G3258 Memory and Overclocking Project Reults -01

[DonkeyOatie]

I have been running through the sorts of tests I expect my Middle School students to perform for their science fair project. The system used is in this thread http://www.tomshardware.com/forum/id-2626627/build-log-middle-school-science-fair-project-system.html and some preliminary overclocking data is here http://www.tomshardware.com/forum/id-2694458/middle-school-science-fair-investigation-input-advice.html

This thread is for the summary of the data I have collected. The raw data will be provided once my students have duplicated my work.

The object of the project is to investigate the effect of memory speed and configuration on the G3258 at stock and overclocked speeds, with an without a GPU, using a variety of benchmarks to simulate real-life use.

I'm looking at three use scenarios.

1. General-purpose light productivity, gaming, and editing. PCMark 7 and PCMark 8 Accelerated and Conventional are used to measure this.

2. Heavy gaming, or a least as good as the system will do. 3DMark FS, SD, CG, and IS, and Unigine Heaven and Valley are used to measure this.

3. Heavy numeric processing and computing. Caselab's Euler 3D fluid dynamics simulation is used for this, and tested at 1 to 8 threads to test processor load and operating characteristics. This test has no graphics component.

A. CPU OVERCLOCKING
3.2 Ghz to 4.2 Ghz (as good as I could get stable at reasonable voltages on stock cooler)
Theoretical performance improvement = 4.2/3.2 or +31.25%

1. GP use PCMark 7 = + 11.34%, PCMark 8 C = + 12.28%, PCMark 8 A = + 16.72%

2. Gaming, IS = +8.60%, CG = + 9.74%, SD = +0.99%, FS = +3.33%, UH Low -0.61%, Ultra +0.00%, UV Low +0.26, Ultra +-0.57%

3. Numeric 1 = + 15.26%, 2 = 14.31%, 3 = 13.87%, 4 = 21.64% 5 = 11.29%, 6 = 14.45%, 7 = 13.05% 8 = 14.32%

The G3258 will overclock on most (all?) motherboards and 4.2Ghz is less than many have gotten. I have a second G3258 and will see if it is any better or worse. This kind of overclocking is cheap, safe-ish..

B. iGPU OVERCLOCKING
Running at 3.2Ghz, but overclocking the iGPU to 1500Mhz. (stock is 1100 Mhz)
Theoretical performance improvement = 36.36%

1. GP use PCMark 7 = + 3.92%, PCMark 8 C = + 2.90%, PCMark 8 A = + 2.19%

2. Gaming, IS = +16.71%, CG = + 18.50%, SD = +28.56%, FS = +30.26%, UH Low + 28.40%, Ultra 33.03%, UV Low +29.06%, Ultra +30.68%

3. Numeric 1 = -0.99%, 2 = -2.20%, 3 = -1.30%, 4 = -1.74% 5 = -0.66%, 6 = -1.53%, 7 = -0.66% 8 = -0.56%


C. Memory OVERCLOCKING
Running CPU at 3.2Ghz and iGPU at 1100Mhz but overclocking memory from 1600Mhz to 2400Mhz.
Theoretical performance improvement = 50%

1. GP use PCMark 7 = + 2.78%, PCMark 8 C = + 2.29%, PCMark 8 A = + 6.30%

2. Gaming, IS = +2.64%, CG = + 4.04%, SD = +2.70%, FS = +5.64%, UH Low + 3.58%, Ultra 1.82%, UV Low +3.40%, Ultra +2.27%

3. Numeric 1 = +10.82%, 2 = +12.90%, 3 =+11.63%, 4 = +19.82% 5 = +11.44%, 6 = +12.82%, 7 = +10.99% 8 = +12.55%

D. SHOOT FOR THE MOON
Running the CPU at 4.2 Ghz, the iGPU at 1300Mhz, and the memory at 2400Mhz. Any attempt to increase one, resulted in a drop in something else to reach stability on the stock cooler.

