Few Phenom II questions

TheGame240 · Aug 6, 2009

A friend of mine ordered a PII setup that I'm assembling for him. He wants it overclocked as much as possible and I want to make sure I understand everything before the rest of his parts come in.

I'm not experienced at all with AM2/3 platforms, but I've read all the stickies and some of Dolk's threads and just want to make sure I've got everything down.

System specs:
PII 720 BE
GIGABYTE GA-MA790X-UD4P
G.SKILL 4GB (2 x 2GB) 240-Pin DDR2 SDRAM DDR2 1066 (PC2 8500) (2 sets 8gb total)
CORSAIR CMPSU-750TX
XFX 275
WD 750GB HD

So far I've just played around a little with it. He's ordered a better cooler so I can't do too much overclocking with just the stock cooler. I've managed to unlock the fourth core and get it stable with +.075v. I've also played around with overclocking just the 3 cores and gotten it to post at 3.4 with stock volts. I'll make more serious attempts once the cooler arrives. To get to the questions.

1. If it's prime stable with small FFTs, but fails blend then that's the memory controller, correct? Should I adjust the NB voltage or CPU-NB voltage? I'm pretty sure it's the CPU-NB I should be increasing, but I want to double check. Also what does the NB voltage increase since I was almost certain AMD didn't use a NB chip any longer?

2. ACC. Info on this is pretty scattered and most of what I've found seems to be speculation. My best guess is that is some sort of clock skew where if say core 2 was always failing prime/occt but every other core kept chugging along, I could increase the ACC percentage of that core and try to achieve stability that way instead of upping the voltage to all the cores. Anyone have any concrete data on this feature other than the ability to unlock the fourth core?

3. Ganged vs unganged. Ganged forces all the memory into one allocation heap. Increasing raw memory throughput, but decreasing multitasking performance. Is that correct? And for real world use I would assume unganged mode would be the best selection?

4. FSB overclocking. Does overclocking the bus increase performance by any noticeable amount? There's a lot of conflicting information out there about this and I want to know if it's worth the effort.

5. This one is going to be opinion based, but I'd like some input. How many MHZ is the fourth core worth? For instance if I can take it to 3.6 as a tri-core and 3.4 as a quad, then it's pretty safe to say I'm better off leaving it as a quad. But if I could only go to 3.2 as a quad then it get a little murkier. So what's everyone else's opinion on this?

6. What's the max, 24HR, voltage? While he's excited about getting a good overclock out of this chip, I would like for it to last a while. I've read where some are feeding them 1.5 so would 1.4 be considered a safe max (assuming temps are under 55C load)?

7. This last one is just for my personal info. I'm currently running an [email protected]. I've pre-ordered Windows 7 Pro and would like to be able to use XP mode. So I started looking at processors. The cheapest being ~$160 (which wouldn't be much of an upgrade). While the cheapest quad was ~$200. So then I got to thinking, I could just sell my MB/CPU and pick up a 720+MB for myself and come out better than I would with just a CPU purchase. How does a PII compare to a C2D clock for clock? Pretty well even? Is virtualization support about the same on both platforms? I figure it would be a step up even if I only break even on the clock speed I would still have an extra core or even a quad if I get lucky. I know this is the AMD forum but I just want the best bang/$ so if there are downsides to this idea I'm not seeing I'd like to hear about them.

mxthunder · Aug 6, 2009

You are on the right track.

best way to OC these is combo of multi and fsb. I stick with pure multi tho, as there are many things to watch when you start upping the FSB. ~1.5 is max 24/7 voltage, you can push a little higher if you have really good air cooling, also dependent upon your ambient temps.

ACC is not a way of cheating to maintain stability. It does help it legitimately though. Dont know too many details about it off the top of my head, but you could actually run into issues with it if you DONT want to unlock cores, but want to increase stability.

Im not too sure what this 'virtualization' craze is about and how its relevant to CPUs. They can all run virutal machines...

Neuromancer · Aug 6, 2009

The virtualization thing is over rated unless something has changed. The ability for chips to support virtualization is that software recognizes it so instead of your host OS processing all CPU duties, it just passes them straight to the CPU. Supposed to increase performance to near equal levels, although last benchmarks I saw, it appeared to actually run worse when VT was enabled.

