XFR is speed above base and boost isn't it.
Because of their extremely low Tcase demand. The other 95W parts got a Tcase above 70C.
Robert Hallock said something regarding required temperatures for XFR to lift off. The bigger cooler and lower Tcase is not there to hide a TDP but to make room for XFR and avoid a marketing disaster if users put some 40mm fans on their Ryzen chip and claim an XFR dysfunction on Reddit, getting further flak in this very forum and a Hibben article.
How do you know that they truly require a Wraith?
Have you seen data showing that the regular 95W cooler won't dissipate all the heat? Or are you just speculating?
It will require all of that 120mm AIO when I get my dirty hands on it
Unlocked Skylakes are the same - 91W TDP, but 130W level cooling required according to Intel (or 125W, not sure now). So this is nothing new.So it's "95 W" but it needs a 125 W rated cooler. Which honestly isn't that big of a deal although I guess you have to wonder why they don't just make it 125 W.
I think XFR should kick in right away no matter what cooler you have, its just that it would throttle down pretty quickly as well. Sound about right? It takes a few seconds for a cooler to get overwhelmed doesn't it? I mean it doesn't reach equilibrium instantly. I'll have mine under water anyway.
Also, I'm curious, does anyone have any idea if XFR boosts all cores, or just like 1 or 2? I have a feeling most OCers won't be using that. My only hope is that there is some sort of speedstep implementation. I'd like to have an OC on all cores but still have them throttle down when the CPU isn't being used, much like Intel CPU's do now. As long as CPU's last these days, I'm pretty sure speedstep has saved thousands upon thousands of CPU's from dying an early death.
My E8400 was at 4.0 all the time (no speedstep) and it degraded after about 2 or 3 years, requiring more and more voltage to stay stable. My 3930K is about 300 years old now and still stable as can be. I'd hope the same to be true of Ryzen chips.
It will require all of that 120mm AIO when I get my dirty hands on it
I think XFR should kick in right away no matter what cooler you have, its just that it would throttle down pretty quickly as well. Sound about right? It takes a few seconds for a cooler to get overwhelmed doesn't it? I mean it doesn't reach equilibrium instantly. I'll have mine under water anyway.
Also, I'm curious, does anyone have any idea if XFR boosts all cores, or just like 1 or 2? I have a feeling most OCers won't be using that. My only hope is that there is some sort of speedstep implementation. I'd like to have an OC on all cores but still have them throttle down when the CPU isn't being used, much like Intel CPU's do now. As long as CPU's last these days, I'm pretty sure speedstep has saved thousands upon thousands of CPU's from dying an early death.
My E8400 was at 4.0 all the time (no speedstep) and it degraded after about 2 or 3 years, requiring more and more voltage to stay stable. My 3930K is about 300 years old now and still stable as can be. I'd hope the same to be true of Ryzen chips.
Actually if I understood their explanation right... this might be the holy grail. If my understanding is correct, their sensors duplicate some critical paths in the chip in order to "test" where they fail before the actual logic does so. They of course give it a small margin because there's uncertainty involved, but it could come very close to manual overclocking.My understanding of XFR is that it is based on power consumption, temperature, and actually has some type of clock cut-off based on the temperature.
Being so tightly intertwined, though, you could simplify the operation by temperature alone.
>=80C - XFR OFF
>=70C - XFR DOWN
<=65C - XFR UP / ON (if off)
You would jump in 25MHz increments above the turbo clock every couple of milliseconds until you hit the temperature limit then you would back down again a little more slowly, but maybe faster depending on the distance above XFR_T_THROTTLE (to avoid XFR_OFF).. all of the way up to whatever limit is set for XFR_FREQ_MAX (there MUST be a limit... unless the CPU can detect the conditions which might lead to instability and can actually take you right up to the brink of instability... which would be the holy grail... but would completely ruin overclocking for everyone).
....
As for overclocking and still throttling down... I never do it any other way.
Because of their extremely low Tcase demand. The other 95W parts got a Tcase above 70C.
Actually if I understood their explanation right... this might be the holy grail. If my understanding is correct, their sensors duplicate some critical paths in the chip in order to "test" where they fail before the actual logic does so. They of course give it a small margin because there's uncertainty involved, but it could come very close to manual overclocking.
This might be a situation where manual OC doesn't net you more than 50MHz or so.
I am guessing as the chip exhausts its thermal headroom it will automatically throttle regardless what you believe your OC frequency is. The result would be fewer crashes and lower performance than what you expect. Like, you thought you set the clock at 4.5 GHz and the CPU-Z shows 4.5 GHz before the bench began, but when you come back to check the result you have a score of 3.5 GHz CPU. Basically this is what AMD's GPUs do now, iirc. I am personally not a fan - I prefer an old school crash/BSOD - but I guess this is a progress..I think XFR should kick in right away no matter what cooler you have, its just that it would throttle down pretty quickly as well. Sound about right? It takes a few seconds for a cooler to get overwhelmed doesn't it? I mean it doesn't reach equilibrium instantly. I'll have mine under water anyway.
Also, I'm curious, does anyone have any idea if XFR boosts all cores, or just like 1 or 2? I have a feeling most OCers won't be using that. My only hope is that there is some sort of speedstep implementation. I'd like to have an OC on all cores but still have them throttle down when the CPU isn't being used, much like Intel CPU's do now. As long as CPU's last these days, I'm pretty sure speedstep has saved thousands upon thousands of CPU's from dying an early death.
Of course, I'm skeptical myself. I don't actually believe it yet. Glad you're testing.Yeah... I'm just trying not to hype it up too much ;-) I have a section on my Ryzen review site specifically for testing XFR, so I am ready to find out just how much better I can - or can't - do with manual overclocking.
unless the CPU can detect the conditions which might lead to instability and can actually take you right up to the brink of instability... which would be the holy grail... but would completely ruin overclocking for everyone).
I believe the TCase max is not what the chip's actual maximum operating temperature is, but rather what the cooler must keep the CPU at assuming X frequency. The non X coolers are more lax, but the top end CPU's need more headroom in order for XFR to do its work.
Zen might not even have set p-states. At least I don't the point in having it if the system can just know the optimal way to run. No point in having a pre-set table.It does work this way... Since Carrizo. But with CZ and BR the clock was limited to the higher boosted p-state.
I analyzed a paywalled paper on ieeexplorer somewhere here (i don't remember the thread) some day ago.
In short AVFS have small replicas (500 on CZ and BR and 1300 on Zen) of critical circuits, spread all over the die and periodically the microcode or the SMU trigger a bunch of circuits (10 on CZ and BR) that perform statistics on these small replica, to infer the correct Vcore and frequency table for maximum performance with stability.
On BR and CZ this table is limited to the supported p-state (and frequencies). On Zen i suppose that this table will be extended to higher frequencies...
I didn't claim otherwise?Tcase is the top of the IHS and what the cooler is required to keep this temperature under. Its not a core temperature.
I didn't claim otherwise?
Its not what you wrote either. The CPU can still be ~100C while the Tcase is below 60C.
Yes, but that would require AMD assembling Ryzen IHS at intel fabs.
60Tcase is the same since phenom, isn't it?
Zen might not even have set p-states. At least I don't the point in having it if the system can just know the optimal way to run. No point in having a pre-set table.
I didn't claim otherwise?
it's like comparing a state of the art heat sensor inside of a spaceship to a thermostat in your houseGuys isnt XFR similar like intels Intel® Turbo Boost Max Technology 3.0 ?
XFR is speed above base and boost isn't it.
Not really.
http://support.amd.com/TechDocs/43374.pdf
http://support.amd.com/TechDocs/43375.pdf
See page 18 and forward.
SKUs and their associated coolers and cooling specs. 1800X/1700X/1600X doesn't just require a 95W cooler, but a special 95W cooler.