Phenom 9500 w/ MSI K9A2 Platinum

[KTE]

Power consumption for this system at stock

CPU 2.3GHz @ 1.192V, 1.2VID
NB 1.8GHz @ 1.25VID
HT 1.8GHz @ 1.1V
2x 1GB RAM 800 @ 2.2V
HSF 1289RPM @ 5V - 400mA
Sapphire HD 2600 XT 800/700 @ 1.26V/DK
2x SATA II WD Caviar SE 80GB
Antec Earthwatts 430W

At 28°C ambient.
Hibernate = 9W AC (probably at like 50% efficiency )
Standby = 65W AC
Idle (no CnQ) = 100W AC
Load (P95 large-FFT) = 175W AC
Load GPU (ATi Tool) = 150W AC
Max. software load = 172W AC w/ Everest FPU Julia

Ever wondered what loads the processor the most with stability software? Look at the above software I included, there's a very good hint. The stability testers with high power draw/heat creation are typically computations taking place inside the FPU units of the processor.

100% load in them is extremely rarely achieved in real-life work and 75% of that load is usually achieved by intensive gaming loads.

Temperatures

A few things I'd like to mention quickly with the new Penryn/Phenom strange temperatures around.
Typical case/heatsink ambient is not sub 30°C, it is 30°C-45°C and don't forget about the far East, South East Asia, Africa, Australia, Middle East, South Americas, many parts of EU/US, with 30-50°C summers. This isn't Canada or America, it's the World Wide Web (lol). I travel frequently to many destinations, so I'm aware what is realistic and what isn't for case temps. SoCal would boil for over 5 months plus 30°C inside temps, let alone what your ICs inside a case will give off and make the case ambient.

For instance. When Core 2 was released, I don't know how many of you remember and know the fact that while it needed a lower power consumption than P4 it also had a ~10°C (60-61.4°C) lower max CPU temperature limit (Tc), which meant you still needed the equivalent class cooling to keep it within Intel specs. For X6800/75W TDP it was 60.4°C max and the E6000/65W TDP 4MB lineup it was 60.1°C max (IIRC). For Phenom 9500/9600 95W TDP this limit is 70°C.

Hence, this is the thermal resistance your cooler needs to keep that TDP within AMDs operating specs assuming ambient heatsink is 28°C:

(70-28)/95 = 0.442°C/W

Let me translate that for you: it means the heatsink/CPU needs to be a maximum of 0.442°C hotter than the surrounding air for every 1Watt of heat being dissipated to stay within the thermal specifications of the processor.

We can check processor temperature accuracy to a certain degree this way you know. I need a minimum of that to stay below the CPU max safety limit. Well, let's throw in what my system/monitoring devices were telling me the CNPS9700 at lowest fanspeed was getting at 28°C ambient.

(Bja = (Tj — Ta)/Ph)

(41-28)/95 = 0.158°C/W

Seems about right that, since Zalman 9500 AT at low-full speed has a specified thermal resistance of 0.16-0.12°C/W.

I really wish someone with an "074# MPMW" that was just looking for stability would try these settings... At least for me, these seem to be pretty well tweaked... I call it 2.5.. hehe...

With a BE? That would be easy to replicate with mine.

One very good thing to know is, stability is not a factor of "consistent high static load", just like with PSUs it isn't, although that is also tested, but it's mainly a factor of "consistent fluctuating dynamic loads". This is more stressful on any individual circuitry (hence why power cycling degrades over time).

KTE... Just know that I would never doubt you Bro!!!
But I thought a TBL Error generated a "Machine Check Code"...

Wouldn't be anything new if I was misinformed...

I'm not sure myself, but a hard-lock doesn't have to give any error in actual fact.
Firstly, I've been doing these same tasks every day with the BE with the patched BIOS and nothing happened.
Secondly, like I said, P95 etc could not generate this at all and I stressed for around 28 hours at 2.3GHz/2.5GHz/2.4GHz.
Thirdly, what is it if not a CPU error, and if that, which other error do users around have at stock settings when they can run every bench/pass every stability test?

I'll refresh my memory up on TLB errata again though.
The processor switched off sending information and locked up, hence why the screen blacked out. I've never seen it happen before and I do know quite a bit about Windows workings. I honestly see this as the bug manifested.

