Phenom 720 overclock? FSB/HT or Multiplier?

Summer

Junior Member
Feb 27, 2009
23
0
0
Generally speaking, is it better to overclock using FSB/HT or the multiplier (or both) on a Phenom II Black Edition CPU? I've seen people push their HT to well over 2400+ (stating that the Phenom II L3 is where the gains is at) and was wondering what was the best way to get to 3.6ghz while pushing HT on a Phenom 720?

Where do I start on the HT? How much voltage do I need ? Is there a link?

 

richierich1212

Platinum Member
Jul 5, 2002
2,741
360
126
The HT has actually been shown to not make a difference, whether it's at 1.8GHz or 2.4GHz. The NB is another story, however.
 

Summer

Junior Member
Feb 27, 2009
23
0
0
So how do we overclock the NB? Is there a multiplier in the bios? Do we have to worry about voltage settings?
 

Summer

Junior Member
Feb 27, 2009
23
0
0
What is the stock voltage for the NB (790GX/SB750)? Is there a rule of thumb for each x1 (200mhz)?
 

zagood

Diamond Member
Mar 28, 2005
4,102
0
71
Originally posted by: Summer
Thanks richier1212. Guess I'll wait for anandtech's NB article. :D

DON'T GIVE UP.

you've obviously seen the threads, read the reviews, done your due diligence...now play with it and find out for yourself.
 

poohbear

Platinum Member
Mar 11, 2003
2,284
5
81
Originally posted by: richierich1212
The HT has actually been shown to not make a difference, whether it's at 1.8GHz or 2.4GHz. The NB is another story, however.

what do u mean the "NB is another story, however"?

If the memory controller is on the cpu for the PHII, it shouldnt effect performance that much at all if u overclock the NB. How exactly would uoverclocking the NB give a performance boost since the such an option was never available w/ the A64s.
 

richierich1212

Platinum Member
Jul 5, 2002
2,741
360
126
Originally posted by: poohbear
Originally posted by: richierich1212
The HT has actually been shown to not make a difference, whether it's at 1.8GHz or 2.4GHz. The NB is another story, however.

what do u mean the "NB is another story, however"?

If the memory controller is on the cpu for the PHII, it shouldnt effect performance that much at all if u overclock the NB. How exactly would uoverclocking the NB give a performance boost since the such an option was never available w/ the A64s.

Boosting NB frequency has shown to increase memory bandwidth. Try it out on Super-PI.
Check out SlowSpyder's thread on "uncore"
 

Summer

Junior Member
Feb 27, 2009
23
0
0
Bottom

Doesn't appear to be much gain clocking the NB unless you got the data to use it. Guess I'll be going back to stock. :(
 

SlowSpyder

Lifer
Jan 12, 2005
17,305
1,001
126
Originally posted by: Summer
Bottom

Doesn't appear to be much gain clocking the NB unless you got the data to use it. Guess I'll be going back to stock. :(

I don't see any reason not to keep it bumped up if you can. Depending on what you do and how high you push your cores there will be plenty of times where the increased NB frequency will prvide a boost. If you're using this system as an HTPC and power and noise are the major concerns, there may be no reason to add the heat and power usage. But if I was going to game or do anything else that's fairly intense I'd at least take a little time to push the NB.

Of course in my case my NB speed is only 1.8GHz, I believe your chip runs at 2.0GHz. I think once you start pushing the core speed the higher NB speed helps more than if you keep the core speed lowish.

AT tested one game, FC2 with the increased NB speed and saw tangible gains only with 2x4870's. I wouldn't be suprised if that means the next gen video cards would see gains as well (Radeon RV870 or whatever it is and GT300 or whatever they're calling it).

I can't remember off the top of my head, but I believe I'm running a pretty lowish voltage of 1.125 for my meager 2070MHz NB current NB speed. I believe on my chip that was one bump in voltage from a stock voltage of 1.100. But since I'm pretty sure your chip has a different NB speed from the factory, I can't say if your chip may use a higher voltage as well.

Anyway, increasing the NB speed won't make the chip suddenly an i7 slayer, but I'm fairly confident there will be benefits depending on what you do as well. If you're going to be overclocking anyway than I think there is reason to spend some time tweaking the NB as well.
 

DrMrLordX

Lifer
Apr 27, 2000
21,629
10,841
136
Increasing NB speed should reduce L3 cache and system memory latency. That will reduce the amount of time your CPU will spend waiting for read/write operations to complete. How often this matters varies from one application to another, but it is generally better to have low cache/memory latency. It is for this reason that cache was integrated into chips in the first place. This has been discussed elsewhere and is essential for proper performance scaling if you choose to overclock your cores.
 

