Phenom 9500 w/ MSI K9A2 Platinum

[justapost]

Are we looking at the right errata, 254 is a TLB errata, but I thought the patch in question dealt with 298, which has not been published yet.

Good point. AMD has posted the patch for errata 298 on the x86-64 mailing list here.
But looking at the source code the patch is a workaround for errata 254.
Can be it are two TLB related problems, whom can both be fixed just by disabling the TLB.

[KTE]

I think you gentleman in this thread should be commended for doing such a fine job with all of this! Although I will stick with my original assumption from what I see and read and say the the current Phenom is good for stock or perhaps tops 200mhz over rated speed that you can safely get away with.

Discount my results until I can pick up another since my chip is buggy at stock -> no boot, cannot up and down the NB and so on. This is supposed to be a stock option.

Phenom chips seem to like 10x multi: Max with stock VID NB at 9x (2178MHz)

http://valid.x86-secret.com/show_oc.php?id=298288

also is everyone aware that the watch dog trick works on this board?

I noticed it playing around the other day.

If you get bios lock and system want start you can hold insert when you power it up and start tabbing delete.

It will post saying INS what held and overclock failed but like 7 out of 10 you get back into bios with out having to clear and in some cases all settings are still there...

Works on my board @ times. Not always though but I hardly ever have to clear CMOS anyway but for the things that make it unbootable for 10 hours plus.

New bios out KTE

with A7376AMS.P0F

Quote:
1.Modify "Adjust AM2+ CPU Ratio" function to user friendly. 1.Modify "Adjust AM2 + CPU ratio function to user friendly.
2.Modify "DCT Unganged Mode" default to "Auto". 2.Modify "DCT Unganged fashion" default to "auto".
3.Fixed "Adjusted CPU Frequency" display incorrent when use AM2+ CPU. 3.Fixed "Adjusted CPU frequency display incorrent when use AM2 + CPU.
4.Added "Patch AMD TLB Error" item. 4.Added "patch AMD TLB Error" item.
5.Fixed can't adjust to x6 and x7 ratio when use AM2 CPU. 5.Fixed can not adjust to x6 and x7 use ratio when AM2 CPU.
6.Modify "Current DRAM Frequency" and "FSB/Memory Ratio" function code. 6.Modify "Current DRAM Frequency" and "FSB / Memory ratio function code.
7.Modify display memory channel function at POST when use AM2+ CPU. 7.Modify display memory channel function at POST use when AM2 + CPU.

Thanks for posting the details.
Apart from the TLB patch () that's great news! A lot of the things I asked for there, a lot of them. Thanks goes to Sami here.

Will check it out today when I can.

[justapost]

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
2x SATA II WD Caviar SE 80GB
Antec Earthwatts 430W

I miss the GFX, was the HD 2600 XT I guess.

I'll add values assuming 70W TDP for an dual core K10

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.

Phenom dual core requires 28A -> 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

70W X 2967MHz/2300MHz = 90,3W 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

Other way around:
28A x 2 (cores) = 56A
56A x 1,6v = 90,3W TDP !!

More likely:
56A x 1,4v = 78,4W

[KTE]

Working out K10 native quad and X2 dual MCM quad there Achim.

Yes if it was dual-core it would be as you stated. BUt that's different to what I worked out.

You don't need 1.6V for 3.3GHz on X2 5000 BE+. At 24A per core, many are getting 3.3GHz stock volts, I had mine at 3360MHz stock volts stable, which is 1.35V. If it's dual chips on one die they would both run the same at the same TDP. Like Kentsfield, the TDP is just double (65W x 2 = 130) for the MCM.

[justapost]

Working out K10 native quad and X2 dual MCM quad there Achim.

Yes if it was dual-core it would be as you stated. BUt that's different to what I worked out.

You don't need 1.6V for 3.3GHz on X2 5000 BE+. At 24A per core, many are getting 3.3GHz stock volts, I had mine at 3360MHz stock volts stable, which is 1.35V. If it's dual chips on one die they would both run the same at the same TDP. Like Kentsfield, the TDP is just double (65W x 2 = 130) for the MCM.

