Overclock.net - An Overclocking Community - View Single Post - AMD Kaveri Refresh A10-7870K Apu arrives.

View Single Post
post #355 of (permalink) Old 10-08-2015, 03:23 AM - Thread Starter
New to Overclock.net
Himo5's Avatar
Join Date: Dec 2011
Posts: 968
Rep: 87 (Unique: 57)
This excersize to find the overclocking potential of the A10-7870K is based on 1404 runs of 39 benchmarks producing 2880 results. The benchmarks were run with the CPU, RAM and iGPU set to standard values with their potential forecasted on overclocking the CPU to 4750MHz, the RAM to 2616MHz or the iGPU to 1140MHz, expressed as a percentage of the three stock operating frequencies of Cpu:3900MHz, Ram:1600MHz and iGpu:866MHz.

The installation used for the excersize was based on an Asus A88X-Pro motherboard with an A10-7870K Apu, 2x4GB G.Skill 2666MHz Cl11 Dimms, Phanteks PH-TC14pe Air Cooler and a Seasonic 850Watt KM3 Gold Psu.

The excersize consisted of two tests; the first was conducted in Windows 8.1Pro to compare results from raising RAM through 1600/1866/2133/2400/2616MHz while keeping CPU at 3900MHz and iGPU at 866MHz. The G.Skill 2666MHz 2x4Gb Dimms were set to the required frequencies using the following - fairly loose - timings:

The second test took place in Windows 10 Pro between the 1st and 5th weeks after launch on July 29th, 2015 and compared the results with RAM held at 1600MHz while the CPU was raised through 3900-4500MHz and the iGPU through 866-1040MHz. Note that the Base Clock of both tests was the same and that the W10 results all out-performed those under W8.1Pro.

With the exception of the 2616MHz Ram Test all overclocks were chosen to be well within standard parameters and the only alteration from standard voltages was made automatically in BIOS when VDDNB is raised to 1.3875V when setting the iGpu to 1040MHz.

Each benchmark is run four times in each overclock and an average of the four results is taken forward as a percentage of the average result of the Base Clock, Cpu:3900MHz; Ram:1600MHz; iGpu:866MHz. The first five overclocks test the effect of raising Memory Frequency through 1600/1866/2133/2400/2616MHz and the next four test raising the CPU and iGPU from the Base Clock to 4500MHz and 1040MHz respectively. In benchmarks where lowest result is best, marked %(-), the sign of subtractions is reversed.

The resultant overclock potential is expressed both as an observed example and as a forecast extrapolated on part of that example. Thus, the Cpu forecast is calculated as (%(4500x866)-%(3900x866))*((4750-3900)/(4500-3900)), and the Gpu forecast as(%(3900x1040)-%(3900x866))*((1140-866)/(1040-866)). In the Memory Test the 5th result - for 2616MHz - is not used to calculate the observed potential, which is based on the result for 2400MHz. The 2616MHz result is given as the forecast.

Assuming a 4% tolerance Overclock Potential is given as one of three value marks: [x] Non-existant to Low (anything below 4%); [+] Measurable (4%-10%); and [*] Significant (anything above 10%). Fractions are prevented from rounding up through the two critical values, 4% and 10% so that marginal results cannot cross these boundaries.

I am leaving the implications of what this all means for now, just to concentrate on introducing the findings, so the overall results divide into 15 groups of overclocking potential as shown in the following Spoiler together with a comparison chart for the three overclock types:

The benchmark suites involved in the excersize are listed below with links that in most cases will acquire the versions I used (I wasn't able to find a link for RatGPU v0.8.2). However, it matters more that the same benchmarks were run in all the tests than what results were achieved. I collected a full set of proofs for the Ram tests but the Cpu/iGpu tests were initially only meant as a preliminary set of single tests to help plan a more ambitious program and so no proofs were collected when I decided to expand them into the full test instead. However, with the exception of the 2616MHz Ram Test all the results should be reproducible.

Cinebench R10 [http://downloads.guru3d.com/CineBenc...load-1708.html]
Cinebench R11.5 [http://downloads.guru3d.com/CineBENC...load-2475.html]
Cinebench R15 [http://www.guru3d.com/files-details/...-download.html]
3DMark [http://www.guru3d.com/files_details/..._download.html]
3Dmark11 [http://www.guru3d.com/files_details/..._download.html]
3Dmark Vantage [http://www.futuremark.com/benchmarks/3dmark-vantage]
PcMark7 [http://www.guru3d.com/files_details/..._download.html]
PcMark8 [https://www.techpowerup.com/download...rk-8-v2-4-304/]
FFIV Heavensward [http://na.finalfantasyxiv.com/benchmark/]
Unigine Heaven [https://unigine.com/products/benchmarks/heaven/]
Unigine Valley [https://unigine.com/products/benchmarks/valley/]
Unigine Tropics [https://unigine.com/en/products/benchmarks/tropics]
Unigine Sanctuary [https://unigine.com/en/products/benchmarks/sanctuary]
RatGPU v0.9.1 (was v0.8.2) [http://ratgpu.com/]
Furmark v1.15.1.0 [http://www.geeks3d.com/download/416]
RealBench [http://rog.asus.com/file/RealBench_v2.4.zip]
Black Hole [http://www.guru3d.com/files-details/...-download.html]
SuperPi [http://www.putme.net/SuperPI190.zip](Run as administrator)
WPrime [http://www.putme.net/wPrime210.zip](Run as administrator)
Combpubench [https://compubench.com/windows-download]
NovaBench [https://novabench.com/download.php]
Luxmark v2 [http://www.luxrender.net/wiki/LuxMark_v2]

The following are the detailed results.

Cinebench Results

3DMark Results

PcMark Results

Unigine Results

Furmark, Heavensward & Black Hole Results

Maths Benchmark Results

Realbench, Compubench and Novabench Results

Luxmark v2 Results

I've just spotted a mistake in my chart generator where I forgot to specify the parameters of the 2616MHz overclock. Instead of being Cpu:3900;Ram:2616;Gpu:866;Os:W8.1 that part of each table should read Cpu:3924;Ram:2616;Gpu:850:Os:W8.1 to conform with the 2616 overclock used during the tests. Apologies for the oversight.

Last edited by Himo5; 04-26-2019 at 03:54 AM. Reason: Removing Spoiler commands
Himo5 is offline