Jump to content


Photo
- - - - -

What Is The Multiplier Actually Doing?


  • Please log in to reply
15 replies to this topic

#13 oralpain

oralpain

    Geek

  • Members
  • PipPipPip
  • 870 posts

Posted 18 November 2005 - 05:23 PM

Alown, that is not how it works. The multiplier does not divide anything into frames.

However, I can't help but think that it's almost pointless to have a high multiplier.  The only thing it helps with is calculations that can be stored in the CPUs cache, no?


It's far from pointless to have a higher multiplier. Current (super scalar, pipelined, out of order) CPUs could not possibly process all the data they would beable to recive a 1:1 (cpu:bus) memory and/or system bus, in the vast majority of situations. If you double a modern CPU's data buses and halve the multiplier, you will probably see a nice performance increase. Howerver, if you go further and tripple it while cutting multiplier three fold you would see much less of a boost. Efficency drops at each stage. You will eventually hit a wall where the CPU just cannot process data anyfaster, so will stop asking for it till it needs more.

This is why, with my new S939 dual channel setup, I get only about 50-60% more memory bandwidth than I did with the same memory and settings in a single channel config. Max theoretical bandwidth did double, but my CPU is not fast enough to process all the information it recieves in many cases. I am not normally limited by memory bandwidth even with a CPU clock 12 times as fast as the memory clock. I actually have more *usable* memory bandwidth running the memory at 194MHz with the CPU at 2.71GHZ than I do with the memory running at 250MHz and the CPU running at 1.5GHz, for example.

CPUs do alot of stuff with a given instruction and peice of data. They do not need massive relative bandwidth to reach their limits. It takes alot longer than 1 clock cycle to process most anything.

http://arstechnica.c...t-2/cpu2-1.html

When I raised my multiplier, my arithmatic benchmarks did major jumps but things like 3Dmark only had a few extra marks.  hm.


3dmark tends to be video card limited.

This is where the term bottleneck comes in. If one component is faster at getting a given task done than other components, you have a bottleneck. It's going to vary from program to program and task to task. When recording video my hardrive is the bottleneck. When encoding video, my CPU is the bottleneck. When playing BF2, my video card is the bottleneck.

#14 alown

alown

    Still an ATI fanboy!

  • Members
  • PipPipPip
  • 856 posts
  • Gender:Male
  • Location:WI

Posted 20 November 2005 - 01:19 AM

well, isn't 3dmark 01 and 03 have cpu speed affect them, but 05 is mostly video card.
AMD Thuban X6 1055T @ 4.0Ghz
Corsair H70 & Cool Master HAF 932 (painted)
8GB of PC12800 OCZ Reaper @ 1900Mhz
ASUS Crosshair IV
2 WD Blacks in 0 RAID
AIT 6870 @ 960 gpu 1140 mem
Cooler Master GX 750W

#15 skarphace

skarphace

    New Member

  • Members
  • 23 posts
  • Location:PA, USA

Posted 23 November 2005 - 06:56 AM

This is where the term bottleneck comes in. If one component is faster at getting a given task done than other components, you have a bottleneck. It's going to vary from program to program and task to task. When recording video my hardrive is the bottleneck. When encoding video, my CPU is the bottleneck. When playing BF2, my video card is the bottleneck.

View Post

That's the problem I was hitting when I first got my X850Pro, my core system was the bottleneck, not the video card. Which is one of the main reasons I started getting into OCing. So I was expecting a larger jump in performance. However, when I ran PC mark, my CPU scores weren't too much better either, really only Sandras Dhry and Whet scores went way up. In fact, I beat the 64 bit that's in the house. Needless to say I was jumping for joy, heh.

Thanks for all the great info, it's been very helpful.
128 Processor servers. Prepare to be boarded!
-------------------------------
Athlon Xp 2500+ Barton(2.36Ghz - 12.5 x 189Mhz @1.7v)
--AeroCool HT101 heat pipe/sink w/80mm
1GB PC2700 (512+256+256(fried))
ASUS A7V880(Via KT880)
80GB EIDE - 8MB cache - 7200rpm
120GB EIDE - 8MB cache - 7200rpm
Radeon X850Pro 256MB GDDR3
SB16 4 channel Audio

#16 oralpain

oralpain

    Geek

  • Members
  • PipPipPip
  • 870 posts

Posted 26 November 2005 - 01:32 AM

well,  isn't 3dmark 01 and 03  have cpu speed affect them,  but 05 is mostly video card.

View Post


Generally yes, but it depends on system.

Even 3dmark 2001 is video card bottlenecked on a TNT2 with any recent CPU.

Likewise 3dmark 2005, is CPU bottlenecked on a 3.9Ghz FX-57, and SLI 7800GTX 512MBs cooled with phase change and overclocked to 800+Mhz core.

However, you are right, for most reasonable systems, 3dmark 2001 is mostly CPU/memory limited, and 2005 is mostly video card limited, with 3dmark 2003 being in between.

Thanks for all the great info, it's been very helpful.


Any time. Just don't take my word as gospel. Do some research for yourself.