I know. Just want to add dual core k8/k10 values. The TDP calculations I made where for an theoretical overclocked dual core k10.
Looking at the dual core values I can understand why they focus on lower power consumption first, as the BE-2400 has a 45W TDP vs. the approximated 70W a dual-core K10 whould require atm.

[KTE]

Looking at the dual core values I can understand why they focus on lower power consumption first, as the BE-2400 has a 45W TDP vs. the approximated 70W a dual-core K10 whould require atm.

Yeah you don't need to tease me with mentioning my favorite daily chip there.

[justapost]

Yeah you don't need to tease me with mentioning my favorite daily chip there.

So still no BE-2350 in sight over there?
I appreciate your comparison as the 65W TDP cpu compares better vs. the phenom due to the similar TDP. I'll order an 5000BE this week for an office-rig and put my 3800EE in it instead.

[KTE]

Had it a few weeks ago, it was DOA (never booted) and I've not had it back since. My procedure is quite lengthy since I don't buy them myself so it's not me dealing with the supplier or retailer, it'll me the IT management at my uncles firm dealing with their distributor or usually AMD, and so it takes longer for me to get something since they cover many queries at once i.e., I drop it back there when I get time (500 miles away), they then work something out with their distributor/AMD, then retake, redelivery to them, I get informed by whoever finds out and I have to wait to drive 1000miles there and back again to pick it up.

BTW, AMDs SB600 gives higher ATA133 IDE performance than P965, P35. However the SATAII performance is worse on my board than the UDMA6 IDE HD.
In 32MB sequential read using HDTach, I get 20MB/s more for the IDE HD than the 2x SATAII HDs I've tried.

TDP compare: No problem.
My main point is, clock per watt, oc'd clock per watt, at the same fab process, one revision earlier, K8 core still clocks higher within the released Phenom TDPs, whereas K10 core stops at a low point. Hence there is something with the K10 core stopping the clocks and it cannot be just design materials (SOI) since it's this same SOI with much worse transistor performance/leakage that can produce higher clocks at the same TDP in K8.

[justapost]

Had it a few weeks ago, it was DOA (never booted) and I've not had it back since. My procedure is quite lengthy since I don't buy them myself so it's not me dealing with the supplier or retailer, it'll me the IT management at my uncles firm dealing with their distributor or usually AMD, and so it takes longer for me to get something since they cover many queries at once i.e., I drop it back there when I get time (500 miles away), they then work something out with their distributor/AMD, then retake, redelivery to them, I get informed by whoever finds out and I have to wait to drive 1000miles there and back again to pick it up.

What is wrong with sending it through the post?

BTW, AMDs SB600 gives higher ATA133 IDE performance than P965, P35. However the SATAII performance is worse on my board than the UDMA6 IDE HD.
In 32MB sequential read using HDTach, I get 20MB/s more for the IDE HD than the 2x SATAII HDs I've tried.

What is your average read speed? With an 500GB Sata II HD i get 56MB/s with my 8 year old 20GB IDE 12,5MB/s.

TDP compare: No problem.
My main point is, clock per watt, oc'd clock per watt, at the same fab process, one revision earlier, K8 core still clocks higher within the released Phenom TDPs, whereas K10 core stops at a low point. Hence there is something with the K10 core stopping the clocks and it cannot be just design materials (SOI) since it's this same SOI with much worse transistor performance/leakage that can produce higher clocks at the same TDP in K8.

They had problems with higher clocks on the first K8's in 65nm too. Can be there is a leakage problem with the L2/L3 cache as there are no 2MB L2 65nm k8's also.

[KTE]

What is wrong with sending it through the post?

I can't pickup new equipment that I find there this way.
I also miss out on viewing any latest tech. they have there and since they have many large departments you can get to see some unreleased products being demo'd or tested at times. Companies send their reps there for presentations and demo's. Also I like to work with the engineers and the professor that's head of lab there along with my uncle and see what they're doing because they have much more knowledge and experience than I do, as well as the right contacts, industry feedback and resources to test. You usually get to learn a lot my being there than by hearing it over the phone and I usually want to meet my uncle anyway. He's an exciting man.