Summer

Junior Member
Feb 27, 2009
23
0
0
CptMorgan's post

2600 NB versus 2000 NB. Good results so far. CptMorgan stated that there is minute differences at 2000~2400 and recommended a full jump to 2600+. Guess we're looking at a typical NB voltage increase of 1.2 --> 1.375~1.4 to achieve 2600.

How safe is it to run the NB frequency at 1.375+? How will the voltage increase temperature? One would think there would be a lot of data regarding NB overclocking but there isn't but I guess it has to start somewhere. :p
 

DrMrLordX

Lifer
Apr 27, 2000
21,629
10,841
136
It's somewhat new territory because the K8 had a memory controller speed synched to the core speed, so there was no way to really mess with it much; furthermore, Phenom Is were such a dud that only a few AMD experimenters played with them.

In short there's no way to know how much the extra voltage will affect the lifespan of the NB or anything controlled by it (IMC, L3) because we haven't had enough test cases. I'd say stick to the +10% rule unless you'd got rock-solid cooling. In other words, 1.325v should probably be seen as the "safe" limit. 1.4v may be pushing it.

Another thing to take into consideration is how much vdimm you're pushing through your RAM. I don't have any experience with RAM overvolting on PhenomI/II systems but I remember on K8s that if the delta between vcore and vdimm got too high, there was a risk of burning out the memory controller (for example, I sort of stressed the memory controller on my k8 to the point that memory overclocks using anything but 1:! or 2:1 dividers just don't work right). So if you are running a Phenom II and have some heavily-overvolted RAM (2.2v+ on DDR2 or 1.7v+ on DDR3) you may need a similar overvolt on the NB and/or cores to avoid burning out your memory controller. There is a similar situation on Core i7 chips so it would not surprise me if the same situation existed on Phenom IIs.

If you really want to see the performance difference of raising NB speed though, try an app like SuperPi. In my experience, it responds very well to reductions in memory/cache latency.

edit: if you've got a board that supports it, try disabling L3 in BIOS (the option may be labeled as a TLB bug fix or something; on some boards, this "fix" got around the old Phenom I TLB bug by disabling the L3 altogether!) and see if that affects your maximum NB speed at any given voltage versus normal operation.
 

SlowSpyder

Lifer
Jan 12, 2005
17,305
1,001
126
Now I have no idea if this is right or not, I'm quite the novice really.

But, I would think that the NB/L3 can live around the same voltage levels as the core. I have my cores at 1.468volts. Everything on the chip is built using the same process and same silicon, my *guess* is that AMD just turned down those frequencies to get 'good enough' performance for the factory core speed. With the cores at say 3.0GHz, they felt that in most scenarios that a 1.8GHz-2.0GHz uncore would be fast enough to not hamper performance, all while being able to save some power/heat over having those parts clocked much higher and with more voltage. As your cores get faster, the need for the NB/L3 to get faster becomes greater, and I would think (but have not tested in anyway, so I could be full of shit here...) that they could take the same to very near the same voltage as the core.

Now I understand that different parts may have different sensitivities to voltage and such, but I'm just going to say again that it's my guess that the NB/L3 can run at 1.4 volts without any long term issues.
 
Dec 30, 2004
12,554
2
76
Originally posted by: SlowSpyder
Now I have no idea if this is right or not, I'm quite the novice really.

But, I would think that the NB/L3 can live around the same voltage levels as the core. I have my cores at 1.468volts. Everything on the chip is built using the same process and same silicon, my *guess* is that AMD just turned down those frequencies to get 'good enough' performance for the factory core speed. With the cores at say 3.0GHz, they felt that in most scenarios that a 1.8GHz-2.0GHz uncore would be fast enough to not hamper performance, all while being able to save some power/heat over having those parts clocked much higher and with more voltage. As your cores get faster, the need for the NB/L3 to get faster becomes greater, and I would think (but have not tested in anyway, so I could be full of shit here...) that they could take the same to very near the same voltage as the core.

Now I understand that different parts may have different sensitivities to voltage and such, but I'm just going to say again that it's my guess that the NB/L3 can run at 1.4 volts without any long term issues.

That is what I have surmised as well.
OC'ing the CPU-NB (AKA L3 $) improves your performance anywhere from 10-20%, 20% for the higher clocks (like 2.6Ghz) and 10% for 2.2-2.4Ghz.