gone into practically zero detail about it other than one very confusing slide and a few words that explain nothing, if it was the best thing since sliced bread AMD would not be so silent about it and deliberately confusing. Its just marketing BS IMO.

Which is fine they are unlocked so its not like we need any kind of help to OC we can just go oldschool, which is how i prefer it anyways.

Seems like most agree with the fact that at the beginning binning will play a large roll and higher end SKU's will OC better, at least at the start.

I also agree with the poster who said that good example will hit 4.5Ghz, at least i hope so.

Its been a while since there has been a chip worth overclocking, intels K series with its 10-20% OC's has been so boring.

 

[unseenmorbidity]

Rumor from someone that supposedly has ES samples,

- IPC is at the least Ivy Bridge-E and higher
- SMT for Ryzen is more efficient then Intel's HyperThreading
- Ryzen has no cold Bug
- Cinebench R15 hits 145 single thread @ 3.4GHz on ES, earlier models hit 130-140, retails should hit 140-150
- Most Ryzen ES samples hit 4.3-4.5GHz MAX on Air with all core enabled
- Intel is testing out Skylake-X, and beats out current 6950X with 8C/16T because it can hit higher clocks.


http://www.overclock.net/t/1623292/lets-talk-about-a-ryzen-es

 

[FrostigerFinger]

I'd be surprised if they didn't bin like crazy at first. Half a year down the line though ...

 

[Arachnotronic]

So... Overclocked?

The problem with the "overclocked" hypothesis is that the silicon needs to be validated to run at the frequencies that XFR will enable. If you validate your silicon for max turbo of 4GHz, then running at 4.1GHz is risky because the silicon might not be able to handle it.

I don't think XFR will be "overclocking beyond the boost limits."

 

[unseenmorbidity]

The problem with the "overclocked" hypothesis is that the silicon needs to be validated to run at the frequencies that XFR will enable. If you validate your silicon for max turbo of 4GHz, then running at 4.1GHz is risky because the silicon might not be able to handle it.

I don't think XFR will be "overclocking beyond the boost limits."

How does one validate that? CPUs have a life expectancy of like 20 years.

How is this different from any other CPU being "validated" at a boost frequency!?

 

[bjt2]

Actually even Bristol Ridge has a limited XFR-like technology: shadow p-states. It's limited because only the fixed p-states (up to max turbo freq) are allowed, but theoretically even at max load the max turbo can be achieved if the cooler is good enough. I explained all here: http://www.bitsandchips.it/52-english-news/8058-xfr-and-its-anchestors

 

[unseenmorbidity]

1600x might be the best pick

 

[CHADBOGA]

I cannot say anything further about the subject. As soon as the reviews are out, the functionality of the XFR will immediately become clear to everyone.

Even to Juanrga?

 

[.vodka]

Even to Juanrga?

You mean, he could actually stop his multi year long ranting on Feb28 - March 3?

Nah... you're asking for a miracle. Do you believe in miracles?

 

[Zucker2k]

The problem with the "overclocked" hypothesis is that the silicon needs to be validated to run at the frequencies that XFR will enable. If you validate your silicon for max turbo of 4GHz, then running at 4.1GHz is risky because the silicon might not be able to handle it.

I don't think XFR will be "overclocking beyond the boost limits."

It very well could and will. Base clock and turbo clocks are determined using certain variables of which thermals are one critical component. What happens if the thermal variable changes significantly? You can dial up both power and frequency to a limited range. XFR, I'm sure when turned on (meaning the only variable the processor has no control over - COOLING - is now favorable), would deactivate normal turbo and as a result should take care of both the turbo job and overclocking job. It'll be an all-in-one solution, clocking a chip to turbo speeds and beyond as long as the right conditions are met.

In the past, we had stock clocks and overclocking. Then Intel introduced Turbo. Now, AMD is introducing what is essentially Super Turbo - extended, limited, opportunistic, automated-overclocking.

