4,2XFR on threadripper. Higher binned/new stepping chips) ?

[Adaa]

I dont think any media reviews has talked about this yet, and all slides i saw so far, only mentioned 4,0boost speed, not that they had a 4,2ghz XFR boost as well. 

 

So in this timestamp, They confirm 4,2ghz XFR clocks on Threadripper, this must mean that we are talking about golden binned chips, OR simply a new stepping with higher clock stabillity and topspeeds on the chips ?.

Capsaicin Siggraph 2017, timestamp to their 4,2ghz XFR confirm.

[DXMember]

what the hell? did they livestream? or they  just filmed for youtube video?

Don't know what media you were watching...

Tom's Hardware published stuff including XFR

http://www.tomshardware.com/news/amd-threadripper-preorder-processor-ryzen,35111.html

 

And I'm also very excited for 4.2, looking forward to all core 4.2 overclock

[Adaa]

1 minute ago, DXMember said:

what the hell? did they livestream? or they  just filmed for youtube video?

Not sure, just found it on youtube, donno who filmed it. ;-) 

 

Well it does come from AMD´s own channel, but perhabs they just uploaded it afterwards, where they could edit in it if needed. 

I doubt it was a (live) stream, prob just recorded to upload after edit and such. 

[EPPHO]

They had a nice amount of time from Ryzen release so this was expected to happen.

[vanished]

well, that's still not much of an improvement but I guess anything helps.  I still don't understand the point of XFR though, so, it has a base clock, and a boost clock (which is already bad enough imo) but then on top of that they add another kind of boost?  

[2FA]

Just now, Ryan_Vickers said:

well, that's still not much of an improvement but I guess anything helps.  I still don't understand the point of XFR though, so, it has a base clock, and a boost clock (which is already bad enough imo) but then on top of that they add another kind of boost?  

It's like GPU Boost, there is a stock clock, a guaranteed minimum boost clock, and then additional boost clock if thermals allow.

[vanished]

Just now, DeadEyePsycho said:

It's like GPU Boost, there is a stock clock, a guaranteed minimum boost clock, and then additional boost clock if thermals allow.

so it really is just pointless marketing then.  I have yet to see a valid reason for breaking the frequency range into more than two pieces - the "normal" or guaranteed range, and then the "you might get this if things are good" range.  Based on your description, AMD should really just do away with the "boost clock" and report the boost clock frequency as the base clock.  I mean, what actually is the difference between a "guaranteed minimum boost" and just a base clock?  As far as I can tell they differ only in name.

[Adaa]

7 minutes ago, Ryan_Vickers said:

well, that's still not much of an improvement but I guess anything helps.  I still don't understand the point of XFR though, so, it has a base clock, and a boost clock (which is already bad enough imo) but then on top of that they add another kind of boost?  

Its to get the most out of the cpu´s, and to prevent noobs(read non-techies) who is using stock coolers, or horrible 10-20USD coolers , from having unstable systems that hang or take damage due to overheating from a to high turbo preset. 

 

what i personally dont understand, is why Ryzen dosnt do dynamical overclocking/boosting on more than 1 core, if it could do 1-4 or 1-6 core dynamic boosts, it would be WAY more competetive with Intels X299 platform.  Since 2 or 4 cores at a turbo 2.0/3.0 speed could be more competitive in games for people who dosnt overclock.

[2FA]

Just now, Ryan_Vickers said:

so it really is just pointless marketing then.  I have yet to see a valid reason for breaking the frequency range into more than two pieces - the "normal" or guaranteed range, and then the "you might get this if things are good" range.  Based on your description, AMD should really just do away with the "boost clock" and report the boost clock frequency as the base clock.  I mean, what actually is the difference between a "guaranteed minimum boost" and just a base clock?  As far as I can tell they differ only in name.

All cores will run at stock, not all will run at boost. Just like Intel's turbo boost if I'm not mistaken.

[ravenshrike]

1 minute ago, Ryan_Vickers said:

well, that's still not much of an improvement but I guess anything helps.  I still don't understand the point of XFR though, so, it has a base clock, and a boost clock (which is already bad enough imo) but then on top of that they add another kind of boost?  

Boost is single core only guaranteed on their fan. XFR will go beyond that to the maximum of the chip/cooling capability. Overclocking properly can OC all cores to max XFR freq or a little beyond and if you know how to play with P-states I think you can force it to boost all cores to max OC freq. Not sure on that though.

