A small warning regarding 100W APUs, stock cooler, small cases and highspeed RAM

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
May 11, 2008
19,300
1,129
126
It seems to have more effect on write rather then read speeds. I'm pushing slightly over 10800MB/s with 9-9-9-27 CR1 @ 1600MHz and NB @ 1800MHz. The "funny" bit is that I have higher write speeds then with the memory at 2133MHz (10500MB/s) and only slightly lower read speed...

I must say that 1.35V Crucial low-profile 1600MHz with 9-9-9-24 timings is looking very tempting right now. Should be an excellent compromise between performance and power consumption... :cool:

I wonder if you would get optimal results with 1866MHz memory when compared to 1600MHz and 2133MHz. It would make sense given the NBfrequency of 1800MHz.
I wonder what would happen if you could set your NB frequency at 1866MHz and memory at 1866MHz.
 
May 11, 2008
19,300
1,129
126
Did you check cpu frequencies and vcore? It seems highly unlikely to me that overclocking ram increases power draw by 30W. I don't think the ram slots can even deliver that much.

I think it is the memory controller in the apu that will consume most of the extra 30W @1,65V.
 

coffeejunkee

Golden Member
Jul 31, 2010
1,153
0
0
Well, A10 6800K is rated to work with 2133MHz ram normally. Seems to me 30W extra would push it out of TDP.

I'm still thinking the mobo is doing some sort of turbo overclocking, like many do with Intel cpu's as well (run all cores at max 4,4GHz and increase vcore as well).
 

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
Did you check cpu frequencies and vcore? It seems highly unlikely to me that overclocking ram increases power draw by 30W. I don't think the ram slots can even deliver that much.

I don't think its the RAM, but the memory controller in the APU that's responsible. Operating at high frequencies and voltage is bound to put out some extra heat, compared to other AMD APUs. Richland is also still 32nm, its the exact same die as Trinity. Remember that 2133MHz is the highest speed memory AMD officially supports. Every other APU model is limited to 1866MHz. Its likely that only a portion of the chips AMD fabs can tolerate running the memory at 2133MHz, mine might just be borderline. It certainly won't run 2400MHz without very loose timings or massively increasing controller voltage.

The average power of a DIMM is something like 2-3W.

It's strange though, I'll grant you.
 

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
Well, A10 6800K is rated to work with 2133MHz ram normally. Seems to me 30W extra would push it out of TDP.

I'm still thinking the mobo is doing some sort of turbo overclocking, like many do with Intel cpu's as well (run all cores at max 4,4GHz and increase vcore as well).

It is out of TDP, according to both HWinfo and Aida64 its drawing 120W+...

As for frequency it work just as intended. When the IGP is active, frequency is limited to 4.1GHz.
 

ehume

Golden Member
Nov 6, 2009
1,511
73
91
The stock cooler can keep temperatures down below 70C with the case lid off.

Perhaps you need a new case. If taking the side cover off improves cooling, you have a bad cooling solution. A good cooling solution begins with a case/heatsink combination that lowers temps when the side panel is attached.

I am wondering if your temperature problem comes from your running up against the thermal limits of your case. As an example, my i7 4770K has moderate temps with Linpack/AVX2 until I get to a certain point. Above a certain Vcore, the temps shoot up wildly high within seconds. It is because the TIM under the IHS will transmit heat from the CPU, but when it reaches a certain point it will transmit no more heat. It is at that point where my core temps go wild.

I am thinking your case/heatsink will dissipate heat up to a point, beyond which the temps get really high. You can test this by running your system with the side panel off, or moving your system completely out of the case.

You may simply need a modern case.
 

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
You may simply need a modern case.

If things where that simple, I'd just put it in a mini-tower case, slap a BIG tower cooler on it and be done with it. Unfortunately I can't do that, because I'm height limited. The case HAS to fit in a ~13cm (~5") high space.

The Silverstone ML03 is a fully modern low-profile case BTW. It just has a slight problem with a cross-bracer due to my ITX board having the cooler located slightly differently then your average mATX board. So you can basically only use the stock cooler without interfering with that. I might try an Alpine 64GT Rev. 2, I think that'd fit. Only that cooler is only rated for 70W and the stock FX cooler is just a little too high...
 
