Max power scaling like that is impossible with a modern SoC,all equal.
If you have ever seen such a result, maybe the benchmark wasn't appropriate for the purpose, maybe the cooling wasn't scaled sufficiently for the 8 cores.
That s not as simplistic as one would think...
FI a FX4350 clocked at the same frequency/voltage than a FX8350 would use 55-60% of the latter s power, and the 4350 has a full 8MB cache, with half the cache to match the core count we would be closer to 50%.
An article of infinity fabric at SA by Charlie
http://semiaccurate.com/2017/01/19/amd-infinity-fabric-underpins-everything-will-make/
CPC did mention 65W for 6 cores but also for 4 cores.
"I'm not aware of any Ryzen ES w/ 45W TDP. All current 4C samples are rated at 65W. 45W TDP would require a max. frequency @ 2.8G (w/ 4C)
"https://twitter.com/CPCHardware/status/819435019588567040
That's a somewhat surprising statement given what's known of Ryzen's performance and that Intel's 7700T is 4c/8th and has a base clock of 2.9 GHz, turbo to 3.8 GHz, AND HD 630 integrated graphics all at a 35 W TDP. Could mean that Intel's process is more efficient lower on the power curve, but AMD's catches up (or surpasses) at the high end.
There is also the 7700 at 65W with a base of 3.6 and a 4.2 turbo.That's a somewhat surprising statement given what's known of Ryzen's performance and that Intel's 7700T is 4c/8th and has a base clock of 2.9 GHz, turbo to 3.8 GHz, AND HD 630 integrated graphics all at a 35 W TDP. Could mean that Intel's process is more efficient lower on the power curve, but AMD's catches up (or surpasses) at the high end.
You won't be the least surprised about it when you get to compare prices (4c Zen vs. 4c KBL). Meanwhile, if Zen top bins can stay within 95W at 3.6Ghz then there's a very good chance even good quality 6c/12t SKUs can do 2.8Ghz @ 45W. All you need is a bit more than 10% voltage drop from 3.6Ghz to 2.8GhzThat's a somewhat surprising statement given what's known of Ryzen's performance and that Intel's 7700T is 4c/8th and has a base clock of 2.9 GHz, turbo to 3.8 GHz, AND HD 630 integrated graphics all at a 35 W TDP. Could mean that Intel's process is more efficient lower on the power curve, but AMD's catches up (or surpasses) at the high end.
There is also the 7700 at 65W with a base of 3.6 and a 4.2 turbo.
Makes 45W at 2.8 max sound kinda' low?
Quick update on our T SKU discussion: got some time to test in Cinebench with a 35W package power hard limit. Max clocks were set to 4Ghz, voltage was stock.
For reference, the 6600T max 4 thread turbo is 3.3Ghz and the max 3 thread turbo is 3.4Ghz. Base clock is 2.7Ghz. The T CPU may well still use less power than the standard SKU in the loads above, but simply clocking higher than the T CPU at the same 35W TDP outlines the flexibility of using standard CPU in combination with TDP limits to obtain higher performance in a DIY SFF system.
- Running CB15 on 4 threads resulted in a score of 595, the i5 6600K averaged about 3.55Ghz while compelting the bechmark.
- Running a custom 3 threaded bench resulted in a score of 496 and an average clock of 3.85Ghz.
- Prime 95 Large FFT load resulted in sustained 2.9Ghz. (28.8 avg)
Where i said "i think"? What post are you referring?Bjt: Thanks, that's exactly what I meant. You dodged the questions again. Proves my point from the beginning.
All very optimistic 'I think' disguised as fact.
Sent from HTC 10
(Opinions are own)
We have factual confirmation of 5.0 on air for RyZen? Seems like that would be big news, even on a single core suicide run?Where i said "i think"? What post are you referring?
If you refer the posts on power consumption, I posted PDF of VRMs and web link of an efficiency diagram of a PSU. So my calculations are upper bound to power consumption. I don't care of 2008 CPUs. I care of comparable CPUs: SMT, similar turbo behavior, similar power saving features, etc... So max 2-3 years old CPUs...
If you refer the post on the max OC, i posted FACTS: +400MHz base clock and less power draw. Moreover there is that 5GHz air OC on one core. Is there on the web at least a suicide OC on air of a 6900K at 5Ghz? If not, this is another argument that make more probable my statement.When I say that Zen can overclock more than BWE i can't be sure, because i don't have an ES to test. But there are these FACTS that let me say that it is probable.
If you refer other posts, i don't remember, so please tell me.
We have factual confirmation of 5.0 on air for RyZen? Seems like that would be big news, even on a single core suicide run?
We hide regularly some easter eggs in the magazine. In March, we had coded "Intel GPU = AMD" in a binary string of the CPU page of the purchase guide. Virtually no one had seen the information at that time when it made a big noise 6 months later. In short, we are teasing. The presence of the chain of the current issue, which decodes "ZenOC @ Air = 5G" in this issue people talking on forums for 2 days. Being unmasked, we owe you some details. First, we did not summarize a test in a few bits. If we had been able to test ourselves overclocking, we would have told you openly in the preview. In spite of everything, we know with almost certainty that the CPU that we used for the tests actually came close to the 5 GHz with an (huge) air-dissipator. The I / O multiplier is not clamped at this time and is configured in steps of 0.25x. One heart, however, was active; The Motherboard VRMs seemed at that time too unstable to test with all of the cores. Other Ryzen ES are currently in the hands of overclockers and you should not delay to learn more: a demonstration of overclocking could occur at the CES if good results are achieved.
