60 Minutes – CPU performance and Energy consumption in Gaming. [AtenRa]

[AtenRa]

Reviews take a lot of time testing the hardware, creating the graphs, writing the review, upload the script and graphs and then publish it. So when reviewing multiple hardware, you simple have to compromise in a lot of things in order to achieve the time targets you have to finish the Review and publish it on the internet or the magazines.

The biggest compromises reviewers are making are in the length of game-play testing (30-60 seconds), the second compromise is the use of pre-scripted in-game Benchmarks. The first one is made for time consuming reasons as its takes a lot of time benchmarking 10-20 different hardware combinations for more than 1-2 minutes each and the second for consistency and time consuming purposes.

But the data gathered by those two methods are often not enough. Benchmarking a game in one area of the game for 30-60 seconds may not give you the same performance as in a different level of the same game after 20-30 minutes of game-play. And pre-scripted in-game benchmarks often are found to be not consistent with actual gaming performance.
One more thing that a lot of reviews lack is the frame latency (Time-Frame) evaluation.

This has been neglected as most reviews are focused on the High-End hardware. But it has become more important for the low-End CPUs as latest games can utilize more than two threads.
So this review is the real thing, one hour of real gaming evaluation of CPU/platform and frame time performance along with total system energy consumption.

This will be an ongoing review, more CPUs and platforms are going to be added along the way.


Testing methodology

The purpose of the review is a full evaluation of CPU performance in today’s games under Windows 10 and system energy consumption during gaming. Evaluation of the CPU importance in Gaming performance (Frames Per Second) and frame latency (Frame Time). Also, total system energy consumption evaluation for the duration of each gaming session.

Playing for one hour straight each game is very time consuming, so a selection of only five games were used for this review that could also have consistency between each run as much as possible.
Different Image Quality settings were used depending of the game and the Game style. For example, Turned-Based games like Civilization can use the highest Image Quality settings and are perfectly playable at 30fps when First Person Shooters like Battlefield 4 in MultiPlayer mode need more than 60fps. Also, AMDs Mantle was used where possible.

FRAPS was used for DX-11 Gaming performance measurements and in-game console command “PerfOverlay.FrameFileLogEnable 0” was used to measure Mantle performance.

Both tools create an excel file, then FLA Calculator was used to open the excel files and create the fps and FrameTime graphs.
Total system energy consumption was measured at the wall by using a Kill A Watt.

Hardware and software

Socket 1155
CPU: Intel SandyBridge Core i5 2500K
Motherboard : Gigabyte GA-Z68A-D3H-B3
Memory : 2x 4GB 2133MHz Kingston Genesis (1600MHz at 9-9-9 1.5V)
GPU : ASUS HD7950-DC2T-3GD5-V2 (1GHz core, 1500MHz Memory, +20% power control)
HDD : Seagate Barracuda ST1000DM003 1TB SATA-6
PSU : Be-Quiet 1000W 80plus


-----------------------------------------


Socket 1151
CPU: Intel Skylake Core i3 6300
Motherboard : ASUS H110M-K D3
Memory : 2x 4GB 2133MHz Kingston Genesis (1866MHz at 9-9-9 1.5V)
GPU : ASUS HD7950-DC2T-3GD5-V2 (1GHz core, 1500MHz Memory, +20% power control)
HDD : Seagate Barracuda ST1000DM003 1TB SATA-6
PSU : Be-Quiet 1000W 80plus


-----------------------------------------


Socket AM3+
CPU: AMD Bulldozer FX 8150
Motherboard : ASUS Crosshair V Formula
Memory : 2x 4GB 2133MHz Kingston Genesis (1866MHz at 10-11-10 1.5V)
GPU : ASUS HD7950-DC2T-3GD5-V2 (1GHz core, 1500MHz Memory, +20% power control)
HDD : Seagate Barracuda ST1000DM003 1TB SATA-6
PSU : Be-Quiet 1000W 80plus



Softawere
Windows 10 Pro 64bit
GPU Driver : Catalyst 15.11Beta
FRAPS
FLA Calculator

Games.
Battlefiled 4 - Mantle
Civilization : Beyond Earth – DX11
Dragon Age : Inquisition – Mantle
Formula 1 2015 – DX11
Ryse : Son Of Rome – DX11

OverClocking.