[vanished]

1 minute ago, DeadEyePsycho said:

All cores will run at stock, not all will run at boost. Just like Intel's turbo boost if I'm not mistaken.

 

Just now, ravenshrike said:

Boost is single core only guaranteed on their fan. XFR will go beyond that to the maximum of the chip/cooling capability. Overclocking properly can OC all cores to max XFR freq or a little beyond and if you know how to play with P-states I think you can force it to boost all cores to max OC freq. Not sure on that though.

 

Ah.  Well it's still an unnecessary distinction, just, the line that doesn't need to exist is the one between boost and XFR.  People shave gotten so used to always running at max turbo (on Intel) that I think they've forgotten that the idea of it was to boost as far as it can, provided enough thermal, current, voltage, etc. headroom.  Sounds like that's what XFR is too, why does it need to be differentiated from "boost clock"?

[2FA]

Just now, Ryan_Vickers said:

 

 

Ah.  Well it's still an unnecessary distinction, just, the line that doesn't need to exist is the one between boost and XFR.  People shave gotten so used to always running at max turbo (on Intel) that I think they've forgotten that the idea of it was to boost as far as it can, provided enough thermal, current, voltage, etc. headroom.  Sounds like that's what XFR is too, why does it need to be differentiated from "boost clock"?

I think it's to save face more than anything as they aren't going to guarantee the clock speed you get with XFR.

[Adaa]

1 minute ago, ravenshrike said:

Boost is single core only guaranteed on their fan. XFR will go beyond that to the maximum of the chip/cooling capability. Overclocking properly can OC all cores to max XFR freq or a little beyond and if you know how to play with P-states I think you can force it to boost all cores to max OC freq. Not sure on that though.

Cept P-states oc´ing is broken on most if not all bios és so far, i only heard about succes doing P-state oc using some ROG board software untility.

sadly. The boards makers are still sleeping and figthing with mem compabillity and, most of them(see actually hardcore overclocking´s channel), have wrong timing auto setup, in the bios´es for a ton of ram sticks, that either prevent best possible performance, or prevent high clockspeed mem oc. 

[Doobeedoo]

Yeah, read about it already. It's great extra increase. Nice to see it on Threadripper that good clocks are to stay with higher core count. 

[PianoPlayer88Key]

Having higher boost clocks is nice, sure.  But, I want to see better IPC in the next generation, like the better of the 2 following examples...

 

In single-threaded mode at 4 GHz, get like 200-250 in CineBench R15.

Or, pass whatever is the then-current Intel architecture in IPC, by how far the FX series had been behind Intel's latest before Ryzen was released.  For example, if Piledriver did 50%  of Intel Kaby Lake (LGA1151), then Ryzen 2 / Threadripper 2 should do 200% of Coffee Lake / Ice Lake / whatever is out at the time.  If it was 25% behind, then it should be 33% ahead.  Basically, leapfrog by the reciprocal or something like that.

[The Benjamins]

10 minutes ago, PianoPlayer88Key said:

Having higher boost clocks is nice, sure.  But, I want to see better IPC in the next generation, like the better of the 2 following examples...

 

In single-threaded mode at 4 GHz, get like 200-250 in CineBench R15.

Or, pass whatever is the then-current Intel architecture in IPC, by how far the FX series had been behind Intel's latest before Ryzen was released.  For example, if Piledriver did 50%  of Intel Kaby Lake (LGA1151), then Ryzen 2 / Threadripper 2 should do 200% of Coffee Lake / Ice Lake / whatever is out at the time.  If it was 25% behind, then it should be 33% ahead.  Basically, leapfrog by the reciprocal or something like that.

That is some wiseful thinking. it took them 4+ years to make that jump.

 

I would expect 5%-10% clock bump and 5%-10% IPC bump, not another 40+% bump.

 

Note: I also find what I expect to be optimistic. it also seems likely that Zen2 will have 6c per ccx which would be a 50% core/multi thread bump.

[Misanthrope]

Could it be the physically larger chips? XFR is just one core and the cooling solutions should be far more robust meaning 4.2 might be easier to achieve on a single core given the space.

 

But maybe they already improved manufacturing a bit though.

[The Benjamins]

1 minute ago, Misanthrope said:

Could it be the physically larger chips? XFR is just one core and the cooling solutions should be far more robust meaning 4.2 might be easier to achieve on a single core given the space.