Last edited:
May 11, 2008
19,300
1,129
126
Maybe this is an idea ?
I once created a cooler solution from two different coolers.
I had a small heatsink that would only support small 70 cm fans.
The heatsink had to fit for socket A. The fan came from an Athlon 64 cpu socket cooler. I took it apart and sawed a part of the plastic fan retention and drilled holes in it for the screws to attach it to the socket A heatsink. This all to make it fit in my case. But with a 85 to 90 mm very silent fan. :)

IMG_0114.jpg


IMG_0112.jpg
 
Last edited:

ehume

Golden Member
Nov 6, 2009
1,511
73
91
If things where that simple, I'd just put it in a mini-tower case, slap a BIG tower cooler on it and be done with it. Unfortunately I can't do that, because I'm height limited. The case HAS to fit in a ~13cm (~5") high space.

The Silverstone ML03 is a fully modern low-profile case BTW. It just has a slight problem with a cross-bracer due to my ITX board having the cooler located slightly differently then your average mATX board. So you can basically only use the stock cooler without interfering with that. I might try an Alpine 64GT Rev. 2, I think that'd fit. Only that cooler is only rated for 70W and the stock FX cooler is just a little too high...

What an interesting problem. You might consider putting your specs in your sig, and posting a pic of your rig in place. Clearance between the top shelf and the case makes a big difference.

Also, with the case under a shelf, no one will see the top of your rig, so you could leave the top panel off. I know that sounds odd, but like yours, my rig sits under a shelf. I cut out the top grills of my prior case to improve airflow.
 
Last edited:

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
Maybe this is an idea ?
I once created a cooler solution from two different coolers.
I had a small heatsink that would only support small 70 cm fans.
The heatsink had to fit for socket A. The fan came from an Athlon 64 cpu socket cooler. I took it apart and sawed a part of the plastic fan retention and drilled holes in it for the screws to attach it to the socket A heatsink. This all to make it fit in my case. But with a 85 to 90 mm very silent fan. :)

Interesting idea. I -was- toying with the idea of making a wooden "adaptor" to fit an 80mm (or perhaps 92mm) fan in the mounting holes the stock fan. That ought to improve airflow over the heatsink significantly.

For the moment, I ordered one of those 64GT's. It is so cheap it won't hurt anything. Its properly not much better then the stock cooler, but at least it ought to be quieter. It also has the advantage of directing airflow in only two directions, where it can be "picked up" by the 80mm fan(s) and exhausted from the case. I have a feeling the design of the stock heatsink is creating a vortex inside the case, where hot air is sucked continuously through the fan.

What an interesting problem. You might consider putting your specs in your sig, and posting a pic of your rig in place. Clearance between the top shelf and the case makes a big difference.

Also, with the case under a shelf, no one will see the top of your rig, you you case leave the top panel off. I know that sounds odd, but like yours, my rig sits under a shelf. I cut out the top grills of my prior case to improve airflow.

If I can't get temperatures under control, that's exactly what I'll do... ;)

Though at the moment I'm only seeing 70C maximum at full load. The Richland chip should be good for 80C, so I don't think that ought to hurt anything.
 
May 11, 2008
19,300
1,129
126
Insert nickname, i was wondering about if 1800MHz might be your memory clock sweetspot. Did you by any chance found some time to test it to see if it really is better than 1600MHz or 2133MHz ?
Can you even get the memory clock at exactly 1800MHz ?
I am curious.
 

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
I got that 64GT yesterday. Now that is one wicked little cooler. Absolutely no problems keeping everything nice and cool. The temperature sensor must be a little off, it records an idle GPU temperature of -4C. Even running full bore (2250RPM) its whisper quiet too... :cool:

Its so good I've been wondering why AMD hasn't licensed the design.

Insert nickname, i was wondering about if 1800MHz might be your memory clock sweetspot. Did you by any chance found some time to test it to see if it really is better than 1600MHz or 2133MHz ?
Can you even get the memory clock at exactly 1800MHz ?
I am curious.

Doesn't seem to make much difference running at 1:1 NB-RAM. Results are pretty much the same as 1600MHz. I can't tell if there really is an improvement because it's within standard testing anomaly. Timings still seem to have far more effect on memory performance then frequency and NB clock. I still get best write speeds from 1600MHz 9-9-9-27 CR1, that seems the sweet spot for some reason. I really wish I had access to a kit that could do something like 8-8-8-24 at 1600MHz, I wonder how that'd perform.
 
