-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
Calculating memory speed CAS vs MHz.
- Thread starter don256us
- Start date
There are none. Even PCMark 8 which should base on popular applications is really bad to compare performance. It reacts better on some generations of hardware and memory performance is barely affecting any scores. Actually storage results are weird in PCM8.
- Cinebench and similar rendering benchmarks are showing some improvements on faster memory but results are like +/-20 points when total score is 1000+.
- SuperPi/HyperPi 32M is good to compare memory performance but hard to call it real world benchmark.
- Realbench is sometimes acting weird and you can get +/- 20% higher or lower results after couple of runs.
- AIDA64/Maxxmem are showing only bandwidth and latency what isn't showing real differences in daily usage and 10GB/s higher bandwidth can give you 1% higher performance in other applications.
- 3DMark physics tests are showing differences between memory kits but it's still not big difference and hard to say if it's worth to buy faster memory based on these results.
- Geekbench has many memory tests and it's free in 32b version. It's also more like bandwidth benchmark.
Simply you have to test it on applications which you are using as it won't give you anything if you are comparing performance on benchmark or game or anything else that you won't ever use for normal work. Memory can give 10% higher performance in some applications ( mainly server ) but do you really care if you won't see it on your applications ?
If you are playing games then compare it in games. Some of them have built-in benchmarks so it's easier to compare.
You can also compare memory performance based on integrated system tool. Simply type 'winsat mem' in command prompt and it should test system memory performance. It's how windows sees memory speed and results are not always the same as in benchmarks.
Personally I see slightly better reaction of my PC when I set higher memory frequency/lower timings but it's nothing that I really need. In most games I can't see any difference if I'm playing on i7 [email protected]+DDR3-2400 or on [email protected]+DDR4-4000. However I play mainly at 1080p and I just don't care if there is 50 or 80 FPS. I can't see the difference anyway.
- Cinebench and similar rendering benchmarks are showing some improvements on faster memory but results are like +/-20 points when total score is 1000+.
- SuperPi/HyperPi 32M is good to compare memory performance but hard to call it real world benchmark.
- Realbench is sometimes acting weird and you can get +/- 20% higher or lower results after couple of runs.
- AIDA64/Maxxmem are showing only bandwidth and latency what isn't showing real differences in daily usage and 10GB/s higher bandwidth can give you 1% higher performance in other applications.
- 3DMark physics tests are showing differences between memory kits but it's still not big difference and hard to say if it's worth to buy faster memory based on these results.
- Geekbench has many memory tests and it's free in 32b version. It's also more like bandwidth benchmark.
Simply you have to test it on applications which you are using as it won't give you anything if you are comparing performance on benchmark or game or anything else that you won't ever use for normal work. Memory can give 10% higher performance in some applications ( mainly server ) but do you really care if you won't see it on your applications ?
If you are playing games then compare it in games. Some of them have built-in benchmarks so it's easier to compare.
You can also compare memory performance based on integrated system tool. Simply type 'winsat mem' in command prompt and it should test system memory performance. It's how windows sees memory speed and results are not always the same as in benchmarks.
Personally I see slightly better reaction of my PC when I set higher memory frequency/lower timings but it's nothing that I really need. In most games I can't see any difference if I'm playing on i7 [email protected]+DDR3-2400 or on [email protected]+DDR4-4000. However I play mainly at 1080p and I just don't care if there is 50 or 80 FPS. I can't see the difference anyway.
OP
- Joined
- Jul 17, 2003
- Thread Starter
- #23
I'll post some screen shots but I used Passmark to test because it didn't charge me to show all of the results. Here is the long of the short of it:
The only real difference was between 8 GB and 16 GB. I attribute this to the dual channel configuration allowing for increased bandwidth.
The difference between the CAS 9 and CAS 11 was actually inverse for the one iteration of the test that I ran. Statisically speaking, they were probably within the same margin of error.
Going from single to dual channel is where I think my increase in performance came.
The only real difference was between 8 GB and 16 GB. I attribute this to the dual channel configuration allowing for increased bandwidth.
The difference between the CAS 9 and CAS 11 was actually inverse for the one iteration of the test that I ran. Statisically speaking, they were probably within the same margin of error.
Going from single to dual channel is where I think my increase in performance came.
- Joined
- Jan 10, 2001
Super Pi is great for testing speed differences. I keep multiple instances of it in different folders so that it retains the scores from different sticks, platforms or whatever. With the same ram sticks, going from AMD to Intel the speeds were way faster with the Intel then the AMD platform. So there are differences even with the same ram but it still has a limit to how fast it does anything without regard to basic timings. Back in the early DDR days I did extensive testing with ram timings and found that 7-3-3-2.5 was the fastest speed of ram when the rage was expensive 7-3-3-2 sticks that cost more but did nothing to increase the actual time it took to do memory calculations. I had two different sets of 3200 that would not run together in my system despite both sets having the same IC's. I got them to work by reading the subtimings with Typhoon burner and fount that one set had no subtimings entered at all and depended on the motherboard to set them. I simply flashed the timings from one set to the other and then they would run together.
The best thing about Suer Pi is that it just times the same iterations and leaves the best time on the page so you don't have to record the time differences between different timings or sticks and you can keep multiple instances open side by side and see the time difference it takes to complete the work. Tip, each instance must have it's own sub folder because it keeps the record in there with it.
The best thing about Suer Pi is that it just times the same iterations and leaves the best time on the page so you don't have to record the time differences between different timings or sticks and you can keep multiple instances open side by side and see the time difference it takes to complete the work. Tip, each instance must have it's own sub folder because it keeps the record in there with it.