Happened twice now in the same situation. Can't reproduce it when I'm looking for it but when I don't want it to occur. From my observations, after loading the same applications, it occurs after 35 or so minutes under dynamic/high loads when a very high load spike occurs, i.e., P95 in-place large FFT max value reached is 175W AC but during these workloads loads, the max it spiked at momentarily was 183W AC after which it fell to 175W AC constant lockup). My power lines are perfect, don't worry, +5V and +3.3V doesn't even fluctuate idle>load>idle and +12V fluctuates -0.03/0.04V from +12.01V to +11.97/11.98V to +12.01V idle>load>idle (on the 4-pin CPU +12V rail) and PSU exhaust grill temperature is 37°C.

well even with or with out the patch I would guess for it to be stable around 2.4 and almost the same as having a 4 core opteron at 2.85ghz (939 like scores IMO)
Any word on it the patch via bios or via software is better?
is there a software patch?

Not that I know of.

That's about what I figured because running 2 cores on both a 2.4 Phenom and a 2.8 5600 the Phenom was a tad faster.

Super Pi isn't anything to go by fully being only a memory intensive mathematical algorithm based on approximating the AGM, but can be used to show if there is any difference in the underlying core at equal speeds, since it is single threaded. For instance, Penryn (non-QX) finishes Spi1M @ 3601MHz in 12.797s and C2Q at the same settings does it in 14s. An efficiency value of 46,082 for Yorkfield verses 50,414 for Kentsfield.

X2 and Phenom @ 2640MHz, same multi/HT ref., RAM as far as possible;
X2 finishes in 33.125s and Phenom finishes in 28.671s. An efficiency value of 75,691 for K10 verses 87,450 for rev.G2 K8. In 8M the difference is 25seconds. So, there are improvements, just not enough to win an industry lead or level up with a very good core fore-running architecture (C2) from a very inefficient one (Netburst).

It also looks like we have a ceiling at around 2.4 - 2.5 from what I gather. I've seen other numbers floating around but from what I gather those 2.5-2.7G ocs are not completely stable.

My 9600 BE is 2530MHz as stable as you can have (nothing more though - I need the errata patch though and that makes it like a 9000, so it's best to run stock ).
My 9500 was 2750-2700MHz stable, this setting was the best and I ran/benched/tested it many many times at least for 4 weeks, and 2640MHz or better, 265x10 (2650MHz) (can't find ss) was every day at stock volts bootup and running fine. The problem with 2.7GHz was, I didn't like using 1.392V which is what I need for perfect stability and preferred 1.240V speeds.
My other two 9600 were not even 2500MHz stable. 1st is around 2400MHz and 2nd would be a little more, possibly around 2440MHz.

Let me explain why this is horror oc'ing> the thing with Phenom to note is this procedure which will be common for many:

*Get your high MHz, get your valids, get your benches, be happy you didn't get a dud, expect more because that was rushed. Then, stress test it until it passes your highest stable. I expect many to get 2.7GHz stable with a lucky shot (non BE).
*Then, restart after many hours whilst you think it's perfect and change the settings, to a lower V/MHz value. Run that for a day or two.
*Next go back in to oc'ing it at bootup with the same settings you had before or through trying AOD.
*Now, most of the time, you won't be able to replicate it, you might even not get a bootup at those settings anymore. Hence what we all have been saying since 2 months.
*Then, on a passing day while you've given up, you try a quick oc.... and you get 2.8GHz 3D benchable stock volts.
*You back down to 2750MHz and stress it to hell. It passes perfectly. You run it for a day until you reboot. You try those settings at bootup, no POST. You try higher volts at those settings, no POST. You try max everything incl. cooling but a 100MHz lower than those settings, no POST. You try 300MHz lower, no POST. You try 100MHz higher than stock, no POST.

Other times, pay attention here, it isn't the clock speed which messes everything up, it's high VID/high voltage (above 1.265VID and 1.35V ish). You can at many times get far higher stable at stock volts than higher than stock volts, I've seen it and so has Dave. At other times it's the clock generator which messes everything up by randomly throwing a core or the IMC at 100Mhz higher than current speed. And yet even more, at other times you get a perfect boot but ATi drivers screw everything up at the time they're supposed to be loading.

Hence why it is very difficult to say anything when someone says "whats max stable?" I've done 2772MHz on the 9500 2hr stability pass at 1.4V 1.2VID, but that wasn't possible after that one time.

OK so this is just very brief. Built the system today, using the Asus board (M3A32-MVP Deluxe), 2 sticks of DDR2-800 CL5, and a 3870 card. Booted to 2.6 Ghz, stock voltage, no problems was able to run several benchmarks (Cinbench R9.5 and R10, no lockups), ran Everest stability (CPU+FPU+Cache) for about 15 mintutes. Could boot to 2.7 GHz but could not run some benches completely, R9.5 completed but R10 did not.

I have not attempted to stress test for stability at this clock yet, so don't get too excited at 2.6 ... it is late, I will be posting CPUID shots, and validations in the next few days.

