why did AMD leave the Phenom II CPU-NB at 2GHz?

[sangyup81]

I see all kinds of gaming benchmarks showing that overclocking CPU-NB does more for performance than overclocking the CPU speed (at least clock for clock)

4GHz Phenom II loses to Core2Quads because those bench markers left CPU-NB at default

My Phenom II at 3.95GHz would never lose to a 3.95GHz Core2Quad due to having a 2.9GHz CPU-NB overclock

You would think that AMD would come out with some Phenom IIs running at 2.4GHz CPU-NB default at least

 

[VirtualLarry]

It was due to power reasons. If they increased the speed/voltage of the CPU-NB, they would exceed standard desktop TDPs. It was bad enough that early Phenom II chips were 140W TDP at the top bins.

 

[Idontcare]

It's also a die-size issue.

To make higher clocked circuits you need more drive current, to get more driver current you have to make the xtors wider, making the xtors wider makes the xtors take up more space, more space means bigger chips which means lower yields and higher manufacturing costs.

 

[formulav8]

Binning and yield costs. Plus power consumption as mentioned.

It wouldn't really be worth it in performance unless they could get the nb to 2600mhz. That is quite a bit more to worry about.

 

[AnandThenMan]

Power consumption for sure, yes raising the NB speed did a lot for performance, but it did even more for power draw.

 

[superccs]

I thought that there was a negative in other types of usage situations.... or maybe that was when you OC'd the hypertransport clock. I think I dabbled with OCing the NB speed and got quickly slapped with stability issues at stock voltages.

 

[soccerballtux]

^was your bus not @ 200mhz? I had makor stab issues until I turned the fsb back to 200, then I hit 2.8ghz on cpu-nb easy.

 

[sm625]

This post shows data that seems to suggest NB OC does almost nothing in a wide variety of tests.

http://www.overclock.net/amd-cpus/816159-phenom-ii-cpu-nb-overclocking-review.html

Are you calling Asmola a liar or a fraud?

 

[waffleironhead]

This post shows data that seems to suggest NB OC does almost nothing in a wide variety of tests.

http://www.overclock.net/amd-cpus/816159-phenom-ii-cpu-nb-overclocking-review.html

Are you calling Asmola a liar or a fraud?

Maybe the things he tested werent held back by the NB to begin with.

 

[LOL_Wut_Axel]

This post shows data that seems to suggest NB OC does almost nothing in a wide variety of tests.

http://www.overclock.net/amd-cpus/816159-phenom-ii-cpu-nb-overclocking-review.html

Are you calling Asmola a liar or a fraud?

CPU-NB seems to do nothing for performance in some applications when over 2800MHz or so. I think around 2600-2800MHz is the best place to have it with a very minor over-volt. If it's on stock voltage, 2400MHz.

 

[Black96ws6]

This post shows data that seems to suggest NB OC does almost nothing in a wide variety of tests.

http://www.overclock.net/amd-cpus/816159-phenom-ii-cpu-nb-overclocking-review.html

Are you calling Asmola a liar or a fraud?

Did you read the entire thread?:

Cause i used almost maximum graphical settings on those game test's, it might look that CPU-NB overclocking is useless. So i did one test with Far Cry 2 minimum graphical settings (DX10, everything high<-lowest setting with DX10

)
and results are much more clear now, also updatet this on first post:

Improvement ~12,5%

So now it's obvious that CPU-NB overclocking really does have huge effect on performance, atleast with lower graphical settings and with lower-end graphics cards.

 

[Black96ws6]

CPU-NB seems to do nothing for performance in some applications when over 2800MHz or so. I think around 2600-2800MHz is the best place to have it with a very minor over-volt. If it's on stock voltage, 2400MHz.

This link seems to indicate that ~2600 is the sweet spot, but then you have guys replying to the thread above who say they notice a difference as it goes higher...

http://www.overclockers.com/the-importance-of-northbridge-overclocking-with-the-phenom-ii/

 

[Kalessian]

Did you read the entire thread?:

I agree, it looks bad for games because the tests he performed were for the most part GPU-limited.