Software companies may have fixed this though, just started out bad or poorly implemented at least.

In my limited experience, overclocking AMD is all about system performance not so much CPU performance. You can run up the CPU speed pretty high with even a mid range cooler, because they are do not throw a ton of heat. However for NB clocking you do get a big advantage from running a massive bolted on cooler. NB gets hot and if not cooled adequately takes everything else out with it.

Get your mem speed up as high as you can, some swear by 1000 MHZ 5-5-5 minimum. Have not done a lot of experimenting with mem clocking yet, I run 900+ 4-4-4 even though my sticks can handle close to 12 at 5-5-5. Whatever you do low voltage really seems to help here.

My Gkill 800s runs great up through 900MHZ 4-4-4 1T with 1.92-1.94 volts and delay 150 channel A 300 channel B

Have not figured out the best setting for my blades yet.

Safe voltage is entirely temp related on AMD

If you are running 55C load you will not be able to run 1.5 volts anyway

So keep it in the 1.4 vitamin range

Keep your HT link no more than stock speed. There are CPU benefits to dropping, but 3D penalties. In my experience in Vantage, the penalties to 3D exceeded the benefits to 2D so 1600 HT link at lowest for crossfired 4850s was the were I stopped seeing benefits.

unganged

What clock speed you benefit from vs core unlock is entirely dependent on what you do with it.

If this is not for you, I would recommend leaving it locked with a profile saved for unlocking the 4th core. Should something like video encoding need to be done

QuietIce · Aug 7, 2009

TheGame240 said:
1. If it's prime stable with small FFTs, but fails blend then that's the memory controller, correct? Should I adjust the NB voltage or CPU-NB voltage? I'm pretty sure it's the CPU-NB I should be increasing, but I want to double check. Also what does the NB voltage increase since I was almost certain AMD didn't use a NB chip any longer?

You should be adjusting the cpuNB voltage and it's multiplier.

The cpuNB is the internal link between the cores, L3 cache, IMC (Integrated Memory Controller), and HT Link. The HT Link communicates with everything else - video, other PCIe, and PCI.

In general the "NB" is really the HT Link's external aspect. It's where the primary PCIe and HT Link come together and also serves as a pass-through to the SB, which is where other PCIe devices and the PCI bus tie into the HT Link. HDD's, Ethernet, sound, etc. still run on a PCI bus.

TheGame240 said:
2. ACC. Info on this is pretty scattered and most of what I've found seems to be speculation. My best guess is that is some sort of clock skew where if say core 2 was always failing prime/occt but every other core kept chugging along, I could increase the ACC percentage of that core and try to achieve stability that way instead of upping the voltage to all the cores. Anyone have any concrete data on this feature other than the ability to unlock the fourth core?

Info on the ACC continues to be scattered because AMD is keeping pretty tight-lipped about it (last I checked). As near as I can tell it's what you've found, a type of clock skew, and is based off a timing signal from the SB.

Adjusting the ACC is very similar to some s939 board's ability to adjust the RAM skew. You can use it to tweak an extra few MHz out of the CPU - if you're lucky. The Phenom I's got a good boost from ACC, Phenom II's have internal ACC but can still benefit as a tweaking tool.

TheGame240 said:
3. Ganged vs unganged. Ganged forces all the memory into one allocation heap. Increasing raw memory throughput, but decreasing multitasking performance. Is that correct? And for real world use I would assume unganged mode would be the best selection?

Correct!

Ganged could benefit someone using programs that do a lot of huge data transfers - like video encoding, large Photoshop projects, and DTM. For most apps, unganged is the way to go.

TheGame240 said:
4. FSB overclocking. Does overclocking the bus increase performance by any noticeable amount? There's a lot of conflicting information out there about this and I want to know if it's worth the effort.