What is your average read speed? With an 500GB Sata II HD i get 56MB/s with my 8 year old 20GB IDE 12,5MB/s.

SATAII-80GB: Burst 188MB/s - Av.R 60MB/s (CPU 1%)
SATAII-80GB: Burst 168MB/s - Av.R 53MB/s (CPU 2%)
UDMA6-80GB: Burst 89MB/s - Av.R 70MB/s (CPU 0%)

They had problems with higher clocks on the first K8's in 65nm too. Can be there is a leakage problem with the L2/L3 cache as there are no 2MB L2 65nm k8's also.

Yeah but K8 is inferior to K10 nFET/pFET transistor performance, these have improved current leakage over K8 rev.G2.

[jonspd]

KTE do you know if there is any information on the vreg stuff on the board. I was thinking on checking all my voltages via DMM. I would like to know what the readings for the following are via DMM compared to software

nb vid
ht vid
sb vid
vcore, and maybe vddim.

[KTE]

KTE do you know if there is any information on the vreg stuff on the board. I was thinking on checking all my voltages via DMM. I would like to know what the readings for the following are via DMM compared to software

nb vid
ht vid
sb vid
vcore, and maybe vddim.

VIDs you can't check by measurement since they're values placed within the core for P-States. They're not voltages. Voltages you can check.

I tried 2.6GHz by just multiplier (1.52V 1.45VID) and absolutely no way on my system (unlike what I said to you earlier). The limit is 2.53GHz with or without multis. My system failed bootup for 10 hours and I kept pencil modding the resistors to change VCore values. It was 1.12V.
After quick modding it only reached 1.19V max but that still did not get a boot. Super I/O chip at that time read wrong values compared to MOSFET measurements, it read 1.08VCore and 2.47VDIMM.

NB CPU voltage = IMC voltage.
NB voltage = RD790 voltage.

On K9A2 Plat. you can measure at the MOSFET legs before the DIMM slots to find the real VDIMM. You can do the same for the CPU but it's better done at the inductors. I haven't found where to measure NB voltage since I never needed to change it from stock and at stock it's at 1.1V. Also it's too cramped in there to measure anyway.

[Swanie]

Hi there i though the link at the bottom could be interesting reading in future but here is what this guy got with his Phenom:

Hi, I thought I would start a thread to gather some K9A2 Platinum specific tips and tricks from folks out there and share my own. I spent much of today pushing my CPU to the limit. I have been able to get it stable at a FSB setting of 240, CPU 1.45V, Mem 2.4V, NB 1.4V, HT 1.4V, HT Speed x5. CPU speed 2760MHz. Memory divider 1:1.66 yields DDR2-800 at this setting. CPU diode temperature at idle with Cool'n'Quiet enabled (throttles the voltage down to 1.3V and cuts the CPU speed in half at idle for each core) is 95F (35C), max load temp with 4x prime95 affinity large FFT is 122F (50C) with the Koolance on auto-fan (speed 3 out of 10) and the pump at about 1/3 speed. Pump at half speed + fan forced to speed 10, temp came down to 110F (43.3C).

http://forum.msi.com.tw/index.php?topic=113902.0

[KTE]

Hi there i though the link at the bottom could be interesting reading in future but here is what this guy got with his Phenom:

Hi, I thought I would start a thread to gather some K9A2 Platinum specific tips and tricks from folks out there and share my own. I spent much of today pushing my CPU to the limit. I have been able to get it stable at a FSB setting of 240, CPU 1.45V, Mem 2.4V, NB 1.4V, HT 1.4V, HT Speed x5. CPU speed 2760MHz. Memory divider 1:1.66 yields DDR2-800 at this setting. CPU diode temperature at idle with Cool'n'Quiet enabled (throttles the voltage down to 1.3V and cuts the CPU speed in half at idle for each core) is 95F (35C), max load temp with 4x prime95 affinity large FFT is 122F (50C) with the Koolance on auto-fan (speed 3 out of 10) and the pump at about 1/3 speed. Pump at half speed + fan forced to speed 10, temp came down to 110F (43.3C).

http://forum.msi.com.tw/index.php?topic=113902.0

People can still get that on water. He doesn't need high NB/HT/RAM volts and the BIOS he's using means 2.7GHz is like 2.3GHz, I hope someone tells him that. Thanks for the link though, all info helps.

[justapost]

I can't pickup new equipment that I find there this way.
I also miss out on viewing any latest tech. they have there and since they have many large departments you can get to see some unreleased products being demo'd or tested at times. Companies send their reps there for presentations and demo's. Also I like to work with the engineers and the professor that's head of lab there along with my uncle and see what they're doing because they have much more knowledge and experience than I do, as well as the right contacts, industry feedback and resources to test. You usually get to learn a lot my being there than by hearing it over the phone and I usually want to meet my uncle anyway. He's an exciting man.

Worked at a german car manufracture in an lab for virtual assembly manufacturing. We had an SGI Onxy II, datagloves and headsets with magentic tracing there. The sgi chefs themselfes stepped by and the chief developer of theit gfx system tought me tricks with the graphic cards. Beside that we had reps for new headsets, tracing and force feedback equipment there showing new stuff.
Unfortunately the lab got cancled together with the project cause it was unprofitable. So I can understand your causes.

SATAII-80GB: Burst 188MB/s - Av.R 60MB/s (CPU 1%)
SATAII-80GB: Burst 168MB/s - Av.R 53MB/s (CPU 2%)
UDMA6-80GB: Burst 89MB/s - Av.R 70MB/s (CPU 0%)

Must dig my benchmark tabulations think i have results from the m2a-vm. I have an asus barebone with an nvidia am2 mobo whom I can test.
I used non ahci mode here for sata II benchmarks, ahci locksup my system after a minute uptime, but that seems to be an sb600 issue.
But the UDMA6 results are really pretty good.

Yeah but K8 is inferior to K10 nFET/pFET transistor performance, these have improved current leakage over K8 rev.G2.

Ah ok, I expected k8's g2 stepping to be a step further. Do you know if those improvements will be used in further k8 steppings?

[dreaminblue]

New bios out KTE


with A7376AMS.P0F

Quote:
1.Modify "Adjust AM2+ CPU Ratio" function to user friendly. 1.Modify "Adjust AM2 + CPU ratio function to user friendly.
2.Modify "DCT Unganged Mode" default to "Auto". 2.Modify "DCT Unganged fashion" default to "auto".
3.Fixed "Adjusted CPU Frequency" display incorrent when use AM2+ CPU. 3.Fixed "Adjusted CPU frequency display incorrent when use AM2 + CPU.
4.Added "Patch AMD TLB Error" item. 4.Added "patch AMD TLB Error" item.
5.Fixed can't adjust to x6 and x7 ratio when use AM2 CPU. 5.Fixed can not adjust to x6 and x7 use ratio when AM2 CPU.
6.Modify "Current DRAM Frequency" and "FSB/Memory Ratio" function code. 6.Modify "Current DRAM Frequency" and "FSB / Memory ratio function code.
7.Modify display memory channel function at POST when use AM2+ CPU. 7.Modify display memory channel function at POST use when AM2 + CPU.

A heads up to anyone that decides to try the new beta bios A7376AMS.P0F, it auto set the ram to 1066mhz and i have 800mhz ram. I thought it was a bad flash untill i took one stick out and it did boot. It also would not save any settings. I tried setting the ram manually to 800mhz but it would then not reboot without a cmos reset. I finally got it working again by booting to usb and flashing back to the 1.13 bios. On the bright side their were some good new options, it allows you to enable or disable the tlb patch and change the multiplier without messing with p-states. Hopefully they will get the bugs worked out and release a similar official release.

[jonspd]

Just a update I'm managing to run prime at a setting that gives 11953mbs in sandra on some pc6400 gskill 2x1gbhz...

Also it's at 272fsb.

