Here's a little teaser....

[Sparky]

Yes it is good but the screenshot above it seems to contradict it if I'm reading it correctly

[rogueagent6]

Yep, Yep!!
Looks like Barcelona's bitten the BIG 1....
So much for AMD!

And you're basing this all on a sandra screenshot of all things? That's bright.

[cky2k6]

And you're basing this all on a sandra screenshot of all things? That's bright.

with a terribly buggy bios at that...

[KTE]

Thanks for the effort guys! Must be crazy getting these working...

BTW fellas (everyone else), the results weren't amazing. The results were poor actually and obviously at least some were very buggy too. Here's what I've heard:
4 steppings so far - B0>B1>BA>B2 BA has the B1 fix. I'm hearing there should be only two steppings out to reviewers, the last two. Can you guys please run CPU-Z?

Dave I know you said you have a B1 but I'd like to see what CPU-Z reports according to core register values please. wPrime reports false frequencies many times, and usually I've seen it report BIOS frequencies and not current CPU frequency. But CPU-Z will report the current frequency given by PLLs/core. If you have buggy/faulty or unsyncd MB PLLs or BIOS, the frequencies will change and be messy (but CPU-Z will report that real-time).

What frequency do you guys get if you run this utility? http://web.inter.nl.net/hcc/J.Steunebrink/chkcpu.htm

I think I know how to recognize the stepping in CPU-Z:



F-2-2 : B2 core
F-2-A : BA core

That's as far as I can work out from AMD K10 PDFs. I know for sure that above screenshot is a B2 core BTW.

s7e9h3n/dave_graham: There's also something seriously wrong with the results/info there, your mem bandwidth and conflicting info in Sandra for one and then comparing Daves results around with those before, like those on Coolaler forums- they all differ. For reference, I ran this about 5 months back on a single core 3.4GHz P4 DDR400.


An Conroe 2.4GHz nicely beats that score. How can these quads get 5.9 in WEI if the encryption/decryption is so slow while very similar to one of the tests ran to determine the end score, and that P4 doesnt even get 3.8 in WEI?

Dave, the 1573KB/s for WinRAR... well E6320 at 2330MHz gets 1302KB/s.

Also, Steven, your memory bandwidth score was the same as a Pentium 4 820 D on Intel 955X chipset with CL5 2x1GB PC2-5300 RAM.
AND your cache/memory speed was the same as a P4 3.4GHz while your bandwidth was less than a 2.4GHz Q6600. For reference, the Q6600 2.4GHz on Intel BadAxe2 2x1GB PC2-5300 gets 41502MB/s.

And Dave, that wPrime....

8 total threads: 32m: 11.406 seconds - 1024m: 364.062

4 total threads: 32m: 23.015 seconds - 1024m: 742.875

1 total thread: 32m: 84.578 seconds

Coolaler forums and previous told scores (IIRC it was Moviemans Xeons):

2GHz Opteron 2332: 32M Test 10.641s (8 cores)
2GHz Opteron 2332: 1024M Test 327.437s (8 cores)
2.66GHz Xeon X5355: 32M Test 8.203s (8 cores)
2.66GHz Xeon X5355: 1024M Test 243.812s (8 cores)

WOW this is all over the shore .... What Gary of AT said 30-08 comes to mind:

"The latest Barcelona chips are B02 steppings with one more to go. Believe me, the reason we did not post any numbers at Computex or since then is the simple fact that the CPU/boards/BIOS have undergone dramatic changes over the course of the summer. If you have an earlier stepping there is a very good chance that HT and the secondary cache is disabled, this will affect the benchmarks dramatically. We expect to see final stepping chips and board revisions early next week, until then, it is all speculation for the most part.

The one caveat that I will add, this chip really does not get into a groove until you get over 2.4GHz and then it scales incredibly well. Also, the first RD790 boards we have will undergo another spin so any Phenom results with those boards are subject to interpretation depending on whether you like AMD or not."

Keep it up! Lets hope some decent BIOS and fixes are made fast.

[planoru]

Oh yeah, forgot to add that there seems to be a problem with the memory detection as well. For some reason or another the 4x2gb sticks of PC5300 are running @ 159mhz SINGLE CHANNEL CL5. That's have a HUGE impact on performance. If anybody's curious, @ 800Mhz on the cpu and 159mhz on the mem, I can run a 39sec Spi 1m LOL

So the cpu is set at 800MHz with the memory @159 and single channel cas5? I understand this well? This means that is really fast, even in Spi. If it would scale just as the K8 this should get it to a max of 30 sec with cpu @2GHz and memory @667/128bits. The platform is really buggy.
Thanks for sharing!

[Piotrsama]

So the cpu is set at 800MHz with the memory @159 and single channel cas5? I understand this well?

Later on, he posted this:

Oops....I was assuming one of the logical procs in socket #1 was running the bench, when in reality the process initiates with cpu1 and then is immediately handed off to cpu6 for some odd reason. Bound the bench to Cpu5 and it results in a 1m ranging from 39s - 41s.

SuperPI was running on a 2 ghz core.

[m411b]

WAIT WAIT.....are you being sarcastic XC? lol
I hope so cuz aint that cache and memory bench like AMAZING?...or am I reading the numbers all wrong...lol
Im confused!!!!
I think that bottom screenie is AMAZING............

LOL, your the only one that understood that I guess!
AMAZING isn't the word for those numbers...

[m411b]

And you're basing this all on a sandra screenshot of all things? That's bright.

You're right I am basing that off of one Sandra screen shot. Those numbers are huge, and I was being Sarcastic. Didn't realize how serious peeps in this thread are getting, sry.

[s7e9h3n]

Can you guys please run CPU-Z?

This is what happens when I try to validate it:

wPrime reports false frequencies many times, and usually I've seen it report BIOS frequencies and not current CPU frequency. But CPU-Z will report the current frequency given by PLLs/core. If you have buggy/faulty or unsyncd MB PLLs or BIOS, the frequencies will change and be messy (but CPU-Z will report that real-time).

False or True Frequency? Answer below



I finally figured out what's REALLY going on with this rig. Thanks to AMD's monitoring software, it became quite apparent what's happening with the cores. Somehow, either the PLL or the bios has screwed around with the P-state of Core0, CPU1. A setting in bios allows the cpu's to enter the OS either in maximum performance levels or in power saving mode. I have it set for max performance. It works as it should, except for ONE minor little bug which has resulted in one of the more interesting screenshots I've ever seen:



Core0 is STUCK @ 800mhz in max power-saving mode (as indicated by the bars on its left) while all the 7 other cores run @ 2G. The problem isn't with the cpu's since it happens only with Core0, CPU1 even when I swap sockets. I'm still wondering how it's possible to have only one core running continuously @ 800mhz and 3 @ 2ghz without crashing the OS. (and YES that core is not being mis-read as a Spi1m with affinity set to Core0 turns out a time of >1minute 30 seconds ) This one's definitely going to have to be swapped out for a replacement......

[PhilDoc]

You did just fine, I think most of us realize you were being sarcastic.

[s7e9h3n]

Also, Steven, your memory bandwidth score was the same as a Pentium 4 820 D on Intel 955X chipset with CL5 2x1GB PC2-5300 RAM.
AND your cache/memory speed was the same as a P4 3.4GHz while your bandwidth was less than a 2.4GHz Q6600. For reference, the Q6600 2.4GHz on Intel BadAxe2 2x1GB PC2-5300 gets 41502MB/s.

Yep, there was something definitely wrong with the bandwidth score. Check out the bandwidth efficiency -> 2%. Octal Opties are capable of some outrageous bandwidth numbers if everything is running properly. Here's what my quad 2218's (2 x 2218 DC) did on the Asus L1N64 (they're capable of more but I haven't benched them hard)....

[tictac]

have you tried crystalcpuid? can you change multiplier with this tools?
http://dl.crystaldew.info/download/C...PUID413x64.zip

look weird core 0 running at 800mhz while 3 other core running at 2Gig

[s7e9h3n]

have you tried crystalcpuid? can you change multiplier with this tools?
http://dl.crystaldew.info/download/C...PUID413x64.zip

look weird core 0 running at 800mhz while 3 other core running at 2Gig

Nope, haven't tried that, but I can't imagine it would work as core0 seems to be totally locked into that P-state. I think either the bios may be corrupted or the PLL is borked on this board....

Yes, it really does look weird but it runs even weirder

[Kenetixx]

Thats some strange happening you have going there ??????
But good work showing us all this, Cheers

[Lightman]

This is what happens when I try to validate it:





False or True Frequency? Answer below



I finally figured out what's REALLY going on with this rig. Thanks to AMD's monitoring software, it became quite apparent what's happening with the cores. Somehow, either the PLL or the bios has screwed around with the P-state of Core0, CPU1. A setting in bios allows the cpu's to enter the OS either in maximum performance levels or in power saving mode. I have it set for max performance. It works as it should, except for ONE minor little bug which has resulted in one of the more interesting screenshots I've ever seen:



Core0 is STUCK @ 800mhz in max power-saving mode (as indicated by the bars on its left) while all the 7 other cores run @ 2G. The problem isn't with the cpu's since it happens only with Core0, CPU1 even when I swap sockets. I'm still wondering how it's possible to have only one core running continuously @ 800mhz and 3 @ 2ghz without crashing the OS. (and YES that core is not being mis-read as a Spi1m with affinity set to Core0 turns out a time of >1minute 30 seconds ) This one's definitely going to have to be swapped out for a replacement......

Se here on XS we could see new AMD feature called independent core clocking in action thanks to small BIOS bug .

Look at this from the other side s7e9h3n, your saving energy!

[justapost]

This is what happens when I try to validate it:

FYI, i can not validate a s3992 board with 2210's with CPU-Z, seems to be more related to the board's chipset.

Yep, there was something definitely wrong with the bandwidth score. Check out the bandwidth efficiency -> 2%. Octal Opties are capable of some outrageous bandwidth numbers if everything is running properly. Here's what my quad 2218's (2 x 2218 DC) did on the Asus L1N64 (they're capable of more but I haven't benched them hard)....

These are amazing results.
With Numa enabled i get 10559 MB/s(int) and 10560 MB/s (float) on the s3992 with two 2210's(1,8 GHZ) and 4GB Reg/ECC Ram running at ~300MHZ.
Without Numa 6013 MB/s (int) and 6004 MB/s (float).

[Motiv]

Not sure but will this be any help with the oddly behaving CPUs

http://www.amd.com/us-en/Processors/..._15258,00.html

[tictac]

Nope, haven't tried that, but I can't imagine it would work as core0 seems to be totally locked into that P-state. I think either the bios may be corrupted or the PLL is borked on this board....

Yes, it really does look weird but it runs even weirder

That tools got a CPU MSR Editor..
in Function tab..
but first beside the cpu name change it to the Processor node 0, core 0

if it is P-State problem...

then open MSR Editor
Read register : C0010062
Bit : 2 - 1 (control the P-State)

set it to 0 (hexadecimal)
0h = P-State0 (Maximum performance)

i hope it helps...

[stealth17]

Things get more interesting by the day. I think it is safe to say that the bios is so buggy and the boards suck so much still that it too early to make any concrete conclusions about Barcelona.

As for the Core 0 issue, this might be pretty cool actually because if that is true, perhaps it is possible to clock each core independently? That would be pretty fricken sweet for when you are pushing balls to the wall and one core is holding you back. Though I'm not sure how that would effect stability or anything....

Thanks for all your hard work and thanks for sharing your results so far.

[dave_graham]

Things get more interesting by the day. I think it is safe to say that the bios is so buggy and the boards suck so much still that it too early to make any concrete conclusions about Barcelona.

As for the Core 0 issue, this might be pretty cool actually because if that is true, perhaps it is possible to clock each core independently? That would be pretty fricken sweet for when you are pushing balls to the wall and one core is holding you back. Though I'm not sure how that would effect stability or anything....

Thanks for all your hard work and thanks for sharing your results so far.

Barcelona allows for each core to clock independent of each other. this was present in early design docs and is still true to this day. to take advantage of this, however, you definitely need the BIOS to support it and the processor driver within windows. Again, this is much more interesting test metric that, should S7 and I be able to figure out a good benchmark for it, would definitely provide how well the Barc. is able to do in power savings over Clovertown.

this has been confirmed by AMD.

DDR3 is not present on the cpu and will require a repackage of the core if/when it arrives (Dozer).

hope that clarifies a few things.

dave

[freeloader]

Very good stuff guys. Keep up the good work.

I really hope AMD irons out problems (if they're with the CPU) before they transition to DDR3 and 45nm.

[PhilDoc]

Barcelona allows for each core to clock independent of each other. this was present in early design docs and is still true to this day. to take advantage of this, however, you definitely need the BIOS to support it and the processor driver within windows. Again, this is much more interesting test metric that, should S7 and I be able to figure out a good benchmark for it, would definitely provide how well the Barc. is able to do in power savings over Clovertown.

this has been confirmed by AMD.

DDR3 is not present on the cpu and will require a repackage of the core if/when it arrives (Dozer).

hope that clarifies a few things.

dave

Yes it does, thanks Dave

Phil

[doompc]

Stephen, try changing p-states with RM Clock:
http://cpu.rightmark.org/download.shtml

Dave, could you run SuperPI to end the PI question?
http://www.xtremesystems.com/pi/

[dave_graham]

Stephen, try changing p-states with RM Clock:
http://cpu.rightmark.org/download.shtml

Dave, could you run SuperPI to end the PI question?
http://www.xtremesystems.com/pi/

my running superpi won't end the PI questinos since i'm running B1's on a single plane board. that's not a fair test to ANYONE.

cheers,

dave

[dave_graham]

Very good stuff guys. Keep up the good work.

I really hope AMD irons out problems (if they're with the CPU) before they transition to DDR3 and 45nm.

this really isn't an AMD problem. it's a very poorly handled platform issue.

cheers,

dave