Using the clock (aka, system clock, ~~FSB~~) to OC is still controversial. Many believe strict multiplier-based OC'ing is the way to go. Being Old School I disagree. A good balance of the two is the best, IMO. I like to use the clock to raise cpuNB speed, which almost always has a lower-than-stock ratio to the max CPU OC. Then I use the CPU multiplier to raise the CPU from there. Using the clock also gets more performance out of your RAM. The difference between DDR2-800CL4 and DDR2-1066CL5 is pretty small to start and once you raise the DDR2-800 RAM above 860± MHz (a 215 MHz clock) it's performing better then DDR2-1066CL5 overall. Since the only way to get a RAM speeds between 400 MHz and 533 MHz is the clock, well let's just say it's obvious to me that clock OC'ing is still the best way to start once the individual component speeds have been tested using their multipliers.

TheGame240 said:
5. This one is going to be opinion based, but I'd like some input. How many MHZ is the fourth core worth? For instance if I can take it to 3.6 as a tri-core and 3.4 as a quad, then it's pretty safe to say I'm better off leaving it as a quad. But if I could only go to 3.2 as a quad then it get a little murkier. So what's everyone else's opinion on this?

That would all depend on exactly what apps you're running (and how many at once!). For Folding and SETI 4x 3000 MHz is pretty much the same as 3x 4000 MHz. That being said, most apps don't use even three full cores so, generally speaking, I'd pull out the same rules that have applied forever - CPU speed is King. Most apps will perform better with 3x 3600 MHz than even 4x 3500 MHz.

But if you commonly run 4-5 apps at once the lines become blurred very quickly. Then it comes back to the specific apps you're running and how much CPU time they're each taking up. If you're gaming on a modern game that actually uses two cores, and you're talking in-game video/pics, and you're running chat, and you're encoding music then the quad configuration might be better. It all depends on the most common usage you have.

TheGame240 said:
6. What's the max, 24HR, voltage? While he's excited about getting a good overclock out of this chip, I would like for it to last a while. I've read where some are feeding them 1.5 so would 1.4 be considered a safe max (assuming temps are under 55C load)?

I wouldn't be comfortable running much over 1.45v in 24/7 unless I was under water or had excellent air cooling and great case flow. But really, as long as your CPU load temps are consistently below 52°C or so, then 1.55v is considered the upper limit for 24/7.

TheGame240 said:
7. This last one is just for my personal info. I'm currently running an [email protected]. I've pre-ordered Windows 7 Pro and would like to be able to use XP mode. So I started looking at processors. The cheapest being ~$160 (which wouldn't be much of an upgrade). While the cheapest quad was ~$200. So then I got to thinking, I could just sell my MB/CPU and pick up a 720+MB for myself and come out better than I would with just a CPU purchase. How does a PII compare to a C2D clock for clock? Pretty well even? Is virtualization support about the same on both platforms? I figure it would be a step up even if I only break even on the clock speed I would still have an extra core or even a quad if I get lucky. I know this is the AMD forum but I just want the best bang/$ so if there are downsides to this idea I'm not seeing I'd like to hear about them.

Again it comes down to your application. Video encoding and Folding/SETI will generally do better on Intel. Most "common" apps will do better on AMD as bang/$$$. Phenom II basically closed the gap up to the Intel QX performance level except as noted. It takes an i7 to beat the Phenom II AM3's and even then games scores are very close with each chip taking about half the titles ...

ChanceCoats123 · Aug 7, 2009

TheGame240 said:
A friend of mine ordered a PII setup that I'm assembling for him. He wants it overclocked as much as possible and I want to make sure I understand everything before the rest of his parts come in.

I'm not experienced at all with AM2/3 platforms, but I've read all the stickies and some of Dolk's threads and just want to make sure I've got everything down.

System specs:
PII 720 BE
GIGABYTE GA-MA790X-UD4P
G.SKILL 4GB (2 x 2GB) 240-Pin DDR2 SDRAM DDR2 1066 (PC2 8500) (2 sets 8gb total)
CORSAIR CMPSU-750TX
XFX 275
WD 750GB HD

So far I've just played around a little with it. He's ordered a better cooler so I can't do too much overclocking with just the stock cooler. I've managed to unlock the fourth core and get it stable with +.075v. I've also played around with overclocking just the 3 cores and gotten it to post at 3.4 with stock volts. I'll make more serious attempts once the cooler arrives. To get to the questions.

