AMD Ryzen (Summit Ridge) Benchmarks Thread (use new thread)

[bjt2]

We have no uarch + process + manufacturing data to model this.

Random guessing aside.

Sent from HTC 10
(Opinions are own)

No. A 2.41GHz neon FPU draw 330mW. A Zen FPU is 4x more powerful. So it should draw 1.33W. A Zen CPU should draw less than 4x its FPU, since the FPU is the most power hungry unit in a CPU. So a Zen CPU at 2.4GHz should draw less than 5W.

This is also confirmed by the excavator low TDP APU that in 15W, on the 28nm BULK process, have a 4 core APU, e.g. the A10 9600P that in 15W gives 4 XV cores at 2.4+GHz and 384 SP@720+MHz... Even if the 15W are all drawn by the 4XV cores at 2.4GHz, they are 7.5W for an XV module (that is comparable to a Zen core) on the shitty 28nm BULK... On the 14nmFF this can safely draw way less 5W... If we consider that the 15W include also a 384SP GPU and the SB and the NB...

EDIT: there is also the FX 9800P that has 4 XV core at 2.7+GHz and 512SP at 758+MHz in 15W... So also 2.7GHz at 5W is feasible...

 

[coercitiv]

This is also confirmed by the excavator low TDP APU that in 15W, on the 28nm BULK process, have a 4 core APU, e.g. the A10 9600P that in 15W gives 4 XV cores at 2.4+GHz and 384 SP@720+MHz... Even if the 15W are all drawn by the 4XV cores at 2.4GHz, they are 7.5W for an XV module (that is comparable to a Zen core) on the shitty 28nm BULK...

Here comes reality check, duck and cover:

Now let's make room for those 384 SP @ 720+ Mhz, shall we?

For example, in “Diablo III”, the CPU and GPU clock speeds start at 1800 and 550 MHz respectively. As the game runs, the speeds drop to 1100 to 1200 (CPU) and 380 to 420 MHz (GPU). The frame rates drop in parallel in our static gaming scene from 72 to 49 fps.

 

[bjt2]

Here comes reality check, duck and cover:

Now let's make room for those 384 SP @ 720+ Mhz, shall we?

It is not clear from the text if the power consumption is for the 35W (A12) or 15W (A10) APU... Moreover it's a notebook review... How they calculate the APU power draw? Are you sure it's not the total power draw? Reading the last paragraph, the 37.5-30W are the whole system power consumption, including monitor, hdd, other stuff in the notebook and power adapter and VRM losses... Nice try...
And it's a translation from a german article...
Since the clocks and the power consumption go down with the time, I suspect there is also some dissipation problem as in first the ASUS Yoga, kicking the overheating protection...

 

[coercitiv]

It is not clear from the text if the power consumption is for the 35W (A12) or 15W (A10) APU... Moreover it's a notebook review... How they calculate the APU power draw? Are you sure it's not the total power draw? Reading the last paragraph, the 37.5-30W are the whole system power consumption, including monitor, hdd, other stuff in the notebook and power adapter and VRM losses... Nice try...
And it's a translation from a german article...

For heaven's sake, look at the clocks, not the system power consumption. It's a 15W TDP CPU and cannot sustain 2Ghz under heavy load, in complete opposition to you claim of 2.4Ghz+ and power to spare for iGPU as well.

 

[bjt2]

For heaven's sake, look at the clocks, not the system power consumption. It's a 15W TDP CPU and cannot sustain 2Ghz under heavy load, in complete opposition to you claim of 2.4Ghz+ and power to spare for iGPU as well.

Have you seen the rewiews of the first ASUS Yoga with the ULV Y-series INTEL CPUs? Also these went in thermal protection and went underclocked.
These AMD APU have a configurable TDP of 10-15W, at OEM discretion. What if HP decided to cap the APU to 10W to use a smaller heat-sink?

Anyway these are on the 28nm BULK HPP process... A 14nm FF low power process is another story... Curiosly at high clock the 14nm FF process is a low power process and can't go too high... On lower clock many of you forget that is a low power process...

And the higher end processor is the FX9800P, that has higher clock and more SPs also in 15W... The other is a lower budget harvested part...

 

[itsmydamnation]

For heaven's sake, look at the clocks, not the system power consumption. It's a 15W TDP CPU and cannot sustain 2Ghz under heavy load, in complete opposition to you claim of 2.4Ghz+ and power to spare for iGPU as well.

