• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

How important is CPUNB speed?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

trents

Senior Member
Joined
Dec 27, 2008
How important is the CPUNB speed to performance? I know that HT speed is not that important but I assume that CPUNB is more critical since it has to do with the ICM. In overclocking my PI X3 to 3.1 ghz with the clock multiplier I have lowered the CPUNB multiplier to keep it at the stock 1.8 ghz. Is this creating a major bottleneck to performance? I guess I should also mention that I am not overclocking the fsb. It is set at 200 mhz.
 

mxthunder

Member
Joined
Dec 27, 2004
Location
Northeast Ohio
Not sure how much difference it makes on the P I's but on my PII it makes a huge difference. Almost as much as the core speed itself. Not only that, but it sometimes helps to stabilize a high OC if you have a higher NB
 
OP
trents

trents

Senior Member
Joined
Dec 27, 2008
Thanks for the input about the higher CPUNB speed helping to stabilize a high OC. I wasn't aware of that.
 

QuietIce

Disabled
Joined
May 7, 2006
Location
Anywhere but there
The higher NB helps the Phenom II but I'm not so sure about the Phenom I's. However, NB speed does help a LOT with memory performance and L3, since it's also on the NB ...
 

QuietIce

Disabled
Joined
May 7, 2006
Location
Anywhere but there
L2 is part of the core, as is L1, and runs at core speeds.

From there data gets transferred to L3, which is on the cpuNB, and may or may not get transferred to RAM through the IMC, which is also on the cpuNB. I often think of L3 as a RAM buffer, similar to the data buffer on an HDD. Thier functions are very similar ...
 

Horn_Bee

Registered
Joined
Nov 5, 2009
Not sure how it will translate to a PI, but on my PII I gained ~4-500 points on 3Dmark06 by increasing my CPU NB frequency 400mhz to suit my overclock. There was also a noticeable performance increase in regards to memory performance. Definitely gains to be had there!

I'd be interested to see how the PI is affected by tweaking the CPU NB...
 

exactopposite

Member
Joined
Dec 3, 2005
Not sure how it will translate to a PI, but on my PII I gained ~4-500 points on 3Dmark06 by increasing my CPU NB frequency 400mhz to suit my overclock. There was also a noticeable performance increase in regards to memory performance. Definitely gains to be had there!

I'd be interested to see how the PI is affected by tweaking the CPU NB...

What is your CPU-NB freq on the rig in your sig?
 
OP
trents

trents

Senior Member
Joined
Dec 27, 2008
What is your CPU-NB freq on the rig in your sig?

When I started the string it was at 1800 mhz, which I believe is the stock CPUNB speed if the CPU was running at the stock 2.2 ghz. Over the weekend I raised the CPUNB speed to 2400 mhz but it took a bump in CPUNB v to make it stable.
 

Horn_Bee

Registered
Joined
Nov 5, 2009
What is your CPU-NB freq on the rig in your sig?

I'm running my NB @ 2400. I can't push any higher without encountering stability problems (Due to running 4 sticks of RAM, I believe).

trents, how did you go once you'd raised your NB? Did you see an improvement in performance?
 

bldegle2

Member
Joined
Feb 18, 2002
Location
Floyd, VA
>Horn_Bee<

>>>>>>>>>>>>>>>>>>>>>>>>>

raise the NB volts, i am running 2800NB (will do 3000, haven't tried higher) on my AsRock with a 965BE.......

bumping the NB does more for performance than anything else you can throw at these AMD slugs....worth it to get get it a bit higher...

laterzzzz................
 
OP
trents

trents

Senior Member
Joined
Dec 27, 2008
trents, how did you go once you'd raised your NB? Did you see an improvement in performance?

Haven't tested it yet. I want to see how far I can push it before I do. I'm not a gamer but in causual computing I don't notice much difference. I guess I'll download SiSoft Sandra when I get around to testing. It has a lot of detailed tests for everything, including cache performance.
 
OP
trents

trents

Senior Member
Joined
Dec 27, 2008
Read my article: http://www.overclockers.com/index.p...the-phenom-ii&catid=57:processors&Itemid=4263

Talks about it at a mid-level view. If you want to learn more just ask ;)

Surprising little difference in performance as the NBCPU speed is manipulated, if I understand the chart. Seems almost random, kind of what you would expect when running several benches without changing the settings.