1. If it's prime stable with small FFTs, but fails blend then that's the memory controller, correct? Should I adjust the NB voltage or CPU-NB voltage? I'm pretty sure it's the CPU-NB I should be increasing, but I want to double check. Also what does the NB voltage increase since I was almost certain AMD didn't use a NB chip any longer?

Blend failure means the cpu/nb and or the ram. So raising cpu/nb voltage can help stabilize.

2. ACC. Info on this is pretty scattered and most of what I've found seems to be speculation. My best guess is that is some sort of clock skew where if say core 2 was always failing prime/occt but every other core kept chugging along, I could increase the ACC percentage of that core and try to achieve stability that way instead of upping the voltage to all the cores. Anyone have any concrete data on this feature other than the ability to unlock the fourth core?

I used to have an option in my bios (even though nvidia shouldn't) and I always ran it on auto, but never noticed any help.

3. Ganged vs unganged. Ganged forces all the memory into one allocation heap. Increasing raw memory throughput, but decreasing multitasking performance. Is that correct? And for real world use I would assume unganged mode would be the best selection?

Unganged for regular usage.

4. FSB overclocking. Does overclocking the bus increase performance by any noticeable amount? There's a lot of conflicting information out there about this and I want to know if it's worth the effort.

I would put a big yes you increasing performance. Here is a way to show. Try using multiplier oc'ing to get to 3.5ghz with a stock ref. clock of 200 and run a super pi 1M test. Mine with a 250*14 oc' was 19 seconds. Your's should be a bit slower (20+ seconds).

5. This one is going to be opinion based, but I'd like some input. How many MHZ is the fourth core worth? For instance if I can take it to 3.6 as a tri-core and 3.4 as a quad, then it's pretty safe to say I'm better off leaving it as a quad. But if I could only go to 3.2 as a quad then it get a little murkier. So what's everyone else's opinion on this?

If I could unlock I probably would, but that depends upon whether or not it requires more voltage for said speeds or if it is unstable in certain areas.

6. What's the max, 24HR, voltage? While he's excited about getting a good overclock out of this chip, I would like for it to last a while. I've read where some are feeding them 1.5 so would 1.4 be considered a safe max (assuming temps are under 55C load)?

I used to run 3.750ghz 24/7 @ 1.52volts with a max temp of 52*C and I had no instabilites there so I guess it really depends upon your comfort level.

7. This last one is just for my personal info. I'm currently running an [email protected]. I've pre-ordered Windows 7 Pro and would like to be able to use XP mode. So I started looking at processors. The cheapest being ~$160 (which wouldn't be much of an upgrade). While the cheapest quad was ~$200. So then I got to thinking, I could just sell my MB/CPU and pick up a 720+MB for myself and come out better than I would with just a CPU purchase. How does a PII compare to a C2D clock for clock? Pretty well even? Is virtualization support about the same on both platforms? I figure it would be a step up even if I only break even on the clock speed I would still have an extra core or even a quad if I get lucky. I know this is the AMD forum but I just want the best bang/$ so if there are downsides to this idea I'm not seeing I'd like to hear about them.

Honestly, Amd has the cake in the server platform so I would have to say the virtualization is better on the Amd side. But I doubt it will impact you. But, lets put it this way, a stock 720 is about equal to a stock e8400. Now compare an e8400 to your current processor and then imagine the impact your cpu at 3.6ghz+ would be. I say 720+mobo!!!!

ViperX800 · Aug 7, 2009

Just to add a bit to #7, A Phenom II cannot keep up with a core 2 clock for clock. It takes anywhere from ~200 to 600mhz increase on the PII to best a C2Q depending on the application.
I personally have limited experience with virtual machines but as far as I am aware a PII pulls ahead of a C2D/Q in this regard, with performance very near 1:1(host/VM)

ChanceCoats123 · Aug 7, 2009

I would have to disagree with viper in this case. Amd cpu's can keep up with intel cpus clock for clock. Granted the Amd cpu's are towards the top end while intel's contender is in the lower spectrum of performance, but the fact remains. I don't mean to offend, but the e7200 is not necessarily a power house chip.

