Phenom 9500 w/ MSI K9A2 Platinum

[Manicdan]

im confused, how does using a phenom cpu make ram timings better? im able to hit 490 stable at 4-4-4-12 with my really cheap patriots 6400 at 2.3v. im scared to see what they could do with your cpu

[KTE]

im confused, how does using a phenom cpu make ram timings better? im able to hit 490 stable at 4-4-4-12 with my really cheap patriots 6400 at 2.3v. im scared to see what they could do with your cpu

It won't do really.
What you get with an X2 is v.close if not exact what you'll get with Phenom with respect to RAM timings.

The RAM I'm using is very good. It constently hits 1200 4-4-4-4 2.2V with Intel CPUs.

[Daveburt714]

KTE, are you still running the 1.13 bios?

I'm really starting to wonder if I have a bad board/chip... When I was having problems last night I set every thing back to stock, it booted into windows and then locked up as soon as I went to open AOD... Guess it could have been a problem with AOD because I lost it and had to delete/reinstall...

I've been OC'ing for years and I've just never seen such odd behavior from a chip before... One time it will crank and next time it won't boot...

Oh well, guess I'll try MSI's 1.1 bios (board came with 1.0) when I get off work...

[Daveburt714]

I'll let you know the basic Phenom pattern.

If you want similar overclocks to Mr.Q:

Get >1.25VID
VCore +1.5V
Low NB/HT

If I've read things correctly, upping the CPU Vid from 1.2 to 1.25 increases the current to the chip right? (I haven't touched that yet...)
Actually, I don't think I've had Vcore over 1.4...

I'm not trying to be a wimp, but doesn't that seem like alot of power for a 65nm chip? My Opty 185 (90nm) only needs 1.43v's to run 3.0Ghz....
Maybe I'm not Xtreme enough for this forum...

All kidding aside, I bought this Phenom to play with, and my 3Ghz Opty is still sitting here if things go really bad....

I've only been playing with it for a few days, but it seems like keeping NB/HTT clocks close to each other gives better results, ie:same multi (of course I could be totally wrong!)......

As long as I keep them both around 1800, would that be OK.. or should I lower them even more? Thanks for any advice in advance!!

BTW, I decided to stay with the 1.113 bios for now...

[Brother Esau]

@ Tye hey what are you're thoughts brother Phenom 9500 they seem to do as well as the next as it would seem until B3 comes to bat and they are $189.00 on the EGG or is the 9600 any better probably not right?

[KTE]

KTE, are you still running the 1.13 bios?

Yep. Trying some new ones soon.

I'm really starting to wonder if I have a bad board/chip... When I was having problems last night I set every thing back to stock, it booted into windows and then locked up as soon as I went to open AOD... Guess it could have been a problem with AOD because I lost it and had to delete/reinstall...

Lockups when opening AOD occur when its trying to initialize the 790FX clock generator. That's an AOD problem it seems, I had it myself at first. Could be the board too though.
Check your AOD log files for errors.

I've been OC'ing for years and I've just never seen such odd behavior from a chip before... One time it will crank and next time it won't boot...

Same here.
http://www.xtremesystems.org/forums/...2&postcount=23

Oh well, guess I'll try MSI's 1.1 bios (board came with 1.0) when I get off work...

1.1 is better than 1.0, it was much better for me anyway.

If I've read things correctly, upping the CPU Vid from 1.2 to 1.25 increases the current to the chip right?

It should be 1.25VID stock, but at times the chip automatically decreases VID. Yep, it decides P-States, which decide amps fed to to each core as well and thus stability.

I'm not trying to be a wimp, but doesn't that seem like alot of power for a 65nm chip? My Opty 185 (90nm) only needs 1.43v's to run 3.0Ghz....

True, many K8s oc'd far higher than this but K10 isn't a mature core, the quad core manufacturing is new and unmatured and this is a quad at the same node as K8 was, so it can't get much lower. Especially with native quad, bins are no where near as good as dual-core and in decent yeilds as dual MCM will be. I didn't expect higher, but 1.45V 3GHz should still be achievable on some K10s. The problem isn't a power/volts one the way I see it but a core design problem.

1.25VID is stock though, giving around 19A per core at 1.232V which is within the 95W TDP. So it is still good, stock CPU power draw is not bad. There's many screaming it's this or that. None can run a few simple tests and provide me feedback to compare with themselves since online reviews are just so differing in variables and tenbd to take things too far in extreme tangents than what we want to know.

I've only been playing with it for a few days, but it seems like keeping NB/HTT clocks close to each other gives better results, ie:same multi (of course I could be totally wrong!)......

Sounds good to me.

As long as I keep them both around 1800, would that be OK.. or should I lower them even more?

1800 after oc'ing is fine, even 1900 is fine in many cases, but not before oc'ing as oc will make it go higher up quite obviously.

@ Tye hey what are you're thoughts brother Phenom 9500 they seem to do as well as the next as it would seem until B3 comes to bat and they are $189.00 on the EGG

Phenom is doing well in EU, don't know about US. Price is much cheaper in US all round than in EU for devices. I sold mine cheaper than you can get an used E6750 here. Price drops are decent around the EU for Phenom but not much on Kentsfields at all so the difference increases and most people feel about what they have to pay there and then and the damage it does to their pockets rather than long term or logically. It still occupies a market niche, a segment orientated specific displacement but you can't say it competes with Core 2 offerings at stock vs stock.