May 11, 2008
19,300
1,129
126
Doesn't seem to make much difference running at 1:1 NB-RAM. Results are pretty much the same as 1600MHz. I can't tell if there really is an improvement because it's within standard testing anomaly. Timings still seem to have far more effect on memory performance then frequency and NB clock. I still get best write speeds from 1600MHz 9-9-9-27 CR1, that seems the sweet spot for some reason. I really wish I had access to a kit that could do something like 8-8-8-24 at 1600MHz, I wonder how that'd perform.

Interesting, well thanks for trying :).
I will try to get my memory to run as well @ 8-8-8-24 CR1.
I will let you know how it performs, if my pc will pas memtest.
 
May 11, 2008
19,300
1,129
126
I have not yet tried to run the memory @ 8-8-8-24, 1600MHz.

But i did take some power measures which i was also interested in.
A10-6700.
No discrete GPU.

I measured power at the 230V connection. I measured with a voltcraft energy check 3000.
125319_BB_00_FB.EPS_1000.jpg


I measured power draw during various moments and it is interesting to know that W7 is already reducing power consumption at the log in screen.
Off : 1,1W.
BIOS : 99W - 101W .
Windows 7 login screen : 33,8W - 42,4W. Fluctuations because of HDD read activity.
W7 after login and no HDD activity : 33W.
Running prime 95 with 4 worker threads(3,4GHz - 4,2GHz) : 121,1 W - 122W.
Running prime 95 with 1 worker thread (@4,2GHz) : 80,9W.

With these numbers i would say a good cooling solution is really needed. Amazing. And the VRM on the board do a great job keeping power stable while power draw jumps up and down rapidly.
 
Last edited:
May 11, 2008
19,300
1,129
126
I noticed that doing activity such as browsing or forum reading while playing music, the power consumption is around 35W -38W at the 230V inlet.
 

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
I noticed that doing activity such as browsing or forum reading while playing music, the power consumption is around 35W -38W at the 230V inlet.

I'm on 230V too (Denmark. Actually I have 238V input, as I live pretty close to the transformer station. Too close for my poor light-bulbs... :(). PSU used is a Corsair CX430 (great for a cheap PSU BTW). At the wall, I'm using between 25 and 35W, browsing, using the desktop. Full idle is 22W. Video decoding is quite dependent on what is being decoded, and range from 25W to 30W doing 1080p youtube to slightly below 40W watching bluray. But I'm using an ITX board, so its pretty much in line with your numbers.
 
May 11, 2008
19,300
1,129
126
While i was making food, i did some testing with timings 8-8-8-24 @1600MHz CR1 and did not see an increase in memory speed. I did saw a reduction in latency as can be expected. I have not done all test so at the moment i am running the memory again at timings 9-9-9-24@1600MHz CR1.

It seems i even lost bandwidth or i need to do the tests again.
I do have a virusscanner running.

cachemem88824.png


gpgpu88824.png
 
May 11, 2008
19,300
1,129
126
I'm on 230V too (Denmark. Actually I have 238V input, as I live pretty close to the transformer station. Too close for my poor light-bulbs... :(). PSU used is a Corsair CX430 (great for a cheap PSU BTW). At the wall, I'm using between 25 and 35W, browsing, using the desktop. Full idle is 22W. Video decoding is quite dependent on what is being decoded, and range from 25W to 30W doing 1080p youtube to slightly below 40W watching bluray. But I'm using an ITX board, so its pretty much in line with your numbers.

I have an Antec PSU earthwatts EA-380D.
The MB is a microATX from gigabyte : GA-F2A75M-HD2.

Interesting that you have even lower idle numbers than me.
I think that the 100W richland APU's are really the best binned parts.
I think that the 65W richland APU is second best.
Especially when i notice that my power consumption is quite high with 4 cores running at maximum , 120W ! :eek:
 
May 11, 2008
19,300
1,129
126
120W input.
At 85% efficiency for my PSU : 102W output.
DIMMS = 2x 3W estimated.
HDD = 5,5W estimated.
A75 chipset = 7W estimated.
Fans 2,4W.
102 - (6+5,5+7+ 2,4) = 81.1W

VRM at 90% efficiency : 81.1 * 90% = 72.99W consumption of APU power while turbocore is active.
I would suggest that the TDP number is strange as more people have discovered.
 
Last edited:

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
I have an Antec PSU earthwatts EA-380D.
The MB is a microATX from gigabyte : GA-F2A75M-HD2.