1. GP use PCMark 7 = + 16.42%, PCMark 8 C = + 17.22%, PCMark 8 A = + 22.52%

2. Gaming, IS = +23.01%, CG = + 24.89%, SD = +20.84%, FS = +23.33%, UH Low +20.25%, Ultra +19.09%, UV Low +20.73, Ultra +18.18%

3. Numeric 1 = + 33.08%, 2 = 32.20%, 3 = 31.41%, 4 = 40.13% 5 = 31.63%, 6 = 32.66%, 7 = 30.01% 8 = 32.80%

I'm not suggesting that any of these options are particularly sensible for gaming. The Firestrike gets to 500 at best, which is a slow slide-show and the Unigine Heaven and Valley at Ultra would be unplayable. However, for general use and numeric calculations, the improvement with overclocking of CPU and memory is comparatively cheap for a big increase in performance.

The memory that can be overclocked to 2400Mhz comes at a 25% price increase.

 

[Frozen Fractal]

Wow, that's one wonderful data. Might wanna do graph charts for each stage to demonstrate gain & loss on the benchies - should come helpful.

 

[DonkeyOatie]

Here's what happens when a discrete GPU is added. The GPU was donated from a student build when the GPU was upgraded to a GTX 980. The GPU I got is the ASUS GTX970-DCMOC-4GD5, a 1.09Ghz compact GTX 970.

STOCK with GTX 970
Running CPU at 3.2 Ghz and memory at 1600 Mhz, and GTX970-DCMOC-4GD5 replacing iGPU.

1. GP use PCMark 7 = + 3.44%, PCMark 8 C = + 26.54%, PCMark 8 A = + 89.95%

2. Gaming, IS = +116.37%, CG = + 169.35%, SD = +704.41%, FS = +1694.62%, UH Low +1376.38%, Ultra +2461.82%, UV Low +962.30, Ultra +2105.68%

3. Numeric 1 = -0.19%, 2 = +1.23%, 3 = +0.88%, 4 = +7.10% 5 = +1.38%, 6 = +0.64%, 7 = +1.32% 8 = 1.10%

FULL BLAST
4.2Ghz CPU, 2400Mhz memory and GTX970 GPU

1. GP use PCMark 7 = + 14.67%, PCMark 8 C = + 45.04%, PCMark 8 A = + 110.26%

2. Gaming, IS = +170.70%, CG = +230.48%, SD = +831.42%, FS = +1899.23%, UH Low +1714.11%, Ultra +2486.36%, UV Low +1285.86, Ultra +2507.39%

3. Numeric 1 = +27.20%, 2 = +28.59%, 3 = +26.13%, 4 = +36.48% 5 = +23.46%, 6 = +27.91%, 7 = +25.50% 8 = 27.96%

So a really good GPU is great for gaming, not so much for anything else (unless your math uses CUDA)

 

[Frozen Fractal]

FULL BLAST
4.2Ghz CPU, 2400Mhz memory and GTX970 GPU

2. Gaming, IS = +170.70%, CG = +230.48%, SD = +831.42%, FS = +1899.23%, UH Low +1714.11%, Ultra +2486.36%, UV Low +1285.86, Ultra +2507.39%

Holy moly!

 

[DonkeyOatie]

I find that the less than doubling (170%) for Ice Storm to be more interesting. By looking at the details of the raw data (which I have not published), we can identify the tasks that are GPU, memory, and CPU limited.

 

[Frozen Fractal]

I think you should plot charts for these tests. That'll help visualize the research better.

 

[DonkeyOatie]

Here's how memory configuration affects performance. For this data 2 x 4Gb is the 'standard' The memory speed is 1333Mhz, which is the default for the G3258 and can be achieved on any motherboard.