And what workload runs the CPU that hard while also running the GPU as hard as possible as well?

You wouldn't what to see my haswell 4300u fall off the cliff with even simple games like diablo 3 at lowest rez+settings.

 

[KTE]

^Like I've discussed with you before, you're jumping many hurdles to arrive at a desired goal there. Mass manufactured semicon reality vs paper extrapolations don't work.



Sent from HTC 10
(Opinions are own)

 

[bjt2]

^Like I've discussed with you before, you're jumping many hurdles to arrive at a desired goal there. Mass manufactured semicon reality vs paper extrapolations don't work.



Sent from HTC 10
(Opinions are own)

30W for a full system, subtracting 90% efficiency for the PSU (and i was generous), 90% for the VRMs (and again i was generous), they are 24W useful power for all the notebook. Subtract 3W for the SSD, 7W for the LCD panel, a few watts for the SODIMMs, and few watts for ancillary chips (e.g. WiFi), fans and leds, and we are left with a little more than 10W (if not less) for the whole APU... 4 excavator cores, 384 SPs, north bridge and southbridge... And this is also not the top model, but a "waste"... The top model is the FX9800P with 2.7GHz base clock and 512SPs in 15W...

 

[Sweepr]

A10-9600P can't sustain its Turbo clockspeeds at 15W TDP, let alone the 'top model':

The A10-9600P does not fulfil its potential. AMD specifies a clock speed range of 2.4 to 3.3 GHz. However, the test APU cannot perform at the upper ends of this range continuously due to the limiting TDP value. This behavior exhibited in the Carrizo models as well. The APU often starts the benchmarks at a high clock speed. Sadly, these high clock speeds are reached by surpassing the 15-watt TDP limit, which means that after a short period of time (around a minute), the performance drops by up to 20%.

In CPU only MT loads it's barely above 2 GHz on all cores after 5 minutes. When both CPU + iGPU are taxed it's a disaster (1.1-1.2 GHz CPU + 380-420 MHz iGPU). And yes, they did test a 15W model.

In games, the TDP limit is temporarily exceeded. However, after a maximum of two minutes (depending on the previous load and temperatures of the laptop), the 15-watt limit strikes back, dropping clock speeds. For example, in “Diablo III”, the CPU and GPU clock speeds start at 1800 and 550 MHz respectively. As the game runs, the speeds drop to 1100 to 1200 (CPU) and 380 to 420 MHz (GPU)

At full load, the 15-inch model consumes a max 37.6 watts, which drops to around 30 watts (typical ULV levels) when the TDP limit kicks in.

http://www.notebookcheck.net/Bristo...0-9600P-Against-the-Competition.168477.0.html

 

[coercitiv]

And what workload runs the CPU that hard while also running the GPU as hard as possible as well?

You wouldn't what to see my haswell 4300u fall off the cliff with even simple games like diablo 3 at lowest rez+settings.

Let's be clear on one thing - since I'm feeling the anti-AMD label sticking on my forum persona (which is quite ironic to say the least), I'm not criticizing the APU, in my view it's normal for combined CPU and iGPU loads to go well over a low 15W budget, simply because it's optimal that way - this ensures all kinds of loads are approached with the maximum performance the APU can offer, be it CPU centric, iGPU centric, or combined load. (both Intel and AMD should and will do this) I said it before and I'll say it again: I kinda like the ideas behind construction cores and I would have loved to see construction based APUs being ported on a modern process (with work on power management IP being continued), if only to see how it really stacked against Haswell and nephews. But that's another subject altogether.

However, what is being discussed here is not that - the XV APU worthiness, but a claim made earlier in this thread that 1 XV module consumes no more than 7.5W to deliver 2.4Ghz+ with power room to spare for the iGPU. This is what I responded to, an outlandish claim with no real world data to back it up, and all the real world data to go against it. And the minute I posted hard facts I was greeted with lots of spins and a "nice try" comment. Why try so hard to invalidate the data from the review I posted when any of the thread participants could bring more data to light, either from alternative sources or even personal systems and show everyone plus dog the truth? That's how it works in productive circles.

You have the 4300U? Great! Let's indulge in a slightly OT comparison. Get Prime 95, run a blend test (which is not the most intensive of them), and see how your CPU will hang around 1.9Ghz-2Ghz while being TDP throttled. My 4510U stays around 2Ghz in the same test, so base clocks are more or less equal to Prime 95 clocks as long as laptop cooling is adequate and CPU is allowed 15W TDP.