 

But maybe they already improved manufacturing a bit though.

no XFR is a all aspects boost. it means it can boost base clock on all cores by 200Mhz and the single/2 core boost by that amount too. but it is not guaranteed.

[Adaa]

Just now, Misanthrope said:

Could it be the physically larger chips? XFR is just one core and the cooling solutions should be far more robust meaning 4.2 might be easier to achieve on a single core given the space.

 

But maybe they already improved manufacturing a bit though.

Or simply golden binned chips. 

Unlike most recommending 1700 (non x)advices, due to techchannels calling them 100% identical, I still belive theres a big binning proccess, it migth be same chip with all the ryzens, but its differently not same 4+Ghz % betwean them, hell my 1700, struggles to get stable above 3,8 at any voltages below 1,4 and 3,9 seems impossible on air and at safe voltage below 1,425v. 

[Misanthrope]

6 minutes ago, Adaa said:

Or simply golden binned chips. 

Unlike most recommending 1700 (non x)advices, due to techchannels calling them 100% identical, I still belive theres a big binning proccess, it migth be same chip with all the ryzens, but its differently not same 4+Ghz % betwean them, hell my 1700, struggles to get stable above 3,8 at any voltages below 1,4 and 3,9 seems impossible on air and at safe voltage below 1,425v. 

I was reading elsewhere that this are actually not rejected Epyc bins (Or viceversa) AMD says 2 out of the 4 units are just there for architectural reasons (i.e. probably less heat to deal with) so I think there might be more to it than just really good binning

 

https://www.overclock3d.net/news/cpu_mainboard/amd_clarifies_why_threadripper_uses_4_silicon_dies/1

[The Benjamins]

6 minutes ago, Adaa said:

Or simply golden binned chips. 

Unlike most recommending 1700 (non x)advices, due to techchannels calling them 100% identical, I still belive theres a big binning proccess, it migth be same chip with all the ryzens, but its differently not same 4+Ghz % betwean them, hell my 1700, struggles to get stable above 3,8 at any voltages below 1,4 and 3,9 seems impossible on air and at safe voltage below 1,425v. 

silicon lottery article  had the 1800x as twice as likely to get to 4Ghz then the 1700.

 

and the TR chips seem to be binned even better.

[Misanthrope]

1 minute ago, The Benjamins said:

silicon lottery article  had the 1800x as twice as likely to get to 4Ghz then the 1700.

 

and the TR chips seem to be binned even better.

But there's just 3 SKUs, wouldn't that mean a lot of wasted ones? Or lots of the 8 core ones that would probably never sell because really if you're spending on X399 why not get at least the 12 core you already paid quite a bit.

[Adaa]

3 minutes ago, Misanthrope said:

But there's just 3 SKUs, wouldn't that mean a lot of wasted ones? Or lots of the 8 core ones that would probably never sell because really if you're spending on X399 why not get at least the 12 core you already paid quite a bit.

Anything that dosnt quallify for 8 stable cores at 3,25ghz full core boost + XFR (R7 1700) with 1 core stable at 3,7, is highest likely being devoted to R5/R3 series chips, since they can find the weak cores, and simply disable those that lack stabillity over a certain clockspeed.

 

So if 1 core isnt stable over 3ghz, they will just lock the core + the worst 2nd one, and then make it a R5 1600/1600x, depending on stable speed on the rest of the cores. 

[The Benjamins]

1 minute ago, Misanthrope said:

But there's just 3 SKUs, wouldn't that mean a lot of wasted ones? Or lots of the 8 core ones that would probably never sell because really if you're spending on X399 why not get at least the 12 core you already paid quite a bit.

I would expect they build towards the demand. the 8 core parts is 2 dies with 2 cores per ccx they are not flawless. they are defect dies but can still hold high clocks when the R5 are defective dies that can't.

[Misanthrope]

Just now, Adaa said:

Anything that dosnt quallify for 8 stable cores at 3,2ghz (R7 1700x) with 1 core stable at 3,7, is highest likely being devoted to R5/R3 series chips, since they can find the weak cores, and simply disable those that lack stabillity over a certain clockspeed.

 

So if 1 core isnt stable over 3ghz, they will just lock the core + the worst 2nd one, and then make it a R5 1600/1600x, depending on stable speed on the rest of the cores. 

Can they rehuse the same die on a different socket chip though? I thought that wasn't really a thing but well that'd make sense.