SCFAdamVcG · Aug 7, 2009

also a little question from me
why does this chip require such a big amount of voltage increase to move up?

ViperX800 · Aug 7, 2009

A rather in depth article some may be interested in reading, if they haven't already.
http://rage3d.com/reviews/cpu/deneb/

ChanceCoats123 · Aug 7, 2009

What do you mean? I need 1.4volts for 3.5ghz and 1.52volts for 3.75ghz (small fft's stable), but most chips in the family scale like that. It could have to do with the fact that my cooling isn't quite top notch. Lower temps means less volts. And remember all chips aren't the same.

SCFAdamVcG · Aug 7, 2009

Well I wouldn't imagine it should need so much even at 47c loaded for 1.53 volts vcore..47c isn't even close to hot.
60c is hot ,and even my old chip ,the pentium d 925 .. ofcourse was intel can even do 3.8ghz on 1.360 volts .. and thats a dinosour :S
I was reading through that article and got completely lost .. going on about I/O and the xbar :| ,it's some crazy **** and I don't know why amd left us to be our own technical engineer, I know there two different companies but honestly ,amd need to catch up for efficiency .. the heat output is phenominaly crazy considering I even had outside temps and the cpu was still hot.. seems a bit wierd to be honest.. i'll figure it out eventually but damn I'm certainly not an electronics engineer,my Dad's a genius with mechanics but not hardware too much.

ChanceCoats123 · Aug 7, 2009

You realize that the intel Pentium D 925 is a 90nm dual core while the 720 is a 45nm Trike. That is one more core and a lot more transistors.

SCFAdamVcG · Aug 7, 2009

actually I really didn't..
So that means that the total voltage given to the end of all those transistors is actually dropped to what would be 1.36 volts for a stupid frequency?
you might not understand me and it really doesn't matter..

velozzity · Aug 7, 2009

The Game,

I just noticed one thing, are you aware that the 1066 memory that you bought or your friend bought will most likely not work at 1066 with all 8 gigs installed, I have never personally experienced it but almost every memory forum that I have read where a person asks why his 8 gigs of 1066 wont run stably give a quick reply that it is not supported and or extremely hard to get it to run without loosening timings alot or dropping down to 800-900mhz range. Best 2 go with just one kit (4 gb) unless using some app that will really need more than 4gb.

ChanceCoats123 · Aug 7, 2009

SCFAdamVcG said:
actually I really didn't..
So that means that the total voltage given to the end of all those transistors is actually dropped to what would be 1.36 volts for a stupid frequency?
you might not understand me and it really doesn't matter..

Well, there is less voltage, but a processor is not necesarily a simple device (to us), but I do somewhat understand that all those transistors need to be working in a set order (which is why it is so hard to get higher speeds when IBM got a single transistor to 500ghz). But it also takes more voltage to get things working correctly in the right order.

QuietIce · Aug 8, 2009

velozzity said:
The Game,

I just noticed one thing, are you aware that the 1066 memory that you bought or your friend bought will most likely not work at 1066 with all 8 gigs installed, I have never personally experienced it but almost every memory forum that I have read where a person asks why his 8 gigs of 1066 wont run stably give a quick reply that it is not supported and or extremely hard to get it to run without loosening timings alot or dropping down to 800-900mhz range. Best 2 go with just one kit (4 gb) unless using some app that will really need more than 4gb.

Good catch - but I just figured he'd run them at an OC'ed DDR2-800 spec. I do.

SCFAdamVcG said:
also a little question from me
why does this chip require such a big amount of voltage increase to move up?

I've never really been much for Intel CPUs so I don't know what kind of voltage you're talking about but what does it matter what the voltage is? By itself, voltage is just an electrical specification - nothing more.

Considering the huge heat difference between processors what you might be asking yourself is, "How do AMD's run so fast with so much less heat output?" ...