I hope this means we can agree that one 22nm Haswell core at 2Ghz uses more than 5W in Prime. For the sake of comparison let's say the 2 cores themselves use 2x5W and the rest of the package uses 5W as well - not accurate but indulge me for now. I don't have real world data on Skylake, but since the i7 has 2.5Ghz base, we can safely assume a 14nm Skylake core at 2.5Ghz uses 5W in Prime. (I could really use more accurate data on Skylake in Prime 95 if anyone can run a 2 min test would be great, since clocks may be a bit higher).

That having been said, let's go back to Zen 2.4Ghz at 5W. Presumably Zen has Haswell like IPC, hence 14nm Zen would be 20% more efficient than 22nm Haswell and even come close to 14nm Skylake (2.4Ghz vs. 2.5Ghz with some IPC handicap). Any watts we take back from the rest of the Intel package and put into core power usage will make Skylake/Haswell look even worse (say 6W for Hasell core at 2Ghz, Zen becomes 40%+ more efficient). This means the claim of Zen 2.4Ghz at 5W puts the new AMD core very close to Intel's best in mobile. Whether that is true or not, it remains to be debated and proved, but I hope this little napkin math shows you we cannot take such a claim lightly, since it puts efficiency close to the best high performance core available on the market.

This 5W Zen @ 2.4Ghz claim must be met with "strict scrutiny" and extrapolating from dubious approximations of XV power usage in mobile is definitely not the way to start this conversation, at least not when real world data blatantly contradicts them.

30W for a full system, subtracting 90% efficiency for the PSU (and i was generous), 90% for the VRMs (and again i was generous), they are 24W useful power for all the notebook. Subtract 3W for the SSD, 7W for the LCD panel, a few watts for the SODIMMs, and few watts for ancillary chips (e.g. WiFi), fans and leds, and we are left with a little more than 10W (if not less) for the whole APU... 4 excavator cores, 384 SPs, north bridge and southbridge... And this is also not the top model, but a "waste"... The top model is the FX9800P with 2.7GHz base clock and 512SPs in 15W...

The idle power usage of the HP Pavilion being used in that review was between 6-9W, you are so off the mark in you approximations that your LCD panel power consumption alone would completely exhaust idle power budget and leave nothing alone for the rest of the system. Even when approximating SSD power usage you conveniently consider the unit is continuously writing data during benchmarks, otherwise you would use the average idle power which is only 100 times lower than what you posted, at 30mW.

A really efficient laptop design with a 15W TDP CPU and no active dGPU will use around 3-5W while idle and 27-28W while under load after CPU will have exhausted turbo time (so it will start at 30-35W and go down after a while). Even a less efficient model will stay at 5-8W while idle and hang around 30W under load (after turbo period).

There's nothing wrong or special about the HP Pavilion in the review, except you're trying to inflate some power numbers in order to deflate others.

 

[bjt2]

A10-9600P can't sustain its Turbo clockspeeds at 15W TDP, let alone the 'top model':



In CPU only MT loads it's barely above 2 GHz on all cores after 5 minutes. When both CPU + iGPU are taxed it's a disaster (1.1-1.2 GHz CPU + 380-420 MHz iGPU). And yes, they did test a 15W model.





http://www.notebookcheck.net/Bristo...0-9600P-Against-the-Competition.168477.0.html

I have shown that on that particular model the power was capped at at most 10W. The same happened on the first asus yoga with the intel Y series.

Are you familiar with the new concept of surface temperature? Latest (mostly) fanless convertible, that uses all the chassis to dissipate the power, like a tablet, with surface temperature limited to 40-45C to not burn the user that usually put the convertible on his/her legs... This explain also the higher power draw at the start of the benchmark: if the temperature is below 40C the CPU is uncapped... 7 more W that translate in 5 more W on the APU (after the losses)... Right 15W at the bench start, that went to 10W when the surface temperature was too hot... And probablily less when the temperature went higher...

 

[coercitiv]

I have shown that on that particular model the power was capped at at most 10W. The same happened on the first asus yoga with the intel Y series.

Are you familiar with the new concept of surface temperature? Latest (mostly) fanless convertible, that uses all the chassis to dissipate the power, like a tablet, with surface temperature limited to 40-45C to not burn the user that usually put the convertible on his/her legs...

