Opteron 165 o/cing

[nealh]

Originally posted by: Gldm
DFI LanParty NF4 SLI-DR. It turns out the 2.9 range was too unstable. I could boot as high as 2925, but superpi or other intensive apps would bluescreen eventually. I've got it stable now at 2835 running on around 1.62v. The mult is 9.0x315. RAM is running at 5/6 which gives about 262.5. I haven't readjusted my RAM timings to the best performance yet, it's still running on auto. I'm losing a bit of RAM speed between 2700 (270) and 2835 (262.5) but I think overall the higher CPU speed will work out and I might be able to make up the difference with tighter memory timings. I've got HT running stable at 4.0x315 = 1260, northbridge and HT bus are stable at their lowest setting (to help keep temps down). I'm getting about 36 idle 41 load on the CPU with a Swiftech Apex Ultra on it.

You better be very careful with that vcore..the DFI SLI DR overvolts under load..at least 0.025v

you will fry your chip with so much vcore...I would never give a Dual opteron 1.7v(unless under pashe and even then risk of killing that chip is very high)...1.6v+ is a big risk

 

[VERTIGGO]

Originally posted by: TigerClaw27
I have an 0550, it does about 2.6 on stock Vcore. Make sure you have good cooling in your rig too (had it running on an Asus A8N-E but the MOSFETs were reporting insane temperatures, 110 - 120C at load, so I switched to the DFI LP Ultra-D).

On a related topic, anyone know how to short the temperature sensor on the stock Opteron HSF to get it to run at max speed, and control the speed in bios? My stock HSF is having troube going past 3200 RPM, which i think is limiting my overclock (my chip will reach 60C and the damn thing is still spinning at only 3200 RPM, sometimes even slower).

just a note, when temps over 100C are reported, it's often an indication that there is no sensor for that component.

 

[Gldm]

Originally posted by: nealh

You better be very careful with that vcore..the DFI SLI DR overvolts under load..at least 0.025v

you will fry your chip with so much vcore...I would never give a Dual opteron 1.7v(unless under pashe and even then risk of killing that chip is very high)...1.6v+ is a big risk

Actually mine undervolts. I set 1.65 and get 1.61-1.62. I don't see it being a problem as long as the chip stays under 50C. Voltage isn't the killer, heat is.

 

[atybimf]

Got a Opteron 165 CCBWE 0551UPMW from TankGuys here. Does 2.5 on 1.325v. Does 2.6 on 1.375v - 9 hours dual Prime stable in a few minutes. I'm content at 2.6 and I like lower voltages and less heat so I haven't tried anything higher. Very very pleased. This is on a DFI LP Ultra-D with Mushkin XP4000 RAM.

Edit: Thought I might mention that I upgraded from an Opteron 148 to the 165, and I am glad I did. Dual core all the way!

 

[CosmoHorizon]

Hi people, i've got mine CCBWE 0551VPMW, and i am currently investigating the OC possiblity with my rig.
Basically, i am currently sitting on 2.4 @ 1.3V (yes, i down volted the cpu) with 300*8
I have been sitting on stock volt with 2.4GHZ for a while and i decided to try out exactly how much volt will it run prime stable when clocked to 2.4GHZ. However, I noticed a problem of mine failing prome wiht the 320*7.5 = 2.4GHZ, which is exactly the same, but not quite getting it even with stock core of 1.35V (speedfan displays 1.38V)

When I get my new rig (P180 and Ninja), I'll have a look at bumping votage up to 1.4 and have it run 320*8 = 2560 or something along the line of that. Wouldn't try to push mine beyond the 2.6ghz barrier.

Cheers!

 

[nealh]

Originally posted by: Gldm

Originally posted by: nealh

You better be very careful with that vcore..the DFI SLI DR overvolts under load..at least 0.025v

you will fry your chip with so much vcore...I would never give a Dual opteron 1.7v(unless under pashe and even then risk of killing that chip is very high)...1.6v+ is a big risk

Actually mine undervolts. I set 1.65 and get 1.61-1.62. I don't see it being a problem as long as the chip stays under 50C. Voltage isn't the killer, heat is.

That is 100% wrong voltage will kill a chip..there is an Intel engineer around with a nice thread..

link to thread

http://forums.anandtech.com/messageview...atid=28&threadid=1773169&enterthread=y

Sorry, it's hard to gauge how technically I should answer a question. If I don't make it technical then I feel like I'm not really being very accurate... I think it's an engineering trait, I notice my co-workers do this a lot too.

Sorry, I can't do l33t speak. So I'll have to try English.

There are 4 ways that chips break:
electromigration: metal atoms in a wire move around because lots of lots of electrons hit them. Move enough metal atoms and the wire will either short with another wire, or have a hole in it and stop working.
TDDB: the insulation material that separates two key parts of a transistor breaks down creating an electrical short - which breaks the transistor. In most cases, if any one of the hundreds of millions of transistors breaks, it will break the chip.
hot-e: a key parameter of the transistor called "Vt" (threshold voltage) shifts over time - which essentially slows the transistor down. If it slows down enough, then the chip will calculate an incorrect value.
BTI: Similar to hot-e but for a different type of transistors and happens for a different reason. Usually fixed in the factory.

In all but hot-e, increasing the temperature a little makes the chip a little more likely to break (this is all statistics... there is no "do this and this will kill your chip..." it's all a matter of probability). In hot-e, lowering the temperature makes it worse.

In all of these, increasing the voltage a small amount makes them much more likely to break the chip.

As to why 10% more voltage is much worse than 10% more temperature, well let's take the example of electromigration and look at it in detail.

Wires are made up of atoms all lined up. Electrons flow through these atoms. An electron is a very small thing, and atoms are a lot bigger. So the idea of an electron moving an atom around is a lot like someone trying to move a car (atom) by shooting a BB (electron) at it. Clearly to ever hope to move a car by shooting BB's at it, you would need a lot of BB's... a storm of BB's. But if you get enough, the car will move. Millions of BB's and that car will likely start getting pushed around. The temperature of the chip could be thought of as how slippery the road is. A little bit more slipperiness isn't going to help a BB move a car. It helps a little but not a lot. On the other hand, the voltage determines how many BB's you have, and it's not like 10% more voltage is 10% more BB's (electrons), you get a lot more than 10%. And worse than that, because increasing the voltage in a chip also increases the temperature (all things being equal, like same heatsink, same air temp, etc.), increasing the voltage is a double-whammy.

The others are similar... but more complex to explain (I'm still not sure that the experts really completely understand the low-level details of both TDDB and BTI...).

Your chip you do what you feel comfortable

Chips will dies with too much vcore no matter temps
chips will die with too high heat no mater vcore

Both will more rapidly kill the chip

I have no idea how long it takes..these opteron I think will do well overall

You say your undervolts are you basing that on voltimeter or winxp reading

the winxp readings are wrong with all programs and read vcore wrong

cpu-z reports 1.456 v but I have bios set to 1.5v

if you set bios to 1.65v...at idle you are getting 1.65v and load...1.68v or more...check with a volitmeter...this has be proven at DFI Street and XS forums

 

[TigerClaw27]

Originally posted by: VERTIGGO
just a note, when temps over 100C are reported, it's often an indication that there is no sensor for that component.

The reported mosfet temperatures were within reason when the chip wasnt under load ( 60 C ) or wasnt overclocked (80 C). So I suspect the temperature was being read, but perhaps over estimated?

TC27

 

[nib95]

Originally posted by: Gldm
I've got a 0551UPMW here. It went to just above 2.5 on stock voltage with the boxed cooler, 2.7 ROCK SOLID @1.500v. This was an easy 50% OC. I mean like Celeron 300A @450 easy.

I've replaced the stock HSF with a Swiftech Apex ultra. So far it's up to 2907 @1.75v. I'm not sure how much more it can go, but I'm not done with it yet.

I have that same stepping with an Opteron 170, I've heard its the best stepping out there for OC'ing.
Once my PC is built (should be Friday) I will have fun trying to OC that baby.
I have the Asus A8N32-SLI Deluxe, Mushkin 2x1GB XP4000 and a Thermaltake Big Typhoon for the job. Hopefully I can hit 2700mhz on air. That would do me more then well enough.

 

[Gldm]

Originally posted by: nealh

Your chip you do what you feel comfortable

Chips will dies with too much vcore no matter temps
chips will die with too high heat no mater vcore

Both will more rapidly kill the chip

I have no idea how long it takes..these opteron I think will do well overall

You say your undervolts are you basing that on voltimeter or winxp reading

the winxp readings are wrong with all programs and read vcore wrong

cpu-z reports 1.456 v but I have bios set to 1.5v

if you set bios to 1.65v...at idle you are getting 1.65v and load...1.68v or more...check with a volitmeter...this has be proven at DFI Street and XS forums

Even if I had a voltmeter, I wouldn't know where on the board to measure it. I was going by the BIOS and Everest readings, which agreed with each other. But I'll take your word for it as I haven't been on the forums you mentioned and I've found the DFI board has quite a few quirks.