[Daveburt714]

A heads up to anyone that decides to try the new beta bios A7376AMS.P0F, it auto set the ram to 1066mhz and i have 800mhz ram. I thought it was a bad flash untill i took one stick out and it did boot. It also would not save any settings. I tried setting the ram manually to 800mhz but it would then not reboot without a cmos reset. I finally got it working again by booting to usb and flashing back to the 1.13 bios. On the bright side their were some good new options, it allows you to enable or disable the tlb patch and change the multiplier without messing with p-states. Hopefully they will get the bugs worked out and release a similar official release.

Thanks for the info Jon and Dreamin!! That sounds close to the Bios I've been waiting for, especially if it will totally disable the TLB patch...

My memory is ddr2/1000 I might actually give it a try if I'm feeling squirley later on tonight...

[tictac]

Just a update I'm managing to run prime at a setting that gives 11953mbs in sandra on some pc6400 gskill 2x1gbhz...

Also it's at 272fsb.

Any higher than that? 280MHz? 290MHz? 300MHz?

[Blacklash]

I may try another AMD chip that overclocks better. Phenom overclocking is just depressing. Not to mention, I have been hanging around in the Intel section where folks are getting 4.7GHz with the new E8500s.

[jonspd]

http://www.jonspd.com/jonspd_uploads...nies/test2.GIF

also ran 3dmark05 and got a 16428 but it froze on the desktop idling there after.


On stock cooler... awaiting mounting hardware for a xp-90c

[KTE]

I may try another AMD chip that overclocks better. Phenom overclocking is just depressing. Not to mention, I have been hanging around in the Intel section where folks are getting 4.7GHz with the new E8500s.

While C2 or the QX oc's far better air/water (there's no doubt about it since long) with retail <E8400/Q9550, I doubt it. Not what I'm seeing at all, stuck sub 3.7-4GHz <1.5V. I'll pick up a retail E8400 very soon if not this weekend anyway. Intel started the games again with this commanding position of offering crap locked multis even though the chips are capable of more with a higher multi. But E8500 here is no where found, none of them are yet and the E8500 is 200USD more than the 9500 here. <E8400 will be same price as 9500 here.

http://www.jonspd.com/jonspd_uploads...nies/test2.GIF

Have you never tried max HT of the chip yet?
272MHz is very high. My chip did 270ish max and it ran 265HT perfectly at 10x, so try for more. Maybe you need better cooling.

[jonspd]

Nope I haven't. Maybe I should try a 9x muilti tomorrow. Yes my temps do concern me at this point on air. I did however manage to run alittle lower vcore.


off to bed and fold for a while gotta turn in a unit tom sometime also got the F@H 2653 smp wu's down under 12min a frame.

[Daveburt714]

I may try another AMD chip that overclocks better. Phenom overclocking is just depressing. Not to mention, I have been hanging around in the Intel section where folks are getting 4.7GHz with the new E8500s.

You know what... Phenom is mysterious, I've had it for almost 3 weeks, I've kinda treated it like a new car. I'll be nice to it most times, but I'm not afraid to take it out on the highway either, and I'm not gonna Drive 55!

I'm probably crazy... But I always feel like theres that little something missing that will make this chip shine, whether it be Bios, OS, Drivers, Software, SB700 or my Tweaking ability....

I can't find the link right now, but I think it was Tom's Hardware (Don't Hate!) that showed the Phenom kicking butt on UT3, seems like the programmers spent alot of time on multi-thread for that game and it did really well on Phenom.....

But your absolutely right.. That 5000 BE is looking pretty tempting for $99!

Bet I could hit 3.5 on that bad boy....

[Daveburt714]

Nope I haven't. Maybe I should try a 9x muilti tomorrow. Yes my temps do concern me at this point on air. I did however manage to run alittle lower vcore.


off to bed and fold for a while gotta turn in a unit tom sometime also got the F@H 2653 smp wu's down under 12min a frame.

So your running 272 HT with an 8x NB, WOW..

Sounds like you got a good one Man!! What was that stepping again?