1333Mhz Memory
There are four data sets, in order they are 1 x 4 Gb, 1 x 8 Gb, 2 x 8 Gb, and 2 x 4 Gb + 2 x 8 Gb (with 2 x 4Gb being the reference)

1. GP Use
a. PCMark7-- -10.88% -09.78% +0.32% -0.11%
b. PCMark8C -07.27% -07.59% -0.44% -0.69%
c. PCMark8A -07.13% -07.21% -0.04% -0.32%

2. Gaming
d. IS--- -14.4% -14.27% +0.34% -0.35%
e. CG-- -09.77% -10.45% -0.31% -0.59%
f. SD-- -08.73% -09.18% -0.67% -0.78%
g. FS-- -11.37% -11.63% -0.78% -1.03%

h. UHL -19.12% -18.81% +0.00% -0.31%
j. UHU -08.33% -09.26% +0.00% -1.85%

k. UVL -17.15% -19.00% -1.06% -1.58%
m. UVU -10.34% -10.92% +0.57% +0.00%

3. Numeric
n. 1. -05.08% -08.24% -1.48% +3.22%
p. 2. -12.46% -13.73% -2.16% -2.12%
q. 3. -10.96% -11.98% -1.10% -1.21%
r. 4. -12.29% -13.66% -1.66% -1.81%
s. 5. -12.84% -13.51% -1.64% -1.71%
t. 6. -12.22% -13.28% -2.05% -2.12%
u. 7. -11.36% -12.40% -1.57% -1.53%
w. 8. -08.17% -09.62% +2.55% +2.51%

All of these benchmarks are run as the ONLY task on the computer, so the benefits of extra memory for additional processes is not tested; that is for another time.

 

[DonkeyOatie]

I think you should plot charts for these tests. That'll help visualize the research better.

The chart is very dense and a bit confusing too. I think that most people would want to know the effect of a specific configuration rather then the sort of comparison a chart provides.

 

[DonkeyOatie]

Here's how memory configuration affects performance. For this data 2 x 4Gb is the 'standard' The memory speed is 2400Mhz, which is can be achieved for a G3258 on the Z97 board (but not necessarily on the cheaper H81, B85, or H97 boards)

2400Mhz Memory
There are only three data sets, in order they are 1 x 4 Gb, 1 x 8 Gb, and 2 x 8 Gb (with 2 x 4Gb being the reference). The 2 x 4 + 2 x 8 configuration was eliminated because the first series of tests did not show much difference between that and the 2 x 8. Again, this may be an area for future investigation.

1. GP Use
a. PCMark7-- -03.24% -04.49% -0.04%
b. PCMark8C -03.10% -04.00% -0.39%
c. PCMark8A -03.10% -03.96% -0.39%

2. Gaming
d. IS--- -05.25% -06.68% -0.08%
e. CG-- -04.48% -06.29% -0.79%
f. SD-- -03.22% -05.37% -0.64%
g. FS-- -06.55% -10.19% -2.91%

h. UHL -06.80% -08.58% -0.59%
j. UHU -02.68% -05.36% +0.00%

k. UVL -05.82% -07.34% -0.51%
m. UVU -03.89% -05.56% +0.56%

3. Numeric
n. 1. -03.72% -05.98% -1.86%
p. 2. -08.26% -11.14% -1.82%
q. 3. -07.07% -09.89% -2.40%
r. 4. -08.49% -11.41% -1.71%
s. 5. -08.57% -09.91% -1.73%
t. 6. -07.94% -11.30% -1.95%
u. 7. -07.53% -09.11% -1.42%
w. 8. -08.39% -10.99% -2.20%

All of these benchmarks are run as the ONLY task on the computer, so the benefits of extra memory for additional processes is not tested; that is for another time.

 

[Frozen Fractal]

The chart is very dense and a bit confusing too. I think that most people would want to know the effect of a specific configuration rather then the sort of comparison a chart provides.

Oh sorry, I meant to plot graphs, not charts

 

[DonkeyOatie]

The chart is very dense and a bit confusing too. I think that most people would want to know the effect of a specific configuration rather then the sort of comparison a chart provides.

Oh sorry, I meant to plot graphs, not charts

OK. That makes more sense. I may do that later, although you are free to help .