EDIT: Oh, yeah... using cruddy cooler, the AMD stock cooler that came with CPU and the pre-applied TIM. So 2.7 is almost certain, 2.8 likely (I hope). Oh yeah, BIOS is 503 -- dated 11/12/07. Asus did not implement TLB patch until 701 from what I understand.

Nice. Hope it passes benchmarking and stability OK, I'd like you to pick a MHz and stress test it first, then boot with it for a few days, then game and do some intensive real work. Just to eliminate and distinguish if your CPU behaves the same as what we're seeing -- or if what we're seeing maybe a MB thing (doubt it).
A few details we'd want from you though: stock CPU VID, stock NB VID, stock CPU VCore (idle/load), stock temps, power if you can, ambient temps, voltage/settings needed for booting,


TBVH with you all, I've backed out trying to oc 9600 BE since the stock volts 2530MHz, when I saw that it had 2.6GHz running but required volts/vid which made it pull 225W AC more than stock idling, and 347W AC more when in full load. My PSU would trip at that point so I didn't continue until I can change the PSU once again to something even more beefier. 430W is not enough for the amps available on the +12V with a low end card.

Look at it this way, the comparison;
Q6600 G0 95W TDP having 582m transistors in 286mm² at 65nm and Tc at 71°C -> x power consumption, x core voltage, x core current, x thermal resistance needed and x core temperatures with a Zalman 9700.
Phenom 9500/9600 95W TDP having 450m transistors in 285mm² at 65nm and Tc at 70°C -> tad higher power consumption, lower core voltage, similar current, same thermal resistance needed but lower core temperatures.
K8 rev.G2 5000+ 65W TDP having 154m transistors in 126mm² at 65nm and Tc 55-68°C.

So going from 65nm native dual to native quad, voltage has remained similar, transistor count is tripling, Tc has increased rather than decreased and TDP is 30W higher (with a 300MHz deficit).

Phenom 96500/9600 core requires 19A -> for 2.3GHz at 1.192/1.232V/1.240V.
X2 5000+ core requires 24A -> for 2.6GHz at 1.350V.
*Dual MCM would theoretically be 130W TDP for 2.6GHz.
Phenom 9900 is 140W TDP for 2.6GHz.

X2 5000+ BE, most cores can do 3300MHz 1.35V stable.

Speed:
3300/2600 = 27%
x by 2300 = 2967MHz

TDP:
65 x 3300/2600 = 82.5W TDP
95 x 2967/2300 = 122.55W TDP

If it could scale at stock amps/volts.
But it can't. So let's feed it the voltages and amps that X2 65nm needed for 3.3GHz.

*Amps:
24 x 4 = 96A
96 x 1.35V = 130W TDP

But at 130W with Phenom, you still only get ~2.4GHz whereas you'd have 3.3GHz with 2x rev.G2 X2 MCM.

Using figures of what 130W TDP K10 and 82W TDP X2 rev.G2 gives us respectively for clock speeds; MHz/Watt oc'd ->

82.5W x 2 = 165W TDP

Rev.G2: 40 MHz/W (2600) - 40 MHz/W (3300)
↑ MCM: 20 MHz/W (2600) - 20 MHz/W (3300)
K10: 24.2 MHz/W (2300) - 18.46 MHz/W (2400)

So to put it clearer for you all, K10 core 65nm compared to K8 core 65nm is worse in overclock/frequency at the same TDP.


I'm going to make a short list of the problems/bugs I'm experiencing which could possibly cause the odd processor behavior I'm seeing with this setup, and really messes the oc'ing:

-Cannot change NB FID lower than 9x/CPU FID above 12x through the BIOS: lockup.
-Cannot put high VID on NB/high VID on CPU through BIOS.
-Cannot have high VID/volt bootup.
-Cannot have low NB/CPU VID.
-Cannot have a plus 12.5x CPU FID even in Windows at low HT.
-Cannot boot plus 220HT reference at any volts.
-Cannot boot using NB/CPU DID.
-Cannot have 1066 RAM without errata patch.
-Random lockups.
-Random RD790 clock generator lockups.
-ATi display driver causing problems loading Windows (any of the last 4)
-ATi drivers make GPU load 50% at stock idle at desktop (thus fan is always running high and temps high - doesn't happen with X2/C2D/C2Q).


jonspd: Yep that's definitely possible. My chip should get to that with 12x NB multiplier (has done) .. but just take a look at the power consumption/load temps with those VIDs/Volts. It's scary.

Check AMD Power Monitor for your NB VID, it will be around 1.3 I think.