Battle against heat

[Mo5]

Quite some time ago, I've ranted how my motherboard PCI-e slots have been laid out in a way that isn't ideal for multi-GPU configurations - bottom card blocks most of the airflow coming towards top card. Anyway... this isn't the point of this topic. The point is - what is the most effective way to reduce heat while still achieving acceptable gaming performance.

 

I'll start by telling the basic system configurations and we'll go from there. My PC is running a AMD Phenom II X4 955, 4GB of RAM and (most importantly) 2 ATi HD5750 in Crossfire. All of those components are packed in a Cooler Master HAF 912 Plus.

 

I'll dive right into the results and explain how I got to them afterwards:

 

Most of the table is self explanatory, but let's go through it:

-Under CPU you can see 2 values - the first represents the amount of cores enables (in BIOS), the second number is clock speed. All the tests were conducted on stock voltage of 1,325V.

-Under GPU it's similar - first number corresponds with the number of GPU running while the other is core clock and then memory clock.

-The temperatures for CPU were taken with CoreTemp after a 45 minute Prime95 run.

-The temperatures for GPU were taken from MSI Afterburner after 3 runs of DMC4 benchmark (to simulate a realistic load) - top card then bottom card

-Both games come with their built-in benchmarking utility so I used that. In both cases I did 3 runs and averaged those out. In DMC4 I've taken Scene 2 and 3 (lowest and highest values) and in GRID Autosport I've taken the minimum and average FPS.

 

Both games were running at 1080p. In DMC4 all things cranked to the max (including 8xMSAA), in GRID Autosport I was running the high preset.

 

The results itself are interesting, particularly in dual core runs. With CF there was considerable amount of bottlenecking happening, but strangely one card at stock clocks (700/1150) performed just as expected. More strange results happened in DMC4 - the dual core configuration consistently scored the highest results than 3 and 4 cores. You can observe and analyze the other results yourself. I still find it amazing that even with one 5 year old midrange underclocked GPU and only 2 (Phenom II) cores you can get "console grade" experience in Grid Autosport - frame rate of around 30 on 1080p

 

So what is the bottom line? Well you get most performance out of 4 cores and 2 non-underclocked GPUs (shocker right?), but in my case I'm getting the best balance of thermals and performance with the dual core and a single stock clocked GPU.

[TheFlyingSquirrel]

Upgrade your cooling, first get a Noctua NH-D14 and see what you get, then change the rest to Noctua NF-A14(or some variation)

 

http://www.amazon.co.uk/NH-D14-Processor-LGA1366-LGA1156-LGA1155/dp/B002VKVZ1A/ref=sr_1_1?s=computers&ie=UTF8&qid=1419961185&sr=1-1&keywords=noctua+nh-d14

 

http://www.amazon.co.uk/Noctua-NF-A14-PWM-case-fan/dp/B00COH30A4/ref=sr_1_4?s=computers&ie=UTF8&qid=1419961282&sr=1-4&keywords=noctua

 

I run get 24C at idle and 48 at peak running the rig in my setup with games with BF4.

[Mo5]

Upgrade your cooling, first get a Noctua NH-D14 and see what you get, then change the rest to Noctua NF-A14(or some variation)

 

http://www.amazon.co.uk/NH-D14-Processor-LGA1366-LGA1156-LGA1155/dp/B002VKVZ1A/ref=sr_1_1?s=computers&ie=UTF8&qid=1419961185&sr=1-1&keywords=noctua+nh-d14

 

http://www.amazon.co.uk/Noctua-NF-A14-PWM-case-fan/dp/B00COH30A4/ref=sr_1_4?s=computers&ie=UTF8&qid=1419961282&sr=1-4&keywords=noctua

 

I run get 24C at idle and 48 at peak running the rig in my setup with games with BF4.

Getting a better cooling will just move the heat elsewhere (in the room). I want to remove it in the first place (plus 56°C isn't a bad result for a CPU).

[TheFlyingSquirrel]

Getting a better cooling will just move the heat elsewhere (in the room). I want to remove it in the first place (plus 56°C isn't a bad result for a CPU).

 

I am not sure I follow.  Better cooling will keep temps in the case down more efficiently; allowing for good overclocking, giving you better performance per temp.

[Mo5]

I am not sure I follow.  Better cooling will keep temps in the case down more efficiently; allowing for good overclocking, giving you better performance per temp.

I'm talking about heat all together - not just case and core temps. TDP if you wish.

[TheFlyingSquirrel]

I'm talking about heat all together - not just case and core temps. TDP if you wish.

 

TDP of your entire room?!   Well your room is a controlled environment, AC in the room? :unsure:

[Mo5]

TDP of your entire room?!   Well your room is a controlled environment, AC in the room? :unsure:

Nope, no AC and the summers do get quite warm. The reason why I'm doing this testing in the winter.

[TheFlyingSquirrel]

Nope, no AC and the summers do get quite warm. The reason why I'm doing this testing in the winter.

 

If you can't control the room temperature then I would focus on the computer environment.

[Mo5]

If you can't control the room temperature then I would focus on the computer environment.

isn't that exactly what I'm doing

[TheFlyingSquirrel]

isn't that exactly what I'm doing

 

These are my thoughts, you have two areas of the heat, the components inside the rig and the ambient temperature of the room in which it is in.   With air cooling the former is directly effected by the latter, my core temps are minimum of 5C hotter in the summer; you touched upon this above when you mentioned testing in winter. 

 

You can't eliminate the room temperature out of the equation as you have no AC, but you can minimise its impact with water cooling both the CPU and GPU with closed units and on exhaust fan at the back; this would give you cooler temps than air cooling during your very hot summers.

 

The reason I recommended improving the cooling is that CPU and GPU are great sources of heat that if neutralised improve the temps in the rig.

 

I may be misunderstanding what you are trying to do though.  

[Mo5]

These are my thoughts, you have two areas of the heat, the components inside the rig and the ambient temperature of the room in which it is in.   With air cooling the former is directly effected by the latter, my core temps are minimum of 5C hotter in the summer; you touched upon this above when you mentioned testing in winter. 

 

You can't eliminate the room temperature out of the equation as you have no AC, but you can minimise its impact with water cooling both the CPU and GPU with closed units and on exhaust fan at the back; this would give you cooler temps than air cooling during your very hot summers.

 

The reason I recommended improving the cooling is that CPU and GPU are great sources of heat that if neutralised improve the temps in the rig.

 

I may be misunderstanding what you are trying to do though.

Air or water, the heat produced by components will get into the room - fact.

What I'm trying to do is lower the heat produced by components by decreasing their performance (lowering their TDP effectively).

[TheFlyingSquirrel]

Air or water, the heat produced by components will get into the room - fact.

What I'm trying to do is lower the heat produced by components by decreasing their performance (lowering their TDP effectively).

 

Who said it wasn't fact.

 

Why the hell would you want to do that, architectural speaking you have a huge task, transistors output heat and accurately measuring how much is produced by each that any one time would be a guess at best.   You would need very specialised software/