You have shown no such thing, the HP Pavilion is in no situation similar to a convertible/tablet, it is a 15" notebook with a sizeable chassis and adequate cooling for a 15W TDP CPU and a small dGPU on the side. This is how the unit looks like on the inside, with both a decent sized fan and radiator.

You really need to stop inventing reasons for this review/unit to be flawed and come up with another source of data that directly shows A10-9600P or higher 15W TDP SKU running Prime 95 at base clocks or higher.

And btw, keep in mind the minimum cTDP for A10-9600P is 12W, not 10W.

 

[bjt2]

Let's be clear on one thing - since I'm feeling the anti-AMD label sticking on my forum persona (which is quite ironic to say the least), I'm not criticizing the APU, in my view it's normal for combined CPU and iGPU loads to go well over a low 15W budget, simply because it's optimal that way - this ensures all kinds of loads are approached with the maximum performance the APU can offer, be it CPU centric, iGPU centric, or combined load. (both Intel and AMD should and will do this) I said it before and I'll say it again: I kinda like the ideas behind construction cores and I would have loved to see construction based APUs being ported on a modern process (with work on power management IP being continued), if only to see how it really stacked against Haswell and nephews. But that's another subject altogether.

However, what is being discussed here is not that - the XV APU worthiness, but a claim made earlier in this thread that 1 XV module consumes no more than 7.5W to deliver 2.4Ghz+ with power room to spare for the iGPU. This is what I responded to, an outlandish claim with no real world data to back it up, and all the real world data to go against it. And the minute I posted hard facts I was greeted with lots of spins and a "nice try" comment. Why try so hard to invalidate the data from the review I posted when any of the thread participants could bring more data to light, either from alternative sources or even personal systems and show everyone plus dog the truth? That's how it works in productive circles.

You have the 4300U? Great! Let's indulge in a slightly OT comparison. Get Prime 95, run a blend test (which is not the most intensive of them), and see how your CPU will hang around 1.9Ghz-2Ghz while being TDP throttled. My 4510U stays around 2Ghz in the same test, so base clocks are more or less equal to Prime 95 clocks as long as laptop cooling is adequate and CPU is allowed 15W TDP.

I hope this means we can agree that one 22nm Haswell core at 2Ghz uses more than 5W in Prime. For the sake of comparison let's say the 2 cores themselves use 2x5W and the rest of the package uses 5W as well - not accurate but indulge me for now. I don't have real world data on Skylake, but since the i7 has 2.5Ghz base, we can safely assume a 14nm Skylake core at 2.5Ghz uses 5W in Prime. (I could really use more accurate data on Skylake in Prime 95 if anyone can run a 2 min test would be great, since clocks may be a bit higher).

That having been said, let's go back to Zen 2.4Ghz at 5W. Presumably Zen has Haswell like IPC, hence 14nm Zen would be 20% more efficient than 22nm Haswell and even come close to 14nm Skylake (2.4Ghz vs. 2.5Ghz with some IPC handicap). Any watts we take back from the rest of the Intel package and put into core power usage will make Skylake/Haswell look even worse (say 6W for Hasell core at 2Ghz, Zen becomes 40%+ more efficient). This means the claim of Zen 2.4Ghz at 5W puts the new AMD core very close to Intel's best in mobile. Whether that is true or not, it remains to be debated and proved, but I hope this little napkin math shows you we cannot take such a claim lightly, since it puts efficiency close to the best high performance core available on the market.

This 5W Zen @ 2.4Ghz claim must be met with "strict scrutiny" and extrapolating from dubious approximations of XV power usage in mobile is definitely not the way to start this conversation, at least not when real world data blatantly contradicts them.

The idle power usage of the HP Pavilion being used in that review was between 6-9W, you are so off the mark in you approximations that your LCD panel power consumption alone would completely exhaust idle power budget and leave nothing alone for the rest of the system. Even when approximating SSD power usage you conveniently consider the unit is continuously writing data during benchmarks, otherwise you would use the average idle power which is only 100 times lower than what you posted, at 30mW.

A really efficient laptop design with a 15W TDP CPU and no active dGPU will use around 3-5W while idle and 27-28W while under load after CPU will have exhausted turbo time (so it will start at 30-35W and go down after a while). Even a less efficient model will stay at 5-8W while idle and hang around 30W under load (after turbo period).

There's nothing wrong or special about the HP Pavilion in the review, except you're trying to inflate some power numbers in order to deflate others.

7W for the LCD was the only data i found after deep google digging.

Also in this digging I didn't find any other A10 or FX9800P review with power numbers. I found only the Acer E5-553G, that has also a discrete GPU. Anyway on cinebench R15 draw 40W, so we are close to INITIAL draw and performance of your review... My theory is that the APU went rapidly in thermal throttling to keep the surface temperature below 40-45C and that we must consider the maximum clock, obtained in non power constrained situation...
I also specify that 90% of efficiency for VRM and PSU is an exaggeration.

For the PSU i have found that if it's not marked 80 plus, it's efficiency is between 50% and 80%. Let's say 75%.
For the VRMs i have found up to 90% in ideal conditions. Let's say 85%
So from the 30W, we must subtract the 25%, having 22.5W from the PSU.
Then we subtract 6W (?) for the panel. 16.5W.
Then we subtract 2W for leds, battery recharge circuit, other chips (audio, wifi, trackpad controller), other stuff (cmos, cmos battery recharge) and the SSD.
We are left with 14.5W. Apply the 85% and we have 12.3W. Subtract the SODIMMs (2 i suppose) and we are at at most 12W, that is the minimum configurable TDP of the APU.
To have 15W for the APU we must draw 40W as in the other review, in which we arrived at 40W because also other components are stressed more (SSD, LCD, VRMs PSU)...

So we have 12W for 4XV core, the GPU, sitting mostly idle, because cinebench is a CPU load, same thing the SB, if the SSD isn't working, while the NB is at full load, as the SODIMMs. The NB should draw a few watts, but not much in a low power APU, so we are stick with at most 10W for 4 XV cores, on a 28nm BULK high power process.

So 5W for 2 XV cores at 2.1GHz on cinebench R15.
And this is not event the top model, that is the FX9800P.

Anyway going to a 14nm FF LOW POWER process, we should gain a few MHz or draw less power...
Zen should draw at most like 2 full XV cores (an XV module), having half the decoders, half the AGUs and less L2 cache...
So a Zen core @2.1GHz on the 28nm BULK should draw at most 5W.
Take this on the 14nm FF (were at low clock we had on a neon FPU more than double clock) and I think that 2.5GHz should be feasible in 5W... And so 32c Zen@2.5GHz in 180W...

 

[bjt2]

You have shown no such thing, the HP Pavilion is in no situation similar to a convertible/tablet, it is a 15" notebook with a sizeable chassis and adequate cooling for a 15W TDP CPU and a small dGPU on the side. This is how the unit looks like on the inside, with both a decent sized fan and radiator.

You really need to stop inventing reasons for this review/unit to be flawed and come up with another source of data that directly shows A10-9600P or higher 15W TDP SKU running Prime 95 at base clocks or higher.

And btw, keep in mind the minimum cTDP for A10-9600P is 12W, not 10W.

I adjusted for the 12W, but since is a full blown notebook, what is the reason for the lower power draw and clock after 5 minutes? The only reason I see is thermal throttling, going on 12W TDP... The numbers even adds up: 40W with cold CPU and 30 with hot CPU (12W on the APU). If we include the efficiency, 10W on the AC adapter translates in about 6W on the APU, giving an initial TDP of 18W...

 

[Dresdenboy]

And what workload runs the CPU that hard while also running the GPU as hard as possible as well?

You wouldn't what to see my haswell 4300u fall off the cliff with even simple games like diablo 3 at lowest rez+settings.

Notebookcheck sometimes provides CPU/GPU clock frequency graphs while running a game. This could help here.

And the steady reduction over time could be related to STAPM and increasing power consumption with rising temps.

 

[coercitiv]

For the PSU i have found that if it's not marked 80 plus, it's efficiency is between 50% and 80%. Let's say 75%.
For the VRMs i have found up to 90% in ideal conditions. Let's say 85%
So from the 30W, we must subtract the 25%, having 22.5W from the PSU.
Then we subtract 6W (?) for the panel. 16.5W.
Then we subtract 2W for leds, battery recharge circuit, other chips (audio, wifi, trackpad controller), other stuff (cmos, cmos battery recharge) and the SSD.
We are left with 14.5W. Apply the 85% and we have 12.3W. Subtract the SODIMMs (2 i suppose) and we are at at most 12W

The math you have just presented applies to any laptop, even to my Toshiba Z30-A. Considering that both notebookcheck and myself measured a maximum continuous power usage of the unit to around 27W, I take the opportunity to congratulate you in proving my i7 4510U is in fact a 10W TDP CPU! Except it isn't.