I'd like to see more tests done with lowered settings / older games.

12.5% increase in Far Cry 2 hit it on the head for me.

 

[sm625]

12% on a 50% overclock is not exactly screaming performance scaling. A 50% overclock on the cpu would get you a lot more, no? I doubt a 50% NB overclock could even make up for a 5% core underclock.

 

[busydude]

12&#37; on a 50% overclock is not exactly screaming performance scaling. A 50% overclock on the cpu would get you a lot more, no? I doubt a 50% NB overclock could even make up for a 5% core underclock.

You you have power consumption figures? NB takes up relatively small die area.. so I figure the impact on temps and increase in power consumption not to be that great.

 

[Kalessian]

12&#37; on a 50% overclock is not exactly screaming performance scaling. A 50% overclock on the cpu would get you a lot more, no? I doubt a 50% NB overclock could even make up for a 5% core underclock.

The op may have overstated its usefulness, but 12% on a chip that otherwise won't overclock anymore is nothing to sneeze at. The way I look at it is a free IPC boost, enough to put it over Penryn (I think). I do agree with op in that reviewers don't seem to know the best way to overclock PhII, I also suspect that they don't optimize the ram for PhII either.

Assuming performance scales linearly with clockspeed (ok, a bad assumption), a 12% boost on a 4.2 ghz would be ~500 mhz. Not bad.

Anyone have any starcraft numbers? I know PhII scaling is poor in that game, could low cpu-nb speeds be the problem there?

Edit: Right, http://www.anandtech.com/show/3877/...investigation-of-thuban-performance-scaling/7

16%

 

[toyota]

I agree, it looks bad for games because the tests he performed were for the most part GPU-limited.

I'd like to see more tests done with lowered settings / older games.

12.5% increase in Far Cry 2 hit it on the head for me.

so you would oc the NB within an inch of its life creating more heat and power usage to get 12% increase at settings that you would never use?

 

[LOL_Wut_Axel]

so you would oc the NB within an inch of its life creating more heat and power usage to get 12&#37; increase at settings that you would never use?

You'd probably get the same result as 3000MHz with 2600MHz CPU-NB, so it's not a big issue. You can get 2600MHz out of a C3 most of the time by raising the NB voltage by one notch only.

 

[soccerballtux]

You'd probably get the same result as 3000MHz with 2600MHz CPU-NB, so it's not a big issue. You can get 2600MHz out of a C3 most of the time by raising the NB voltage by one notch only.

Supposedly 2.6ghz is enough for like 4ghz or something CPU.
Actually it was 2.66ghz now that I recall. Somebody had calculated it. I run at 2.8 for my 4ghz. Going from 2.0 to 2.8 made a HUGE impact on the minimum FPS dips during the intro panning video in WoW (not prerendered).

 

[Kalessian]

Supposedly 2.6ghz is enough for like 4ghz or something CPU.
Actually it was 2.66ghz now that I recall. Somebody had calculated it. I run at 2.8 for my 4ghz. Going from 2.0 to 2.8 made a HUGE impact on the minimum FPS dips during the intro panning video in WoW (not prerendered).

i'd be interested in that calculation if you could find it

 

[Anarchist420]

I think it's ridiculous that AMD doesn't have different black edition levels. Due to AMD's poor binning practices, I'm switching to an intel system soon. I would've paid $300 for a 3.2GHz PII X4 that had a 2.6 GHz NB as long as it ran cooler than my PII X2 555 BE and as long as it worked with DDR2 memory. Even better would've been if they sold well binned X6s Black Eds and then you could disable two cores so that the vcore could be reduced stably.

I'm pretty sure that the super slow NB speed (2.4GHz according to wikipedia) of Bulldozer is going to hold it way back. AMD makes some stupid design choices IMO.