 

[DonkeyOatie]

Here's how memory configuration affects performance with a GTX970 GPU. For this data 2 x 4Gb is the 'standard' The memory speed is 1333Mhz, which is the default for the G3258 and can be achieved on any motherboard. CPU speed is 3.2Ghz

1333Mhz Memory + GTX 970
There are three data sets, in order they are 1 x 4 Gb, 1 x 8 Gb, and 2 x 8 Gb (with 2 x 4Gb being the reference)

1. GP Use
a. PCMark7-- -00.76% -01.31% +0.43%
b. PCMark8C -01.12% -02.33% -1.15%
c. PCMark8A -02.61% -03.25% -2.00%

2. Gaming
d. IS--- -03.18% -06.05% -0.68%
e. CG-- -00.97% -02.69% +0.00%
f. SD-- -01.46% -03.30% -0.25%
g. FS-- -00.64% -02.27% +0.07%

h. UHL -02.44% -05.95% -01.38%
j. UHU +00.90% -00.82% -11.26%

k. UVL -03.16% -05.35% -10.62%
m. UVU -00.88% -03.96% -10.54%

3. Numeric
n. 1. -04.72% -06.89% -0.19%
p. 2. -11.85% -15.04% -3.04%
q. 3. -10.31% -13.49% -1.16%
r. 4. -11.84% -14.72% -1.07%
s. 5. -12.84% -14.26% -1.07%
t. 6. -11.52% -14.85% -0.94%
u. 7. -10.11% -12.64% -0.45%
w. 8. -10.04% -13.85% +0.59%

Here's how memory configuration affects performance. For this data 2 x 4Gb is the 'standard' The memory speed is 2400Mhz, which is can be achieved for a G3258 on the Z97 board (but not necessarily on the cheaper H81, B85, or H97 boards)

2400Mhz Memory + GTX 970 This is a maximum run, so CPU speed is 4.2Ghz.
There are three data sets, in order they are 1 x 4 Gb, 1 x 8 Gb, and 2 x 8 Gb (with 2 x 4Gb being the reference).

1. GP Use
a. PCMark7-- -00.77% +00.18% -0.07%
b. PCMark8C -02.57% -01.25% -2.32%
c. PCMark8A -01.15% -01.01% -2.18%

2. Gaming
d. IS--- +00.01% +00.53% -0.26%
e. CG-- -00.56% -00.29% -1.84%
f. SD-- -01.38% -01.26% -1.12%
g. FS-- -01.17% -00.45% -2.15%

h. UHL -04.65% -01.83% +0.32%
j. UHU -00.07% +00.49% -0.21%

k. UVL -05.52% -01.13% -4.74%
m. UVU -04.95% -04.16% -4.95%

3. Numeric
n. 1. -09.44% -04.77% -3.20%
p. 2. -12.39% -12.23% -4.27%
q. 3. -10.24% -10.21% -3.61%
r. 4. -12.52% -11.23% -4.62%
s. 5. -09.47% -09.53% -1.76%
t. 6. -11.85% -11.66% -3.96%
u. 7. -10.98% -10.92% -3.36%
w. 8. -12.13% -12.07% -4.67%

All of these benchmarks are run as the ONLY task on the computer, so the benefits of extra memory for additional processes is not tested; that is for another time.

 

[DonkeyOatie]

This is the end of the first data set. I will be changing to the other G3258 CPU to see what difference there is. I hope it overclocks to more than 4.2Ghz. I will be fitting it with the RAIJINTEK Pallas cooler and I'm collecting temperature information at various clock speeds while running Prime95 at the moment, so we can see how much cooling the Pallas provides and whether this will allow higher speed.

 

[Frozen Fractal]

hmmm, waiting for that eagerly.

 

[Outlander_04]

4.2 Ghz . Only a bit slower than if you had just got an i3 .

How much did you spend on cooling?

 

[DonkeyOatie]

Nothing. Running on stock. The fastest modern i3 runs at 3.7Ghz, that's still 500Mhz slower than this G3258.

 

[DonkeyOatie]

The RAIJINTEK Pallas is now fitted and the second G3258 is in. I have it clocked up to 4.4Ghz at the same voltages as the other CPU reached 4.2Ghz, and the temperature is 79C under Prime95. I hope to go further.

 

[Frozen Fractal]

huh, looks like the Pallas is doing its job well. Keep it up!

Oh and are you gonna rebench with the new clocks?