 

[lobz]

I believe they have to pay GF the margin GF would have otherwise gotten per wafer. Given TMSC has gross margins around 50% lets say AMD gets a GF discount so GF margin is 40% ( that seems reasonable to me) then per chip not manufactured at GF AMD will have to pay GF somewhere between $8-$12 a chip (assuming between 5-7k a wafer and 240 usable chips a wafer).

while that may be, it'd still be much better then a choking market with ever rising prices

 

[unseenmorbidity]

Admittingly the XFR slide was extremely confusing, but still the slide doesn't say anything about XFR being an "automated overclocking" feature.
XFR stands for Extended Frequency Range and that's exactly what it is, literally.

Lisa does,

skip to 30 minute mark

 

[lolfail9001]

skip to 30 minute mark

Wording is ambiguous. In fact "enable higher clock speeds" is actually... reaffirming my suspicion.

Damn, i need to have a meal and relax, that starts to worry me.

 

[bjt2]

Wording is ambiguous. In fact "enable higher clock speeds" is actually... reaffirming my suspicion.

Damn, i need to have a meal and relax, that starts to worry me.

Have you read the link i provided?

 

[Dresdenboy]

The problem with the "overclocked" hypothesis is that the silicon needs to be validated to run at the frequencies that XFR will enable. If you validate your silicon for max turbo of 4GHz, then running at 4.1GHz is risky because the silicon might not be able to handle it.

I don't think XFR will be "overclocking beyond the boost limits."

Reliability tracking as introduced with BR might come into play here.

What's new with modern AVFS is knowing in advance when the silicon would fail. With that it's possible to lift voltage and frequency in given, validated limits (esp. V) without risking anything.

Why do we ride XFR that much if there is classic overclocking with a Wattman like tool?

 

[lolfail9001]

Have you read the link i provided?

Yep, but since i do not have access to IEEE article (did not bother with it yet), i have to ask. Is the part where conclusion made on turning these power savings into increased frequency... speculation or mentioned in the article itself? Because if it is the former then entire article describes power saving tech in Carrizo/Bristol/Mullins and then proceeds to make a jump inverting their purpose to increase frequency. Make no mistake, it is certainly possible to do. But the wording of the slide and Lisa's words definitely do not state outright that clock *increases*. Just that it can be increased.

Why do we ride XFR that much if there is classic overclocking with a Wattman like tool?

Because XFR carries a hefty premium tag with it. Got to figure out why.
Inb4 AMD next leveled everyone and put a premium tag on barely useful feature.

 

[Atari2600]

The problem with the "overclocked" hypothesis is that the silicon needs to be validated to run at the frequencies that XFR will enable. If you validate your silicon for max turbo of 4GHz, then running at 4.1GHz is risky because the silicon might not be able to handle it.

I don't think XFR will be "overclocking beyond the boost limits."

That is sound reasoning for XFR only allowing power exceedence if cooling can move the excess heat away but still keeping the upper clock limit.

**IF** AMD don't include the disclaimer "processor longevity using XFR setting 2 is not guaranteed".

setting 1 being operating under the turbo clock limit and setting 2 beyond it.


AMD seem to have the tools there, its just a question of how they use them.

 

[The Stilt]

Because XFR carries a hefty premium tag with it. Got to figure out why.

I wouldn't say that XFR carries any premium. It's the higher clocks that do, same as with Intel.

 

[Jan Olšan]

So basically XFR is just name for the underlying stuff that allowed the chip to have its advertised clocks (in the case of 1800X, 3.6 base 4.0 boost?).

That would actually fall in line with the whole SenseMI group. All of them sound like 'let's give names to functionality that simply has to be in a CPU, so that they can talk about some buzzwords before the reveal' business. Something like "Intel Smart Cache" ( = L3).

 

[Dresdenboy]

Some ppl here seem to be on a mission.

 

[Crumpet]

Some ppl here seem to be on a mission.

I'm actually really excited about XFR and cannot wait to see what it does, how it does it and how well it does that.