SCFAdamVcG · Aug 8, 2009

lol sorry ,Stupid question ,but anyway.. amd cpu's produce less heat at a higher voltage? :O
I was thinking the less voltage the smaller the amount of heat the processor would put out .. :S

TheGame240 · Aug 8, 2009

velozzity said:
The Game,

I just noticed one thing, are you aware that the 1066 memory that you bought or your friend bought will most likely not work at 1066 with all 8 gigs installed, I have never personally experienced it but almost every memory forum that I have read where a person asks why his 8 gigs of 1066 wont run stably give a quick reply that it is not supported and or extremely hard to get it to run without loosening timings alot or dropping down to 800-900mhz range. Best 2 go with just one kit (4 gb) unless using some app that will really need more than 4gb.

QuietIce said:
Good catch - but I just figured he'd run them at an OC'ed DDR2-800 spec. I do.

I've never really been much for Intel CPUs so I don't know what kind of voltage you're talking about but what does it matter what the voltage is? By itself, voltage is just an electrical specification - nothing more.

Considering the huge heat difference between processors what you might be asking yourself is, "How do AMD's run so fast with so much less heat output?" ...

I originally told him to just order 4GB of ram, but there was a problem with his credit card. He had a $200 dollar limit in place and didn't know it. So he started breaking the order down into smaller increments and at the same time had his bank lift the limit. Well newegg apparently processed the smaller order and the large one. So he ended up with two cases, ram kits and wireless cards. He decided to go ahead and use the ram anyway. He pre-ordered Windows 7, so I figure if nothing else, the superfetch function should make some apps appear a lot snappier with 8GB to work with.

I did notice the lack of 1066 support. Stock was fine with the 5.33 memory multiplier, but the moment I went above that, random core failures. That's one of the reasons I was asking about increasing the NB voltage.

I think he's going to be primarily using this for gaming, so I may just stick to leaving it a tri-core and push every last MHz out of it I can. I do have a couple of questions left though.

When pushing these, specifically using the base clock, I should adjust the NB/HTT multiplier to keep them in the 2000-2200 range correct?

In regards to the memory, is the 5.33 multiplier the problem, or is it a certain memory frequency where everything tends to go wrong? I noticed I would get random core failures in the blend test, but FFTs I could leave running 24hr+ and I've ran Memtest for over 20hrs with the memory at 1066.

In regards to myself. I know the e7200 isn't exactly a powerhouse processor, I was mainly commenting on the price/performance ratio. As long as the two are comparable clock for clock (within a couple of percentage points) I think the extra core and hardware virtualization support (required for XP mode in Windows 7 and one of the few reasons to have spent the extra $50 during the pre-order) would make the 720 a more logical upgrade, as much as it could be called an upgrade. I don't do a lot of media encoding on my main rig. I have a second rig I use for that even though the main is much faster. I do multi-task a lot. It's not uncommon for me to have IE/FF/Chrome/Safari/Opera open with multiple tabs, Photoshop, Dreamweaver and a media player of some sort going at the same time.

Thanks for all the input so far. It's clarified a lot of my confusion.

TheGame240 · Aug 8, 2009

SCFAdamVcG said:
lol sorry ,Stupid question ,but anyway.. amd cpu's produce less heat at a higher voltage? :O
I was thinking the less voltage the smaller the amount of heat the processor would put out .. :S

Not from what I've read. Unlike Intel processors/older model AMDs, the Phenom IIs heat output seems to be directly proportional to clock frequency. I would assume it would have to do with a more efficient design. :shrug:

ViperX800 · Aug 8, 2009

SCFAdamVcG said:
lol sorry ,Stupid question ,but anyway.. amd cpu's produce less heat at a higher voltage? :O
I was thinking the less voltage the smaller the amount of heat the processor would put out .. :S

This is true within the same architecture. A 955 at 1.35 volts will output far less heat then the same chip at 1.5 volts.
The problem with comparing Intel and AMD in regards to voltage and heat output is that the 2 companies use very different methods of constructing the chpis.

Few Phenom II questions

Member

Member

Member

Disabled

d20 in a jacket

Registered

d20 in a jacket

Member

Registered

d20 in a jacket

Member

d20 in a jacket

Member

Member

d20 in a jacket

Disabled

Member

Member

Member

Registered

Similar threads