If you do your research on electromigration you'll find it's far less of a problem with copper interconnect CPUs. The main problem with higher voltages is vibration of the atoms causing them to deviate from the ideal lattice. Fortunately we have a way to measure this, it's called temperature. Lower temp = less vibration. So, it's important to maintain lower temperatures than the maximum spec as vcore increases. Most CPUs today are speced to tolerate 80C at the stock voltage. If vcore is 150% then power should be 225%, which will obviously increase the thermal load. As long as the temp is kept well under the rated maximum it should be fine even at the higher power setting IMHO.

I've been overclocking CPUs since the 486 age (Yay AMD DX2/80@120 with 40mhz VLB!). Generally I've found no problems with a 10-15% bump in voltage, they run for years. So I'm quite sure 1.5v isn't a problem if I go back to 2.7. 1.6-1.65v is more on the order or 30-40%, so yes I'm expecting a shorter CPU life even with watercooling. However, I'm not intending to keep this CPU beyond a year or so tops.

I guess I'm probably being reckless with it because I didn't buy it. AMD gave it to me.

 

[Bobthelost]

Gldm, processes change and if you'd read the thread linked to you'd see why none of your comments are all that valid. Electromigration is a problem with current generation CPUs.

 

[CnnmnSchnpps]

Originally posted by: TigerClaw27
On a related topic, anyone know how to short the temperature sensor on the stock Opteron HSF to get it to run at max speed, and control the speed in bios? My stock HSF is having troube going past 3200 RPM, which i think is limiting my overclock (my chip will reach 60C and the damn thing is still spinning at only 3200 RPM, sometimes even slower).

What I did to "fix" the issue was just shorting the two pins of the temperature sensor (the little green thing sticking out of the fan there the wires plug in to it). I did this the high-tech way: aluminium foil and electrical tape. Ideally you would want to solder on a small resistor to make sure nothing gets burned out, but I only need this to last a couple more days until my XP-120 comes in... Then we'll see about some serious overclocking.

edit: This allowed my fan's RPMs to go from 2700 max to about 5100 max.

 

[Tweakin]

Originally posted by: Gldm
I've got a 0551UPMW here. It went to just above 2.5 on stock voltage with the boxed cooler, 2.7 ROCK SOLID @1.500v. This was an easy 50% OC. I mean like Celeron 300A @450 easy.

I've replaced the stock HSF with a Swiftech Apex ultra. So far it's up to 2907 @1.75v. I'm not sure how much more it can go, but I'm not done with it yet.

Those are crazy high vcore numbers...I see smoke in your near future...I wouldn't push any dual core above 1.45 with air...

 

[Adegan19]

I've got a 0551 UMPW, best I can get is 255mhzx9 @ 1.35v, even if I increase core voltage it still doesn't go any higher than it did when at stock. It gets to right before verifying dmi pool data and restarts everytime, think I can get any higher than 2295mhz?

 

[Bobthelost]

Have you dropped the RAM timings down yet?

 

[imported_blkdiamond]

I just set up my 165 rig this week. I have the 0551 UMPW Opteron from Newegg running on an Asus A8N SLI Deluxe motherboard and 2 GB (4x512) OCZ Platinum memory. I have been able to get 2.5 GHz (278x9) at 1.375 vcore. I have it 24 hours prime stable on both cores. The cooler I have is the Arctic Freezer 64 (28 idle, 40 load). I really don't want to push too much more, I am extremely happy with this OC.

 

[Lark888]

I know that increasing voltage/temp decreases the life of a CPU. The question is when does that reduced life have an impact on you. Another factor that affects the life of a CPU is Moore's law (cpu power doubling on a regular basis). I have CPUs sitting in the drawer that were seriously overclocked (thereby shortening the life) such as the Celeron 300a in still functioning condition. In fact, I'm pretty sure one of my server type setups is still running a Celeron 300a chip 24/7 overclocked to 450.

Progress is the real CPU killer.

 

[Ackmed]

Hmm.

 

[koolzb]

Originally posted by: Lark888
I know that increasing voltage/temp decreases the life of a CPU. The question is when does that reduced life have an impact on you. Another factor that affects the life of a CPU is Moore's law (cpu power doubling on a regular basis). I have CPUs sitting in the drawer that were seriously overclocked (thereby shortening the life) such as the Celeron 300a in still functioning condition. In fact, I'm pretty sure one of my server type setups is still running a Celeron 300a chip 24/7 overclocked to 450.

Progress is the real CPU killer.

Very good point, most often the CPU is die of the Moore's Law.