Question about CPUs (4 vs 8 core)

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Lorne

Senior member
Feb 5, 2001
873
1
76
There is still memory bandwidth limitations, If there isnt enough to go around the you get CPU starvation.
So it will depend on the benchmark used, Wether it high computation only, High memory bandwidth or both, There is also the fact as insited above that the first LCPU does all the thred handling and takes a hit the more it has to hand out.

The test the OP states could be done with the FX8350 and dissable 4 cores and rerun the several test while enabling a core each time and recording where the hit/lose starts.
Also if using faster memory on the FX8350 helps.
 

Virgorising

Diamond Member
Apr 9, 2013
4,470
0
0
It's more than that. On my rather slow dual core non HT machine, I can do one instance of super pi 1M in 17 seconds. You would think I could run two different instances (different copies from different folders) also in 17 seconds, but it actually takes 21 seconds to run two. Can you guess how long it takes to run 3? 31 seconds. One reason is because they all share the same cache. I'm pretty sure the entire program, all 3 instances all fit in cache, but each process must be getting stuck waiting. For much the same reason, the 8 core cannot run even a perfectly multithreaded benchmark twice as fast vs a 4 core.

Very interesting....also, makes sense.
 

Virgorising

Diamond Member
Apr 9, 2013
4,470
0
0
If anyone hasn't done video rendering and editing then they haven't used the full benefit of a computer because videos is a huge thing in life especially now that most cameras and even cell phones do videos as well. The problem is that video rendering which you absolutely need to do when you create videos takes a tremendous amount of computer time. Video software is capable of dividing up the work between the processors. That means an 8 core will do the rendering twice as fast as a 4 core and maybe even faster if the 8 cores are slightly faster chips. Time in getting things done is very important and so yes it would definitely be worth getting an 8 core for video rendering which I don't see how anyone can get along without doing at some point or another.
Yes, it doesn't make a damn bit of difference for word processing but anyone who uses a computer will at some point want to do a lot more than that like video rendering of personal videos, converting movies to DVD, converting DVD movies to computer movies and so on. These have been very important things to me and I can't wait for next years Haswell 8 core or whatever it is called.


Above confirms my subjective experience in video encoding re my older system, PD 3.40 (Presler) vs this system with the Lynnfield i7 870 chip.
 
Last edited:

Virgorising

Diamond Member
Apr 9, 2013
4,470
0
0
I can't agree.

With a 32-bit OS, you can address at most 2^32 bytes. Total. All programs together. Plus the OS itself (kernel, drivers, buffers, etc). So even though Virtual Memory is 4GB, the actual amount of RAM that can be used is less. Under Windows it's something like 3.2GB. If you have a browser open, another program can use 0.5GB less. Etc, etc.

If you run a 64-bit OS, the situation changes. The OS (kernel, drivers) still needs some RAM. You still lose memory because of overhead (buffers). But each 32-bit process now can have 3.2GB of RAM, without bothering other processes. Have your browser open with lots of tabs, it doesn't matter if it uses a GB of RAM. Run some more programs in the background, that use RAM. Your 32-bit applications can still use 3.2GB each. Even if you run out of RAM, although less likely with a few 32-bit applications and 6GB of RAM, the OS can just use the page-file.

So yes, running a 64-bit OS with 32-bit applications really can make a difference.

There are the metrics on paper, and there there is the reality of subjective experience. None of my 32 bit apps run slower in my current 64-bit platform, which pleased me, but few run faster.
 
Last edited: