Unofficial Opteron 939 Thread

[ixtapalapaquetl]

NOTICE: DFI DOES NOT SUPPORT THE USE OF OPTERON PROCESSORS WITH THEIR MOTHERBOARDS. IF YOU CHOOSE TO USE AN OPTERON, KNOW THAT YOU ARE EMPLOYING HARDWARE THAT HAS NOT BEEN SPECIFIED FOR YOUR BOARD. THUS, YOU SHOULD NEITHER EXPECT NOR REQUEST SUPPORT FROM ANGRY, RGONE, OR ANY EMPLOYEE OF DFI. THIS NON-SUPPORT STATUS HAS BEEN MANDATED BY AMD, AND IS UNLIKELY TO CHANGE IN THE FORESEEABLE FUTURE.

Ok guys, as more and more of us pick up these wicked-bad Opterons, it only makes sense to consolidate our questions, comments, and information into a single thread. Most of my knowledge of these chips has come from the good folks at Xtreme Systems. They have a lengthy thread over there that is an excellent source for those wanting to learn more about Opterons, including numerous benchmarks. If you are looking for results and/or steppings, you might try this German site (for both single core and a few dual cores) or this Australian site (OCAU - some great clockers down under), as both have done a good job organizing their users' data. Below you will find some general information that I have gathered from the collective experience of many, many knowledgeable people all over - thanks to all for your hard work.

 

 

 

Why do People Keep Talking about Opterons? What's the Big Deal? Before the Socket 939 Opterons were even actually released, their was a good deal of buzz about the opportunity to obtain a relatively inexpensive CPU that would boast a 1Mb L2 Cache. Of particular interest were the lower numberd models (the 144 and 146), which essentially promised to be the first entry level San Diego cores - and that at a price lower than the 3000+/3200+ Venices. Needless to say, people were excited. When the Opterons finally arrived, many of the early adopters quickly realized that their chips were absolutely tremendous overclockers. During those early weeks, forums were flooded with example after example of single core Opterons achieving SuperPi stability at 3GHz (or more) - and doing so on air at just above stock voltages. Results such as these fomented the growth of the initial pre-release buzz into a full blown hysteria, and was the genesis of the current hype over 939 pin Opterons.

An Important Note Regarding the Interpretation of Results: The majority of Opteron owners are posting at XS. It should be known that there is a major cultural difference between XS and dfi-street regarding measuring stability. There, stability is often taken to mean SuperPi 32M completion, whereas here stability is more often accepted to be 8+ hours prime stable. The numerous results at XS at 3.0GHz are predominantly 32M stable, and not necessarily (though possibly) prime stable. You should base your expectations of an Opteron's performance with this information in mind.

 

Any Further Advantages? Since Opterons are designed for the high stress environments of servers, it is speculated that they are cast using better silicon than Athlons. This may explain their apparent success. It is further speculated that they have better on-die memory controllers for the same reasons. These speculations are at least partially confirmed by available results, several users comparing their memory controllers to their FX-57's.

 

Availability: Opterons are rather difficult to come by. The demand far exceeds the supply. Monarch, ZZF, and Newegg are known to carry them on and off, but you should also look at smaller distributors (Ewiz, Compumusic, Tankguys, etc.). There are also independents with access to Opterons – see the For Sale/Trade thread here at DFI-street for such an independent. Don't forget about Froogle. In any event, patience and persistence will be required.Opterons are rather difficult to come by. The demand far exceeds the supply. Monarch, ZZF, and Newegg are known to carry them on and off, but you should also look at smaller distributors (Ewiz, Compumusic, Tankguys, etc.). There are also independents with access to Opterons – see the For Sale/Trade thread here at DFI-street for such an independent. Don't forget about Froogle. In any event, patience and persistence will be required.

 

Word has been floating around the net that AMD has begun to take steps to restrict the availability of Opterons to owners of non-workstation boards, i.e. the overclocking community at large. These rumors have recently been confirmed by the manager of AMD's Opteron Division. At the present time, it is AMD's intention to no longer supply Opterons through regular retail channels once the current supply is exhausted (though this is subject to change in the future). Thus, in the near future it will likely be the case that the only Opterons available to enthusiasts will be the handful that leak through the Gray Market. I therefore recommend that if you are considering purchasing an Opteron, you do so without delay – you may not have the opportunity to do so in the future! Thanks to Ben at TankGuys for this information.

 

Model Numbers:: Multipliers are downward unlocked and upward locked – just like AMD64’s.

 

Single Core Chips: 144:200x9; 146:200x10; 148:200x11; 150:200x12; 152:200x13, 154:200x14

Dual Core Chips: 165:200x9; 170:200x10; 175:200x11; 180:200x12

 

Single Core Steppings: The list below is a general guide to the strength of steppings from best to worst (as determinded by their probability of reaching 3GHz):

 

1. CABNE 05xx (excluding 0540) / CABYE 0540

2, CABNE 0540 / CABYE 05xx (excluding 0540)

3. CAB2E 05xx

4. CABGE 05xx

 

It should be known that CAB2E's are more variable than other steppings. They are a bit hit or miss, a fact which diminishes their probability of hitting 3GHz. 146 CAB2E's seem stronger than 144 CAB2E's do. Recently, weeks later than 0540 have begun to show up, but we do not have many results yet.

Dual Core Steppings: At this time, the single most important factor determining the strength of dual core Opterons is the week number. As of now, there are two steppings: week 0530, and everything else. 0530's have the potential to flirt with 3GHz, though most have fallen shy. Non-0530's thus far tend to clock between 2.6GHz and 2.75GHz. Some interesting new steppings with weeks later than 0540 are now coming out; results pending.

 

Temperatures and Cooling: Opterons run hotter than Athlons. Single core maximum temperatures are rated between 65C and 71C, while dual cores are rated between 49C and 65C (AMD specs). The insane overclocks (over 3.4 GHz) are all on phase. An XP-90 and good case flow may be sufficient for 3GHz on a single core (depending largely on stepping), but in all cases water cooling will likely give you a boost in your overclock. This may be even more true in the case of dual cores since they are not rated to run as hot.

 

Removing the IHS has led to dramatic temperature decreases for many users. For information on removing the IHS, look here and search other forums. Be advised that this is a potentially CPU-killing procedure with inherent risks invlolved. Do your research or you'll be sorry. If you use a clip-on HS such as an XP-90, additional precautions must be exercised. Since clip-ons are very difficult to apply with even pressure, you run the risk of crushing the edges and/or corners of the exposed chip, and that is quite obviously very bad. In a stroke of genius, nomad has suggested a workaround that results in a ghetto screw-on heatsink. Remove your bracket from the board, clip the HS on, then position the HS/bracket unit over your CPU (after applying a small amount of TIM, of course). Then screw the bracket/HS unit back on, turning the screws on each side alternately, a little at a time. For more info, search the above link for "XP-90." For best contact, it should be noted that removing the IHS removes about 1mm of height, so one should shave down the contacts on the underside of the bracket about 1mm. Know that the use of any of the above techniques is solely at your own risk! Also know that removing the IHS will void your CPU's warranty!!!

 

Memory: As with all other socket 939 CPUs, 939 pin Opterons do not work with registered/buffered memory. They require unregistered memory. Don’t confuse these chips with the 940 pin models. Excellent results have been achieved with both TCCD and BH-5. It is widely held that they are very sensitive to both read preamble and max asynchronous latency, so if you are having difficulty, try adjusting these settings. Users commonly use 9/7.5, 9/7, 8/6, and 8/5.5. It appears that Opterons, like Athlons, will not run 4x512 at 1T. Some have had problems using dividers (usually 180 and/or 166). If you are having problems passing SuperPi or Prime95 large FFT's, run your memory at 2T. Subsequent success indicates that your memory is the problem, not your CPU.

 

Motherboard Compatibility and BIOS: I have not seen any reports of incompatibilities with DFI boards (nF3 or nF4). In fact, the vast majority of users are using DFI nF4 boards. All other socket 939 boards should be fine. On nF3 boards, BIOS versions 711, 815, and 824 all work. On nF4, it may be difficult to acheive 7.5 read preamble with BIOS 623-3, but 704-2BT 510-xFIX do not have this problem.

 

Please post any questions, comments, and especially BENCHMARKS here. Also, let me know if I should include any more information above.

 

 

148 CABNE 0528GPMW

 

nF4 Ultra-D results. Still learning this board; will probably tighten up a bit as I gain experience. However, I will probably not be running these settings 24/7 since I can run 3GHz stable at just over 1.4V - a 0.2V increase for a .1GHz gain is probably not a sound decision on air.

 

Prime, 3D2001, Everest read, idle temps:

 

 

 

3D2003, SuperPi 1M, Sandra, Everest write:

 

 

 

3D2005, SumperPi 32M, A64 Tweaker, Everest latency, load temps:

 

[Squid1]

You should change "939 pin Opterons DO NOT require registered memory" to

"as all other socket 939 CPUs, 939 pin Opterons do not work with registered/buffered memory, they require unregistered memory".

[ixtapalapaquetl]

148 CABNE 0528GPMW

 

DFI nF3 Ultra-D results below. This board's sensors report temperatures and voltages very differently than those on nF4's, so direct comparison of these values to nF4 chipsets should be viewed with appropriate skepticism.

 

 

Before/after IHS removal load temps:

 

 

Prime, 3D2001, Everest read:

 

 

3D2003, SuperPi 1M, Sandra, Everest write:

 

 

3D2005, SumperPi 32M, A64 Tweaker, Everest latency:

 

[dinos22]

i've got the middle of the road one......we missed out in Australia on CABNE CPUs

 

i've done a bit of testing on mine since i got it a couple of days ago

 

System Specifications:

Opteron 144 (CABYE 0536GPMW)

DFI nF4 SLI-D (BIOS 704-2BTA)

Corsair HC200 Water Cooling

2x512MB G.Skill LE TCCD 440

X800 434/1146

Wavemaster

Enermax 535W EPS SLi PSU

 

 

**I cannot get it to pass 1M at 3GHz............the highest i've done with 1M was 2970MHz..........

**i think the CPU is very sensitive to MAL & Read Preamble settings.

**I also haven't played much with drive strength settings and some RAM timings.....i think i should be able to do a 1M 3GHz with a bit of tweaking..............

**the CPU is normally very close in MHz stability between 1M, 8M and 32M times with all else equal........

**The CPU is definately hotter than Venice.......i think about 10C or so

**This time round i thought I'd try Ceramique thermal paste and see how that does....don't know if it contributes to higher temps at all

 

I can suicide on at least 3.1GHz......didn't try higher but see no issues

1.36V 2.7GHz 1M SPI

 

1.36V 2.7GHz 8M SPI

 

1.44V 2.8GHz 1M SPI

 

1.44V 2.8GHz 8M SPI

[dinos22]

1.5V 2.9GHz 1M SPI

 

1.52V 2.9GHz 8M SPI

 

i left it on CPU burn in last night for 5 hours with 1.54V 2.9GHz and did a 32M run with these settings now.....i'd say 1.52V would probably do the job as well.......it's a slooooooow time.....lots of tweaking ahead

 

1.54V 2.9GHz 32M SPI

 

My best 1M SuperPI time so far........all above benchmarks were initial tests to see where i'm at.......next round of stuff is memory testing and BEST SuperPI times

 

1.57V 2.935GHz 1M SPI

[Dave_Sz]

Those things sure do get HOT. Does anyone know if they are actually rated to handle higher temps or are they pretty much the same as venice cores with maybe 55* max?

[dinos22]

Those things sure do get HOT. Does anyone know if they are actually rated to handle higher temps or are they pretty much the same as venice cores with maybe 55* max?

yeah they are hotter than Venice and San Diego CPUs.....they make it up in OCing department tho

[ixtapalapaquetl]

Those things sure do get HOT. Does anyone know if they are actually rated to handle higher temps or are they pretty much the same as venice cores with maybe 55* max?

Yes, they are rated higher. Check out the max case temp:

 

 

I'm priming in the above image, and I'm on air, so you can see that the CABNE's behave a little better! I have seen Opteron's max case temp as high as 67C. Note: Vcore is 1.64 according to multimeter - my nF3 has a penchant for prevarication! I have a thermocouple, but haven't figured out how to jam it under the heatsink. Ideas most welcome.

[Squid1]

They are rated for 65 C, IIRC.

 

The heat is really not the problem. The high voltages reduces your lifespan independently of the heat.

[Dave_Sz]

Right, but at 1.52 volts that thing is running almost 8* higher than my 3000 at the same voltages. His temp rises 14-18* at load over idle while all my venice chips only rise maybe 10-12 max.

[rodan]

I like this thread, keep the 939 Opteron going.

Didn't read it all, but I can comment on air coolers. The xp90 runs 5c higher on

load, the 90 copper is better (5c lower) for what its worth. The SD's have

1M cache. I run my 3700 2.75g, 1.5vcore, 11x250x4x,1T, 2.5,4,8,3.

According to sensors, nf4 is 25/37c no/load.

 

How much lower is the vcore required for Opteron, I could probably do 2.9 on

the SD at 1.6 vcore. I did remove the ihs and tightened the retainer on the xp90c.

That alone lowered my temps 5c. This is running a total of 56cfm air through the

closed case. Outside the case, the heat sink is bone cold at these settings and

only 1 degree hotter inside the case. When I had the nf3 board it burned the

core (or the resin of as5) at 1.65 vcore. When switched boards the temps

were dramatically lower. I think these Opterons are based on the SD. Is that

true? My cpu is "in the wild too", its not CABGE, its something else. I got it late

in the SD production batch, 3 months ago. Nobody had my batch numbers.

 

Keep the info coming...

Is this info going to stay in the Amd overclocking thread or will a new 939 Opteron

category be created?

[ixtapalapaquetl]

Screenshots added to first post.