 

[DonkeyOatie]

I have rebenched at 3.2, 3.5, 4.0, and 4.2 Ghz, but the results are almost the same as the first chip. I have benched 4.4Ghz and I am testing the stability of 4.5Ghz at the moment, although I have had to speed up the memory to 2400Mhz, to get more than 2 hrs of Prime95. However the temperature is up to 82C, which is more than my target. I'll try for better in the weekend. With 24hr Prime95 runs, fine-tuning is a pain and takes a long time.

 

[Frozen Fractal]

I have rebenched at 3.2, 3.5, 4.0, and 4.2 Ghz, but the results are almost the same as the first chip. I have benched 4.4Ghz and I am testing the stability of 4.5Ghz at the moment, although I have had to speed up the memory to 2400Mhz, to get more than 2 hrs of Prime95. However the temperature is up to 82C, which is more than my target. I'll try for better in the weekend. With 24hr Prime95 runs, fine-tuning is a pain and takes a long time.

Hmm, I get it mate. There's so much variable to keep constant in fine tuning. Tried doing it myself once, and kicked out the whole idea after I got frustrated with all that.

 

[DonkeyOatie]

This is an extension of the earlier data using a second G3258 which proved to be able to overclock to 4.5Ghz while keeping the temperature under Prime85 2.66 in the area of 80C. All clock speeds were re-benchmarked, but there was no significant difference between the two CPUs, so only 4.4Ghz, 4.5Ghz, and Shoot for the moon II are reported different to the original data. 4.5Ghz required overclocking the memory up to 2400Mhz, so that value is not presented separately.

I'm looking at three use scenarios.

1. General-purpose light productivity, gaming, and editing. PCMark 7 and PCMark 8 Accelerated and Conventional are used to measure this.

2. Heavy gaming, or a least as good as the system will do. 3DMark FS, SD, CG, and IS, and Unigine Heaven and Valley are used to measure this.

3. Heavy numeric processing and computing. Caselab's Euler 3D fluid dynamics simulation is used for this, and tested at 1 to 8 threads to test processor load and operating characteristics. This test has no graphics component.

A. CPU OVERCLOCKING
CPU I: 3.2 Ghz to 4.2 Ghz (as good as I could get stable at reasonable voltages on stock cooler)
Theoretical performance improvement = 4.2/3.2 or +31.25%

1. GP use PCMark 7 = + 11.34%, PCMark 8 C = + 12.28%, PCMark 8 A = + 16.72%

2. Gaming, IS = +8.60%, CG = + 9.74%, SD = +0.99%, FS = +3.33%, UH Low -0.61%, Ultra +0.00%, UV Low +0.26, Ultra +-0.57%

3. Numeric 1 = + 15.26%, 2 = 14.31%, 3 = 13.87%, 4 = 21.64% 5 = 11.29%, 6 = 14.45%, 7 = 13.05% 8 = 14.32%

CPU II: 3.2 Ghz to 4.4 Ghz (as good as I could get stable at reasonable voltages on RAIJINTEK Pallas at 80C or under. 4.5Ghz increased the temp to 83C))
Theoretical performance improvement = 4.4/3.2 or +37.5%

1. GP use PCMark 7 = + 13.09%, PCMark 8 C = + 12.69%, PCMark 8 A = + 17.01%

2. Gaming, IS = +8.70%, CG = + 9.88%, SD = +1.32%, FS = +3.08%, UH Low -0.31%, Ultra +0.00%, UV Low -4.99, Ultra -3.41%

3. Numeric 1 = + 17.82%, 2 = 14.31%, 3 = 13.76%, 4 = 21.15% 5 = 14.52%, 6 = 14.10%, 7 = 13.48% 8 = 13.52%


The G3258 will overclock on most (all?) motherboards and 4.2Ghz is less than many have gotten. This kind of overclocking is cheap, safe-ish and quick.

B. iGPU OVERCLOCKING
Running at 3.2Ghz, but overclocking the iGPU to 1500Mhz. (stock is 1100 Mhz)
Theoretical performance improvement = 36.36%

1. GP use PCMark 7 = + 3.92%, PCMark 8 C = + 2.90%, PCMark 8 A = + 2.19%

2. Gaming, IS = +16.71%, CG = + 18.50%, SD = +28.56%, FS = +30.26%, UH Low + 28.40%, Ultra 33.03%, UV Low +29.06%, Ultra +30.68%

3. Numeric 1 = -0.99%, 2 = -2.20%, 3 = -1.30%, 4 = -1.74% 5 = -0.66%, 6 = -1.53%, 7 = -0.66% 8 = -0.56%


C. Memory OVERCLOCKING
Running CPU at 3.2Ghz and iGPU at 1100Mhz but overclocking memory from 1600Mhz to 2400Mhz.
Theoretical performance improvement = 50%

1. GP use PCMark 7 = + 2.78%, PCMark 8 C = + 2.29%, PCMark 8 A = + 6.30%

2. Gaming, IS = +2.64%, CG = + 4.04%, SD = +2.70%, FS = +5.64%, UH Low + 3.58%, Ultra 1.82%, UV Low +3.40%, Ultra +2.27%

3. Numeric 1 = +10.82%, 2 = +12.90%, 3 =+11.63%, 4 = +19.82% 5 = +11.44%, 6 = +12.82%, 7 = +10.99% 8 = +12.55%

D. SHOOT FOR THE MOON
CPU I: Running the CPU at 4.2 Ghz, the iGPU at 1300Mhz, and the memory at 2400Mhz. Any attempt to increase one, resulted in a drop in something else to reach stability on the stock cooler.

1. GP use PCMark 7 = + 16.42%, PCMark 8 C = + 17.22%, PCMark 8 A = + 22.52%

2. Gaming, IS = +23.01%, CG = + 24.89%, SD = +20.84%, FS = +23.33%, UH Low +20.25%, Ultra +19.09%, UV Low +20.73, Ultra +18.18%

3. Numeric 1 = + 33.08%, 2 = 32.20%, 3 = 31.41%, 4 = 40.13% 5 = 31.63%, 6 = 32.66%, 7 = 30.01% 8 = 32.80%

E. SHOOT FOR THE MOON II
CPU II: Running the CPU at 4.5 Ghz, the iGPU at 1400Mhz, and the memory at 2400Mhz. Any attempt to increase one, resulted in a drop in something else to reach stability on the RAIJINTEK Pallas cooler. More may be possible, but it is taking me too long to look for it with 24hrs for each burn-in test.

1. GP use PCMark 7 = + 18.47%, PCMark 8 C = + 20.88%, PCMark 8 A = + 25.65%

2. Gaming, IS = +31.18%, CG = + 33.64%, SD = +29.55%, FS = +31.54%, UH Low +23.93%, Ultra +27.27%, UV Low +23.62, Ultra +24.43%

3. Numeric 1 = + 35.52%, 2 = 35.42%, 3 = 34.06%, 4 = 44.91% 5 = 30.97%, 6 = 37.83%, 7 = 32.21% 8 = 35.31%


I'm not suggesting that any of these options are particularly sensible. The Firestrike gets to 500 at best, which is a slow slide-show and the Unigine Heaven and Valley at Ultra would be unplayable.

The memory that can be overclocked to 2400Mhz comes at a 25% price increase.

If anyone has suggestions as to get more speed with existing equipment, I'd like to hear about it.

 

[Frozen Fractal]

hmm, there's no difference to say

 

[DonkeyOatie]

hmm, there's no difference to say

I don't understand. The second CPU performs better than the first at the higher overclock, There difference is small in some areas, but still significant in GP use and Numerical, and the Shoot for the Moon is better everywhere.

 

[Frozen Fractal]

The difference between two clock settings in the CPU OVERCLOCKING isn't much significant - just 2-3% difference for 4.65% OC. You might say that the numbers per se doesn't look significant. On the other side of the coin, the results are quite close to theoretical gain - and that's good imo.

But one thing to say, don't you think iGP OC for the second CPU with Pallas should be included as a separate chart?

 

[DonkeyOatie]

No. Shoot for the Moon is the best possible with the hardware I'm using. The benches for the overclocked iGPU at 3.2Ghz CPU and 1500Mhz iGPU are the same as the first chip and anything better requires 2400Mhz memory which means I cannot properly compare it. All the Pallas does is reduce the temperature and allow me to get a higher overclock of the CPU.