'I thinks' because,Where i said "i think"? What post are you referring?
If you refer the posts on power consumption, I posted PDF of VRMs and web link of an efficiency diagram of a PSU. So my calculations are upper bound to power consumption. I don't care of 2008 CPUs. I care of comparable CPUs: SMT, similar turbo behavior, similar power saving features, etc... So max 2-3 years old CPUs...
If you refer the post on the max OC, i posted FACTS: +400MHz base clock and less power draw. Moreover there is that 5GHz air OC on one core. Is there on the web at least a suicide OC on air of a 6900K at 5Ghz? If not, this is another argument that make more probable my statement.When I say that Zen can overclock more than BWE i can't be sure, because i don't have an ES to test. But there are these FACTS that let me say that it is probable.
If you refer other posts, i don't remember, so please tell me.
'I thinks' because,
Show us how your method of deducting DC CPU power from AC system power is,
Accurate
Valid
Repeatable
Show us how your method of deducting DC CPU power from AC system power is,
Accurate
Valid
Repeatable
Example - modern system with CPU power management enabled vs. disabled, power measured at wall.Its actually not awfully bad.
If you'd bother to look around, you'd see the Phenom 9850 example you used earlier actually can idle at around 55W without C&Q enabled...
55+60 = 115W, which is not a kick away from 126W measured under (different) load.
That's a somewhat surprising statement given what's known of Ryzen's performance and that Intel's 7700T is 4c/8th and has a base clock of 2.9 GHz, turbo to 3.8 GHz, AND HD 630 integrated graphics all at a 35 W TDP. Could mean that Intel's process is more efficient lower on the power curve, but AMD's catches up (or surpasses) at the high end.
'I thinks' because,
Show us how your method of deducting DC CPU power from AC system power is,
Accurate
Valid
Repeatable
Stick to the subject discussed.
Sent from HTC 10
(Opinions are own)
^ This is the problem right now. There isn't enough data to show any repeatability right now. We need hundreds of Zen CPUs tests to be able to say what the actual voltage, frequency and power parameters of Zen really are. We see so much variation, in most cases, between reviews samples and setups that even reviews are just a 'taste test' of a CPU's potential. AMD can certainly provide this data and eventually will, but verification by actual users gives us the low down on variability.
Example - modern system with CPU power management enabled vs. disabled, power measured at wall.
Enabled - 17W idle, ~90W load in Prime 95
Disabled - 29W idle, ~90W load under Prime 95
Now here's the problem - unless you already know the power management settings in this particular system, the room for error when guessing the needed idle power value is significant.
Let's assume CPU idle power usage is 2W. (deep sleep enabled)
Now let's assume CPU idle is 10W (no sleep, nominal frequency even when idle). Doing the math again leaves us with 63W and 54W.
- In the first case power delta is 73W. Compensating for losses 73x0.9x0.8=>53W. Add 2W and we get 55W
- In the second case power delta is 61. Compensating brigs us to 61x0.9x0.8=>44W. Add 2W and we get 46W
The correct value, when we guess right, is ~55W. But what happens if we guess wrong? The difference between wrong estimates can be 8-17W, or 15-30%. That's quite a bit of room for error and debates.
So using only this methodology and selectively choosing idle power usage that fits one's agenda, two opposing fans could argue that the CPU inside the system described above is either a 45W TDP CPU (optimistic) or a 65W TDP (pesimistic). Neither would be right, but the thread would be full of noise and accusations, even if the two fans used the exact same methodology.
And why is that? Because both ignored missing idle power data to paint desired picture.
In a loose sense, any measure is not repeatable, because there is noise, measure error and other confounding factors that can't be zeroed.
Should I include the error margin on the measure? It's implicit in my statements: i am not saying that Zen will draw 81.2353634W. I am saying that it draw ABOUT 80W.
So these objections are nonsense...
the room for error when guessing the needed idle power value is significant.
Example - modern system with CPU power management enabled vs. disabled, power measured at wall.
Enabled - 17W idle, ~90W load in Prime 95
Disabled - 29W idle, ~90W load under Prime 95
Now here's the problem - unless you already know the power management settings in this particular system, the room for error when guessing the needed idle power value is significant.
Let's assume CPU idle power usage is 2W. (deep sleep enabled)
Now let's assume CPU idle is 10W (no sleep, nominal frequency even when idle). Doing the math again leaves us with 63W and 54W.
- In the first case power delta is 73W. Compensating for losses 73x0.9x0.8=>53W. Add 2W and we get 55W
- In the second case power delta is 61. Compensating brigs us to 61x0.9x0.8=>44W. Add 2W and we get 46W
The correct value, when we guess right, is ~55W. But what happens if we guess wrong? The difference between wrong estimates can be 8-17W, or 15-30%. That's quite a bit of room for error and debates.
So using only this methodology and selectively choosing idle power usage that fits one's agenda, two opposing fans could argue that the CPU inside the system described above is either a 45W TDP CPU (optimistic) or a 65W TDP (pesimistic). Neither would be right, but the thread would be full of noise and accusations, even if the two fans used the exact same methodology.
And why is that? Because both ignored missing idle power data to paint desired picture.
Am i the only one who isn't bothered by TDP at all? I would buy a 2000W chip and happily drill 100mm holes in the wall for the cooling solution if the delivered performance is significantly better.