That's an excellent PSU too. Its actually one of my top choices when I build. Which of the CX430/EA-380D I pick depends entirely on daily price... ;)

Your MB is too small to be a micro-ATX or FlexATX and too large to be DTX. Its more of an ITX-with-extra-expansion-ports. Its a very interesting form factor, and now that I've looked, it seems Gigabyte uses it for quite a number of boards (both AMD and Intel). That will certainly get due consideration in the future. It seems perfect for general use PC, where you might need more then the single expansion port of ITX but not necessarily the 4 ports of mATX, and where power efficiency is king...

Interesting that you have even lower idle numbers than me.
I think that the 100W richland APU's are really the best binned parts.
I think that the 65W richland APU is second best.
Especially when i notice that my power consumption is quite high with 4 cores running at maximum , 120W ! :eek:

Interesting indeed. By rights you should be lower then me. Perhaps there is an efficiency difference between the power delivery circuitry on the MB...?
 
May 11, 2008
19,300
1,129
126
That's an excellent PSU too. Its actually one of my top choices when I build. Which of the CX430/EA-380D I pick depends entirely on daily price... ;)

Your MB is too small to be a micro-ATX or FlexATX and too large to be DTX. Its more of an ITX-with-extra-expansion-ports. Its a very interesting form factor, and now that I've looked, it seems Gigabyte uses it for quite a number of boards (both AMD and Intel). That will certainly get due consideration in the future. It seems perfect for general use PC, where you might need more then the single expansion port of ITX but not necessarily the 4 ports of mATX, and where power efficiency is king...
It is a great MB. It has a dedicated x16 pci-e for a discrete gfx card as well.


Interesting indeed. By rights you should be lower then me. Perhaps there is an efficiency difference between the power delivery circuitry on the MB...?

I did some calculations based on some estimates. See post 69.
I think the A75 chipset might consume a little more, maybe 7 watts. And i do notice that my APU clocks down when the apu gets too hot. So ~73W of power with turbo core overclocking does sound okay while knowing it is an apu with a 65W TDP.

EDIT:
Perhaps i have an APU that just consumes more or the MB has lower efficiency. I do not know. 11W difference (33W -22W) is a lot.
 
Last edited:
May 11, 2008
19,300
1,129
126
What keeps me wondering and thinking :
I know TDP and electrical input power for the APU is not the same but an estimate can be made. But simplified i could explain it like this...
The APU is made up of tiny mosfet transistors and every time the gate capacity of such a mosfet is charged or discharged, a current flows. This tiny (nanoamperes to microamperes) current flows through resistive parts of the chip. Mostly that would be tiny mosfets conducting which have a little resistance between the drain and source called RDSon. Then there also all the resistive interconnects. And since there are millions to billions of mosfets switching at the rate of the clockspeed, we end up with very large input currents. And all that current flows through resistive parts of the chip and combined this causes the current to be dissipated away in heat. And that is where the TDP and appropriate heatsink is needed for. To be able to dissipate away that heat.
 
Last edited:

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,691
136
I think the A75 chipset might consume a little more, maybe 7 watts.

Well, the A75 is a carry-over from Llano. So it -may- be built on an older process node. On the other hand the A75 should be simpler then the A85X... but I digress... :confused:

EDIT:
Perhaps i have an APU that just consumes more or the MB has lower efficiency. I do not know. 11W difference (33W -22W) is a lot.

Do you by any chance use a hard-drive as the system drive? If so that could account for 5-10W extra compared to my SSD, especially the 7200RPM variety. Toss in slightly less efficient power delivery circuitry and a slightly larger board, with some extra ICs to power and it could account for 11W... :hmm:
 
May 11, 2008
19,300
1,129
126
Well, the A75 is a carry-over from Llano. So it -may- be built on an older process node. On the other hand the A75 should be simpler then the A85X... but I digress... :confused:



Do you by any chance use a hard-drive as the system drive? If so that could account for 5-10W extra compared to my SSD, especially the 7200RPM variety. Toss in slightly less efficient power delivery circuitry and a slightly larger board, with some extra ICs to power and it could account for 11W... :hmm:

I use a 7200RPM HDD , i am planning to buy an SSD but have not done so yet. All the reinstalling is what is holding me back.
All combined that might explain it. Factor in that my power meter may also have some tolerance.
A HDD and an older chipset.
 
Last edited: