• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Does more CPU cache give a better performance?

JasonM said:
...
Why is 1333 and 2400 only around 1% difference in most situations? Simple it's because of the cache, and the larger the cache the more chance of cache hits....
Click to expand...
Very good post Jason. I remember a discussion with a prominent overclocker and bencher who was espousing the merits of fast memory and tight timings and how it made the system 'loads faster' and 'more efficient', and when I repeatedly pressed him for some empirical evidence, especially in the realm of video I was only greeted by silence. In truth going from 1600 to 2400 is just a few percent and you've wonderfully explained why. ;)
 
JasonM said:
Other then WinRar (that accessing a lot of memory), the real world difference between 1333 and 2400 DDR 3 is about 1%.

http://www.anandtech.com/show/6372/...-to-ddr32400-on-ivy-bridge-igp-with-gskill/11

Why is 1333 and 2400 only around 1% difference in most situations? Simple it's because of the cache, and the larger the cache the more chance of cache hits.

CPU registers (fastest)
Level 1 cache
Level 2 cache
Level 3 cache
DDR
HDD / SSD (slowest)

When you first switch on your computer all storage is present in HDD / SSD only. However as the computer runs it moves frequent accessed memory closer up the hierarchy of storage. The levels of memory speed above, allow many GB / TB of programs/data to be stored on the HDD/SSD, yet due to the cache levels, there is minimum latency when the CPU executes. The CPU cache controller, and memory management of modern operating system make this possible.

The early PC's used an 8086 CPU that had no cache. Back then memory speed was really important, computers were even advertised with '0 wait state memory' meaning that the CPU did not wait additional clock cycles while memory was accessed. Back then fit the wrong memory speed and you cripple a computer, today you struggle to even notice faster RAM and it's all down to the caching system.
Click to expand...

Your Sir, is a Star.
 
Video encoding is another case where faster RAM helps. Just ran a few tests, the difference between 1600 and 2400 is greater than a 100MHz CPU clock speed bump.

4.6 GHz - 2400-11-13-13-35-1T - AVG FPS 97.43
4.7 GHz - 2400-11-13-13-35-1T - AVG FPS 100.59 +3.14%

4.7 GHz - 2400-11-13-13-35-1T - AVG FPS 100.59
4.7 GHz - 1600-9-9-9-35-1T ----- AVG FPS 96.38 -4.18%

The settings I used for the tests are a middle ground, but the perf gap gets exponentially bigger as settings get more demanding. So these results are closer to worse case and the perf improvement will generally be better than this.

4.18% is nothing to be sniffed at, especially considering a 100MHz CPU clock bump only gave 3.14%. This isn't a kit that tightens up well either so there's even more potential in a kit that runs tight at high MHz.

Video encoding loves high freq and tight timings. The biggest perf gains are in more cpu cores though.

Gains in order of most to least for video encoding would be:
Core count
CPU clock freq (providing the bump was more than 100MHz)
RAM speed (Both Freq and Timmings)

RAM speed might not make much difference in some cases but there are definitely cases where it does.
 
Many thanks for that John. Those differences are too small to be able to make that statement and 1% difference falls within the margin of error. Also you need to run each test say 5 times then take an average. What video program did you use? If you used Handbrake you should test with HT disabled as it can sometimes have a deleterious effect on x264 encoding. Could you please try a few runs of Cinebench and see what you get. Cheers.
 
wam7 said:
Very good post Jason. I remember a discussion with a prominent overclocker and bencher who was espousing the merits of fast memory and tight timings and how it made the system 'loads faster' and 'more efficient', and when I repeatedly pressed him for some empirical evidence, especially in the realm of video I was only greeted by silence. In truth going from 1600 to 2400 is just a few percent and you've wonderfully explained why. ;)
Click to expand...

bigmac121 said:
Your Sir, is a Star.
Click to expand...

Well many thanks :)

My final year degree project was in computer systems architecture simulating 8086 CPU's! Graduated in 1997, first time i've said '0 wait state' in a very long time!
 
Last edited:
John24 said:
Video encoding is another case where faster RAM helps.
Click to expand...

It's a similar situation to the WinRar compression, being the encoding is reading a relatively large amount of different memory addresses. As such the cache hits will be lower then most programs, hence the faster memory is more important here.
 
You must log in or register to reply here.
Back
Top Bottom