Core i5 2500K was OverClocked to 4.3GHz, that is 1GHz higher than its base clock of 3.3GHz. CPU voltage was left to auto, memory was raised to 2133MHz with 11-12-11 timings and 1.65 voltage.

FX8150 was OverClocked to 4.56GHz, that is 1GHz higher than its base clock of 3.6GHz. CPU voltage was raised to 1.465V, Buss Speed was raised to 240MHz and Multiplier to 19. Memory was raised to 1920MHz with 10-11-10 timings and 1.65 voltage. NB frequency was raised to 2400MHz and HT to 2160MHz.

 

[AtenRa]

Battlefield 4

Operation Locker 64 Multi Player map was used with 3200 tickets and 60hz. 3200 tickets insured 60 minutes of continuous gameplay. Also the server is very popular and always full with 64 players, making it perfect for our review parameters. We enabled Mantle and 120fps cap.
Graphics Settings used can be found in the link bellow.

BF4 Graphics Settings

The benchmark run started at the beginning of each new game session of the server with each team at 3200 tickets. The console command “PerfOverlay.FrameFileLogEnable 1” was used to start the measurement. At the same time a measurement of the Kwh was noted at the Kill A Watt on the wall. A countdown of 60 minutes was started the moment we started each game run. We played until the countdown reach 0.0 minutes and the alarm started. Then we stopped the benchmark run with the console command “PerfOverlay.FrameFileLogEnable 0” and took a note of the Kill A Watt again.

Core i3 6300 - The lack of CPU Cores/Threads on the Core i3 6300 is evident even when Mantle is used. There is a lot of stuttering and fps performance is lower than even the Bulldozer FX-8150. The Core i3 6300 is not a good CPU if you have a 120Hz monitor and fast GPU since this game in Multiplayer mode needs a lot of Cores/Threads.

Core i5 2500K – The fps performance for the Quad Core Sandybridge is very good, the Core i5 2500K stuttering is good enough and it can maintain high fps close to 120fps cap.

FX8150 - At default clocks the FX8150 has even less stuttering than the Core i5 2500K but the fps performance is lower than the Quad Core Sandybridge. Mantle helps a lot but the low single thread performance have an impact in the fps performance. Although it cannot reach the same fps performance of the Core i5 2500K, the low stuttering makes it better than the latest Core i3 Skylake.


Total system Energy consumption at default clocks is not that much higher for the 4 year old 32nm Sandybridge Core i5 compared to 14nm Skylake Core i3 6300. For 50 Wh or 25% more, the Core i5 gives you much better performance and smoother gameplay. FX-8150 on the other hand has way higher power consumption than both Intel CPUs as its platform is of an older design with more Chipsets, the ASUS AM3+ board was designed for high Overclocks and the CPU itself is more power hungry.

 

[AtenRa]

Battlefield 4 - Overclocked


When both CPUs are OverClocked, the Core i5 2500K seems to have a little bit of an advantage with less variability than the FX-8150. Also the Core i5 Sandybridge fps performance is better as it can maintain higher numbers closer to 120fps cap. But both CPUs are performing excellent in BF4 when Overclocked.
Total system Energy consumption when Overclocked only raised by 20 Wh more than in default clocks on the SandyBridge Core i5 2500K to 270Wh. But it skyrockets on the FX-8150 to a total of 430Wh for the entire system increased by 90Wh or 26% higher than what we had at default clocks.

 

[AtenRa]

Civilization : Beyond Earth

We used a massive 8 player map saved game at 163 turns. Then we started the benchmark run from turn 164 onwards for 60 minutes of continuously gameplay, measuring the performance with FRAPS and system power consumption with a Kill A Watt on the wall. We only used DX-11 because we cannot record the performance of a real gameplay in Mantle mode.
Graphics Settings used can be found in the link bellow.

Civ BE Graphics Settings

Core i3 6300 - Without Mantle, single thread performance in Civilization BE on the DX-11 mode is king and Core i3 6300 is the better processor here with low stuttering.

Core i5 2500K – Although the Quad Core Sandybridge has the performance lead, it’s a little bit less smooth than the Core i3 6300. But for its age the performance is exceptional.