I make one last appeal to you, find and post relevant data from another source that shows 9600P or other higher SKU to use 15W at base clocks in Prime 95.

So a Zen core @2.1GHz on the 28nm BULK should draw at most 5W.

Here we go again, 28 nm Zen core beating the s**t out of 22nm Haswell core. You don't have a limit to what you can imagine, don't you?

 

[bjt2]

Notebookcheck sometimes provides CPU/GPU clock frequency graphs while running a game. This could help here.

And the steady reduction over time could be related to STPM and increasing power consumption with rising temps.

Another thing could be related to windows 10 power management. The default behaviour (that I have changed on my laptop) is to lower CPU clock, BEFORE increasing fan speed. The other is to increase FIRST fan speed and then lower the clock if it is not sufficient.
Anyway i don't see how a CPU can draw 15W on a full notebook, drawing 30W from the outlet, considering all the efficiencies and the components (first the LCD) in a notebook. And since the maximum consumption is 10W higher with the same load (so not using disk and GPU in any way higher), but in another moment with colder laptop, I suppose that it's in this situation the APU is drawing at least the rated TDP... I can admit that in some situation (cold CPU) the APU can draw more than the rated TDP... Also INTEL does it...

 

[coercitiv]

Notebookcheck sometimes provides CPU/GPU clock frequency graphs while running a game. This could help here.

FPS graph + reviewer comments, it takes about 90s for throttling to occur.

In games, the TDP limit is temporarily exceeded. However, after a maximum of two minutes (depending on the previous load and temperatures of the laptop), the 15-watt limit strikes back, dropping clock speeds. For example, in “Diablo III”, the CPU and GPU clock speeds start at 1800 and 550 MHz respectively. As the game runs, the speeds drop to 1100 to 1200 (CPU) and 380 to 420 MHz (GPU). The frame rates drop in parallel in our static gaming scene from 72 to 49 fps. To some extent, all current Intel ULV chips are also affected by this issue. The game benchmarks above reflect the long-term performance of the test model due to the length of the benchmarks and "hot" state of the laptop.

 

[bjt2]

The math you have just presented applies to any laptop, even to my Toshiba Z30-A. Considering that both notebookcheck and myself measured a maximum continuous power usage of the unit to around 27W, I take the opportunity to congratulate you in proving my i7 4510U is in fact a 10W TDP CPU! Except it isn't.

I make one last appeal to you, find and post relevant data from another source that shows 9600P or other higher SKU to use 15W at base clocks in Prime 95.

Already said that i can't find it.

Here we go again, 28 nm Zen core beating the s**t out of 22nm Haswell core. You don't have a limit to what you can imagine, don't you?

The FO4 is completely different. You can't compare. And Haswell have 256bit FPU that draw much power and leakage. And anyway with 128bit code all INTEL CPUs stays in a quite high turbo state even at full load (but 128 bit)

AMD already managed to beat Haswell on core and GPU clocks. On the 28nm and with the older XV architecture...
I am not saying that it would beat it in terms of performance. At least not with XV... Zen is another story...

Let's use high power components to limit thermal throttling.
The A12 9800 has 4 core at 3.8GHz/4.2 plus a 512 SP GPU at 800/1108MHz in 65W...
On the 28nm BULK.
What is a comparable Haswell APU? For TDP i have found this, but maybe is a notebook SKU...
The top is i7 4790S, 65W 3.2/4.0GHz. GPU 350-1200MHz...

Base clock of both is lower.
22nm vs 28nm...

Obviously the INTEL process is better.

But AMD managed to have higher clock...
Why?
The reason is FO4.
Simple.

EDIT: sorry if i can't find a 65W 2C/4T INTEL CPU... The comparison would have been better...

 

[coercitiv]

Anyway i don't see how a CPU can draw 15W on a full notebook, drawing 30W from the outlet, considering all the efficiencies and the components (first the LCD) in a notebook.

Toshiba Z50 A --> 15" FHD IPS screen, 16GB dual channel RAM, all the usual components including fan, WiFi, LAN, LEDs, audio & video ports, speakers etc --> 30W at the wall

The laptop can consume as much as 30.6 watts in full load, but that will rarely be the case in routine use, and it is still the lowest rate among the comparison devices.