[blargman]

anyone try 1.24 yet? I'm gonna try it when i get home.

[Daveburt714]

Let me us know how they work for you Blargman!!
MSI seems really slow at releasing official Bios....

If you could post a link or attatchment that would be Killer...

Thanks. Dave

[KTE]

Have a new CPU since a day, but don't have the board with me to try. Getting back to flat on Wednesday.

[macci]

MSI board and Black Edition 9600 (and 800SPD DIMMs) - easy clock:





I'll post some Sandra screens next

[Daveburt714]

Those NB/HT bus speeds are Impressive Macci!!!

I didn't realize the NB Multis would be unlocked too.... I can set mine above 9x in p-states, and it will show it as higher in AOD, but it won't actually apply the settings.....

Or do you have some double secret probationary bios?

Edit: Actually it will take the higher Multi, but it locks my HT Ref clock...

[KTE]

Thanks for the peek Sammi

12k Sandra?

That BIOS is needed here.. any help?

With unlocked VIDs/MUltis, this could be an adventure if you know what you're doing.
And throw those timings 4-4-4-4 please, we want to see some latency/bandwidth over K8 here.

[Brother Esau]

macci.....have you taken a look at the DFI 790FX-M2R Yet if so can you explain some of the HT Settings?

@Tye......you get you're BE yet brother?

[KTE]

@Tye......you get you're BE yet brother?

Got it yesterday.
http://www.xtremesystems.org/forums/...&postcount=488

[macci]

same clocks as in the earlier pic:

Sandra:


and then Sandra but with just lower NB clock (2320MHz vs. 2532MHz in the 1st pic):


And same with Everest - notice which tests are affected by the NB clock:

2532NB


2320NB

[Brother Esau]

How is it that Sandra is Higher than Everest its usually the other way around?

[Lightman]

How is it that Sandra is Higher than Everest its usually the other way around?

Because Everest is still testing memory subsytem by loading only one CORE of CPU.
New SP1 of Sandra has reworked testing routines and it's loading all available CORES of CPU thus Phenom memory controller is fully utilized.

[Brother Esau]

Is this thing worth getting?

[vaughanster]

same clocks as in the earlier pic:

Sandra:


and then Sandra but with just lower NB clock (2320MHz vs. 2532MHz in the 1st pic):


And same with Everest - notice which tests are affected by the NB clock:

2532NB


2320NB

Nice to see what a cherry-picked golden sample can do in the hands of an engineer, but what about in reality?

[Brother Esau]

Not to soumd insulting towards Sammi but those were my thoughts as well

[KeZzZu]

Well nice to see some BE action but we need some more details ? new bios for msi? retail cpu?


some mem action with msi mobo for me it's quite impressive thought :P


BTW cpu z only show's correctly cpu speed ... mems are 1:2 divider nb somewhere near 1500mhz or 1750mhz and ht multi 5x

[KTE]

Nice to see what a cherry-picked golden sample can do in the hands of an engineer, but what about in reality?

If you don't trust that for whatever reason, then wait till I or someone else here tests one.

some mem action with msi mobo for me it's quite impressive thought :P

Have you applied the bug patch? Your memory perf and latency is quite bad for 890 and those latencies.
But that maybe because you have a low NB clock.

Also, everyone should check this Phenom in-detail review out! I think this is the first one I really enjoyed for what they tried, dealt with, methods, investigated and wrote in regards.

1st Jan: http://www.lostcircuits.com/cpu/amd_phenom

Contents covered:
-V8, Penryn, Conroe, Pentium, 4x4, X2 (65nm/90nm) all compared.
-Int, FP, gaming benchmarks
-Effect of the AOD "magic button" across benches
-Effect of 1066 RAM over 800
-Power consumption in all states and vitaly when overclocking
-Power efficiency @ workloads
-Virtualization and TLB errata and patch (explanation)
-Effects of memory bandwidth
-Effects of NB clocks

Some things are a little inaccurate in there (one comment IIRC- will talk about it later) but the rest is v.good IMO. Pay special attention to how its shown that oc a lower VID/Amp/Volt binned retail CPU usually gets much lower CPU power consumption than one retailed at that Speed/Volts. This is the same with Pentium, Conroe, A64, K10 and Penryn across the board. Since MFG stability requires higher amps per core and thus higher TDPs/power than customer oc methods. This is why we with 9500/9600 were seeing much lower power consumption figures than to-be retail 9700 and 9900.

[Lightman]

I've read it yesterday. Great article and from power consumption observations it is clear that 9500/9600 are using more power oriented transistors where 9700/9900 and Eng. Samples are using fast transistors. This also explains why Eng. Samples are easier to get to 3GHz. AMD need to improve 65nm process for better speeds/power.

[KeZzZu]

If you don't trust that for whatever reason, then wait till I or someone else here tests one.

Have you applied the bug patch? Your memory perf and latency is quite bad for 890 and those latencies.
But that maybe because you have a low NB clock.

KTE no i havent, and mems are running @ 500mhz :P 2.3v for those but they can run 520 with those timings

but remember i did have very low nb speed on that test, i can rerun with 2ghz nb speed

[gareth170]

i got my 9500 to 2.5ghz on 1.3v. so im guessing if i up the voltage to 1.35v i could maybe get to 2.7ghz