FX-8150 - Im sure that using an NVIDIA card in this game at DX-11 mode will get better results for the Core i5 and FX-8150 than what they get with the AMD HD7950. Since this is a Mantle game the AMD CPU DX-11 performance and especially CPU scaling is not the best the hardware can do. But since this is a turned based game you really don’t have to worry that much about it since the performance of all three CPUs is adequate. But if you have an AMD GCN graphics Card you should play in Mantle mode since it will make the game play smoother for all the CPUs.

Total system Energy consumption at default clocks and the difference between the two Intel Systems is only 10Wh apart. The FX 8150 total system consumption is 80Wh more than the Core i5 2500K.

 

[AtenRa]

Civilization : Beyond Earth - Overclocked

Overclocking helps both the FX-8150 and Core i5 2500K but the Core i3 6300 seems to perform better still in the stuttering department.

 

[AtenRa]

Dragon Age : Inquisition

We began our 60 minutes game play in the Hinterlands, passing through the CrossRoads Village once or twice. We used exactly the same methodology as described in Battlefield 4 above, since both games use the same FrostBite 3.0 engine and we also enabled Mantle.
Graphics Settings used can be found in the link bellow.

Dragon Age Graphics Settings
Advance graphics settings

Core i3 6300 – Although not that bad the Skylake Core i3 6300 does have high stuttering moments during the 60 minutes gameplay. That makes the gameplay a little bumpy and a lot of times it feels sluggish. Fps performance is fine and on par with the other CPUs.

Core i5 2500K - At default Clocks the Core i5 2500K performance is exceptional with very low frametimes with smooth gameplay. Fps performance on par with the rest of the CPUs.

FX 8150 – The FX 8150 frametimes are not that bad, better than the Skylake Core i3 but worse than the Quad Core Sandybridge. Fps performance on par with the rest of the CPUs.

Total system Energy consumption at default clocks department and again we see that the actual power consumption of the 4 year old Quad Core Sandybridge is not that much higher than the 14nm Core i3 6300 skylake. The difference between those two Intel CPUs is only 30Wh. The FX 8150 consumption is 100Wh more than the Core i5 2500K.

 

[AtenRa]

Dragon Age : Inquisition - Overclocked

Overclocking doesn’t affect the average fps but it does have a positive effect by lowering frame times and thus having a smoother game play. The difference in frametimes between the two Overclocked CPUs compared to the Core i3 6300 is substantial making them both having way smoother gameplay than the 14nm Dual Core Skylake.

 

[AtenRa]

Formula 1 2015
We used the Belgian Spa Francorchamps Circuit with maximum graphics settings as shown in the link bellow, also weather conditions were set to light rain.

Our benchmark run begun with all cars stopped on the grid ready for the start of the race. Our car was 3rd on the grid and when the race began we left everyone to pass, then we started the race on the last position on the grid. At the same time we started Fraps and we also took a note of the Wh on the Kill A Watt at the wall.
We played the game for 60 minutes overtaking the cars in front of us and making one pit stop. Then when the 60 minutes countdown reach zero, we stopped fraps and took another note at the Wh at the Kill A Watt at the wall in order to calculate the Wh consumption of each system.

Formula 1 2015 Graphics Settings
Advanced Graphics Settings

Core i3 6300 – The dual Core Skylake does extremely well in Formula 11 2015. Frametimes are low and fps performance the best between the three CPUs.

Core i5 2500K – No problems here either, Quad Core Sandybridge is more than enough for this game even at default clocks.

FX 8150 – The situation with the AMD 8x Thread Bulldozer is a little strange. It does have the lowest stuttering among the three CPUs but, its fps performance has a lot of variance with deeps occurring once on every round.

Total system Energy consumption for both the Intel CPUs is extremely close with Skylake Core i3 6300 using 10Wh less than the old Quad Core Sandybridge. This is another case that the smaller node process has almost zero affect in the total system energy consumption in Gaming workloads. Bulldozer again is using 100Wh more than the Core i5 2500K.

 

[AtenRa]

Formula 1 2015 - Overclocked


Overclocking on the Core i5 2500K only manages to lower the stuttering but fps performance keeps the same. On the other hand the FX 8150 at 4.6GHz has a little worse stuttering compared to default clocks but the fps performance is better and it can almost stay above 60fps the entire benchmark run, outside of a few deeps.

 

[AtenRa]

Ryse : Son Of Rome

For Ryse we used the Campaign Timeline from the Beginning through the S.P.Q.R and to the Trial by Fire until we unleashed the chains of the port tower. We left every video of the story play as if we were playing the game for the first time watching every video. Graphics Settings used can be seen in the link bellow.

Ryse : Son Of Rome graphics Settings
Advanced graphics Settings

We begun our benchmark run from the time Emperor Nero runs inside his palace, when Rome is being attacked by barbarians. At that time, we started Fraps and took a note of the Wh at the Kill A Watt connected on the wall. We played the Timeline until the 60 minutes countdown bell started ringing and then we stopped Fraps and took another note of the Wh at the Kill A Watt.

Core i3 6300 – There is a lot of stuttering with the dual Core Skylake although fps is on par with the rest of the CPUs. Even with graphics settings that make the game extremely GPU bound, the lack or real cores make the Core i3 6300 to stutter heavily making the game very choppy at times. Upgrading to a faster graphics card will only make things worse.

Core i5 2500K – The 4 generation old Core i5 Sandybridge doesn’t stutter that much and it is much better than the dual Core Skylake. The game does feel a little bit choppy at times but nothing like the Core i3. Overall i would say between the three CPUs its the better CPU for the job but what will happen if you upgrade to a faster GPU??

FX 8150 – Although the game feels better with the 8x Threaded Bulldozer FX 8150 than the Core i3, still it’s not as smooth as we would want. The game needs more than 4 threads but it also need high Single Thread performance and that have an impact in gameplay.

Total System Energy consumption is extremely low on the Core i3 6300 with only 170Wh. The Core i5 2500K uses 30Wh more than the Core i3 at 200Wh and then the FX 8150 uses 110Wh more than the Core i5. For the performance of the Core i5 2500K the energy consumption is very good and for once again not that much higher than the 14nm dual Core Skylake. For 2015-2016 gaming the performance of the FX-8150 is fine but the energy consumption of the AM3+ platform when paired with the first generation Bulldozer CPUs is off the scale and extremely high for todays standards.

 

[AtenRa]

Ryse : Son Of Rome - Overclocked

Overclocking the Core i5 2500K to 4.3GHz started to have a nice affect in stuttering until we reached close to the middle of our benchmark run. Then it begun to stutter heavily until the end of the second timeline part of S.P.Q.R. Then in the third timeline part it returned to lower frame times. I have benchmarked this part 3-4 times and it always exhibited the same behaviour with high frametimes and heavy stuttering. So at the end im thinking that the Overclocked CPU may need more fine tuning, perhaps higher vCore but since i had to give the CPU back i didn’t have more time to test it. So i will try to revaluate Overclocked Core i5 2500K again in another time.

As for the FX-8150, things became better with lower frametimes and less stuttering, but i really feel that the game, at those graphical settings, needs a faster GPU than a CPU. At the end, it is very informative to see that even in GPU bound scenarios like this one in RYSE : Son Of Rome, a faster CPU does play a role for a better gameplay without even effecting the fps performance.

 

[AtenRa]

11
Reserved for future use

 

[AtenRa]

12
Reserved for future use

 

[AtenRa]

Ok thats it for now, hope you like this kind of reviewing. Ill try to test more CPUs as soon as possible so stay tuned

 

[dark zero]

Nice thread. Please, add some APUs and if you try, can OC the i3 to 25% more of their values?

Thanks.

 

[maddie]

Very good work. Where do you get the time?

 

[thilanliyan]

Thanks for all your work.

Wow at the power consumption on the FX-8150!! Although it really isn't a surprise...

 

[frozentundra123456]

Interesting data. Unfortunately, 4 of the 5 games you tested are mostly *GPU* limited with the 7950. Not sure that is the best test of cpu performance.

 

[SlowSpyder]

Interesting data. Unfortunately, 4 of the 5 games you tested are mostly *GPU* limited with the 7950. Not sure that is the best test of cpu performance.

But that's how I game, and I think most people do in real life. Right now I'm playing Grim Dawn, it's from a small developer and only uses one core from what I can tell. With my FX @ 4.9GHz my GPU is still almost always at 99% use on a 7970 @ 1050MHz @ 19x12. Sure, some games that's not always the case, but in my experience by far more often than not that's the situation.

Thanks AtenRa, interesting results. Around here I pay ~$0.17 a kilowatt hour, I think. That's a few pennies an hour of gaming in price difference for me.

 

[ozzy702]

But that's how I game, and I think most people do in real life. Right now I'm playing Grim Dawn, it's from a small developer and only uses one core from what I can tell. With my FX @ 4.9GHz my GPU is still almost always at 99% use on a 7970 @ 1050MHz @ 19x12. Sure, some games that's not always the case, but in my experience by far more often than not that's the situation.

Thanks AtenRa, interesting results. Around here I pay ~$0.17 a killowatt hour, I think. That's a few pennies an hour of gaming in price difference for me.

Agreed, I think most people are GPU limited so this is a good real world test/review.

I use a 650watt Platinum PSU on a 4.3ghz 3770k / GTX780 system with a ton of drives and my killowatt shows idle in the 85-90 watt level from the wall and when gaming in the 200-250 watt range most of the time, so not bad at all. My wife and kids use a Haswell i3 cheap box upstairs connected to the 55" LG and that box only sips about 30 watts with normal use. Honestly, getting the wife and kids to turn off lights and fans they aren't using is about the only way for us to reduce our electricity bill without seriously impacting our lifestyle.

We pay $0.09 a killowatt hour in the spring/fall/winter and a variable from $0.09 killowatt to $0.14 during the summer months. Every time I see electricity prices thrown out there it reminds me how much system efficiency really matters in other parts of the nation and especially world wide where costs are usually significantly higher.

AtenRa, thanks for all the effort.

 

[frozentundra123456]

But that's how I game, and I think most people do in real life. Right now I'm playing Grim Dawn, it's from a small developer and only uses one core from what I can tell. With my FX @ 4.9GHz my GPU is still almost always at 99% use on a 7970 @ 1050MHz @ 19x12. Sure, some games that's not always the case, but in my experience by far more often than not that's the situation.

Thanks AtenRa, interesting results. Around here I pay ~$0.17 a killowatt hour, I think. That's a few pennies an hour of gaming in price difference for me.

It is useful but incomplete data. It is like testing a corolla vs a corvette with S rated tires. They will have the same top speed, but it is limited by the tires (gpu) rather than the car (cpu). I am not saying the data is invalid or not useful, just that it does not tell the entire story.

 

[SlowSpyder]

But that's how I game, and I think most people do in real life. Right now I'm playing Grim Dawn, it's from a small developer and only uses one core from what I can tell. With my FX @ 4.9GHz my GPU is still almost always at 99% use on a 7970 @ 1050MHz @ 19x12. Sure, some games that's not always the case, but in my experience by far more often than not that's the situation.

Thanks AtenRa, interesting results. Around here I pay ~$0.17 a kilowatt hour, I think. That's a few pennies an hour of gaming in price difference for me.

Whoops, I pay $0.13111 a kilowatt hour. Even if I gamed 30 hours a month, I probably lose more spare change in that time than the difference in running-cost between my FX system vs. a more power efficient option.

 

[SlowSpyder]

It is useful but incomplete data. It is like testing a corolla vs a corvette with S rated tires. They will have the same top speed, but it is limited by the tires (gpu) rather than the car (cpu). I am not saying the data is invalid or not useful, just that it does not tell the entire story.

That analogy would only be valid if people drove their corvettes in real life with S rated tires for the most part. AtenRa's testing represents how I think most gamers actually use their PC. It isn't isolating just the CPU if it's GPU limited, I get that, you are right. But I think most people actually game at mostly GPU limited settings.

 

[MrTeal]

Whoops, I pay $0.13111 a kilowatt hour. Even if I gamed 30 hours a month, I probably lose more spare change in that time than the difference in running-cost between my FX system vs. a more power efficient option.

A 150W delta for an hour a day would net around 55kWh a year, so $5-$15 depending on your power costs. Not exactly breaking the bank. The difference in idle power would make a larger difference, though that could be mitigated (assuming a home gaming computer that's not a primary work PC) by just sleeping it when you're at work/school and sleeping.

 

[frozentundra123456]

Yes, the extra power consumption of the FX is not a major expense in the grand scheme of things for most people. However, it does not make sense to me to use more power for equal (at best) or (usually) worse performance.

That extra expense is mitigated of course, by the cheaper initial cost of the FX, although when they first came out FX and i5 were close in price, if I recall correctly. But personally I just value efficiency, and using more power for less performance just doesnt feel right to me.