 

[bjt2]

Toshiba Z50 A --> 15" FHD IPS screen, 16GB dual channel RAM, all the usual components including fan, WiFi, LAN, LEDs, audio & video ports, speakers etc --> 30W at the wall

So the APU in the HP notebook is drawing more than 15W at start? Because it draws 40W when starting cinebench... If it draws 15W when drawing 30W in total, then it should draw at least 22W when the notebook is drawing 40W... Almost 50% more than nominal...
There should be something wrong somewhere...

 

[jpiniero]

The HP notebook looks like it uses around 6-8 W idle.

Yeah I tend to believe that Zen Naples is 4x8C using 180 W at 1.44 Ghz.

Edit: Should mention that the Sandra score could be from a dual processor 16C model.

 

[KTE]

I have shown that on that particular model the power was capped at at most 10W. The same happened on the first asus yoga with the intel Y series.

Are you familiar with the new concept of surface temperature? Latest (mostly) fanless convertible, that uses all the chassis to dissipate the power, like a tablet, with surface temperature limited to 40-45C to not burn the user that usually put the convertible on his/her legs... This explain also the higher power draw at the start of the benchmark: if the temperature is below 40C the CPU is uncapped... 7 more W that translate in 5 more W on the APU (after the losses)... Right 15W at the bench start, that went to 10W when the surface temperature was too hot...

But then you say let's suppose "30W for a full system...".

Which that already exceeded even supposedly "capped at 10W", in average CPU loads (+41W).

But it was capped 12W CPU, not 10W BTW.

And here's the kicker of why your simplistic core/clock extrapolating model completely fails;

CB ST
Start 3.2 GHz (37.5 W) - Stable 2.9 GHz (30.7 W)

CB MT
Start 2.6 GHz (41.4 W) - Stable 2.1 GHz (30.8 W)

Prime95
Start 2.3 GHz (41.3 W) - Stable 1.7 GHz (31.1 W)

Diablo 3 CPU/GPU start 1800/550MHz
Stabilization 1100/380MHz (low point)

IF 4Core is max 12W@1.7GHz, you're looking at a budget of +23.5% in clocks for average MT loads and +88% as the advertised max 1Core turbo. A 47% lower full load clock vs low ST load.

IOW an extreme low load vs high load variance of 175MHz/W vs 142MHz/W.

The drop from 2.9->2.1GHz at the same power shows at least +38% difference in 1C vs 4C load clocks even in standard loads.

It also shows if the spread of 1x3.0GHz=12W and 4x2.1GHz=12W, then core clock@power scaling isn't anywhere near linear as you keep suggesting with multicore CPUs.

Assume 1W chip units + 1W idle for the 3Cores from a 12W total, allowing 10W to the Cores. I'm being very generous...

1C 2.9GHz @ 10W
1C 2.1GHz @ 2.75W
1C 1.7GHz @ 2.75W Max Load

Moving at low frequencies, from 2.1->2.9GHz, +38%, is a 364% power rise for 1 Core.

Hence why it's an LPP ULV SoC for mobile, optimized for low power, onchip graphics and low frequencies.

But an LP Mobile SoC is never an indicator to be able to extrapolate to a completely new HP Server/Desktop chip.

Sent from HTC 10
(Opinions are own)

 

[coercitiv]

So the APU in the HP notebook is drawing more than 15W at start? Because it draws 40W when starting cinebench... If it draws 15W when drawing 30W in total, then it should draw at least 22W when the notebook is drawing 40W... Almost 50% more than nominal...
There should be something wrong somewhere...

There's nothing wrong, letting the CPU go over 15W for a limited (controlled) amount of time is by far the best approach for a multitude of tasks, and AMD would be crazy not to do it, as long as it's properly addressed through power management (impose a secondary power threshold, limit max current, check temps etc). Intel has been doing that for a long time, and AMD worked quite a bit on every iteration of their low power APU to improve their power management and achieve similar results.

It just so happens that all this optimization also makes things very hard to evaluate for us enthusiasts, usually requires us to directly test the hardware or get lucky with a meticulous reviewer. For consumers it's a bit simpler, they buy the whole product, not just the CPU.

 

[The Stilt]

FPS graph + reviewer comments, it takes about 90s for throttling to occur.

I've seen that slope before.

Either the system hits the thermal limit (skin temperature) at 100s mark, or HP has reduced the stock duration (200s) in half.
Obviously the first option isn't too likely, since the time is exactly half