Really serves to reinforce the fact that the only thing that really makes much difference in performance in all of our overclocking tweaks is raising CPU speed. Even ram speed doesn't seem to make much difference unless you are running very memory intensive apps and I don't even know what those would be. Maybe file compression? What else?
 

KTE

Member
Joined
Sep 22, 2008
Read my article: http://www.overclockers.com/index.p...the-phenom-ii&catid=57:processors&Itemid=4263

Talks about it at a mid-level view. If you want to learn more just ask ;)
I've just read through this. Three things I need to ask you for the Geekbench tests >

- What was your RAM setup?
- 32b or 64b OS?
- Did you check how the NB+RAM combo is performing in Everest for each setting?

Not speaking on the article but, majorities are making the same mistakes with Deneb as they did with Agena. Clocking NB high and thinking that means better performance - it doesn't. MRL comes into play nearly exactly like this: http://anandtech.com/mb/showdoc.aspx?i=3671&p=6

As Archer said on TR, "it's a zen thing". Meaning, efficiency is the key. It's easily possible to beat a high clocked NB with a lower clocked one. We're talking about daily setups here.

You can't keep the NB delay fixed as that will stop the NB clocks going higher but you can keep it's overall latency below stock and even decreasing (and hence getting faster) in nanoseconds by thinking about what you're doing instead of just slamming the gas at every go i.e.,

2.0GHz NB -> 51 NB Clks -> 25.5ns
2.2GHz NB -> 53 NB Clks -> 24.1ns
2.4GHz NB -> 55 NB Clks -> 22.9ns
2.6GHz NB -> 58 NB Clks -> 22.3ns
2.8GHz NB -> 60 NB Clks -> 21.4ns
3.0GHz NB -> 62 NB Clks -> 20.7ns

Instead of what we >95% commonly see ->

2.0GHz NB -> 51 NB Clks -> 25.5ns
2.2GHz NB -> 55 NB Clks -> 25.0ns
2.4GHz NB -> 58 NB Clks -> 24.1ns
2.6GHz NB -> 65 NB Clks -> 25.0ns
2.8GHz NB -> 70 NB Clks -> 25.0ns
3.0GHz NB -> 75 NB Clks -> 25.0ns

And the poster proclaims:

"ZOMG NB clock gives no performance increase!!! In fact it goes down as you clock higher!!! So this is my theory..."

Well, what do you expect when your 2.8GHz NB is so loose that it's performing like a 2.2GHz NB? :p

That's synonymous to running RAM at 1066 7-7-7 and expecting it to be faster than RAM at 900 4-4-4. You have to fnd the right balance per your setup (heavily BIOS dependent).
 
OP
trents

trents

Senior Member
Joined
Dec 27, 2008
Have you got another link to that article? All I'm getting are Explorer errors trying to open it ... :(

Works fine in Windows 7 for me in both IE8 and FF 3.57.

You know, it occurs to me that this CPUNB speed thing is like any other component overclocking equation. If it is causing a bottleneck then raising its speed will help performance but if the extra bandwidth is not needed then it will not. From Dolk's table it seems that it begins to make a little difference at the higher cpu clocks but not at the low end, where I am at 3.1 ghz.
 

QuietIce

Disabled
Joined
May 7, 2006
Location
Anywhere but there
It's a known problem overclockers.com has with their indexing and IE7. I need another link or I'll have to go searching for the article. :-/



Edit:
But you mentioned FF - I'll give that a try. I don't usually use it but it's installed for occasional use in file transfers ...
 

Archer0915

"The Expert"
Joined
Nov 3, 2008
are there any benchmarks for CPUNB?

Just look at the differences with different NB speeds.

The L3 is a work area for the CPU and that is why some applications run faster than other with L3/No L3. Instead of using slower ram and swapping constantly from the L2 for work, the L3 is use like ultra fast ram.

L1 instruction cache, makes the most difference.

L2 cache is the work area

L3 cache is the fast storage area for the L2 and allows read through to system ram if data is not there. So by speeding up the NB you allow work to be done faster, if the code is properly written and the cache is large enough.