PDA

View Full Version : AMD Shanghai Opteron CPUs On Linux


phoronix
01-02-2009, 08:20 AM
Phoronix: AMD Shanghai Opteron CPUs On Linux

Last quarter AMD introduced their "Shanghai" Opteron processors that join the ranks of Intel's Harpertown Xeon CPUs that are 45nm quad-core server/workstation parts. Initial reviews of these new AMD Opteron processors have been very positive, but how do these chips perform with Linux? In this article we have our hands on a few of the fastest Shanghais, the AMD Opteron 2384 clocked at 2.7GHz, as we see how well they compare to the older "Barcelona" Opteron processors.

http://www.phoronix.com/vr.php?view=13345

d2kx
01-02-2009, 10:06 AM
It's 6 days until the desktop part, the "Phenom II", launches. Just in case someone is interested. I know I am.

Game_boy
01-02-2009, 01:07 PM
Based on the single game test (and the fact that Shanghai is the same silion as the new Phenom II Deneb), the new Phenoms will be 10% faster clock for clock. That would put the 3.0GHz Phenom II against the 2.83GHz Q9550 at $319 and the 2.8GHz against the 2.66GHz Q9400 at $266. Not bad.

Of course, the AM3 Phenoms in February should be slightly faster (~5%) if memory is the bottleneck.

Thetargos
01-02-2009, 05:17 PM
Interesting findings indeed, alas, without proprer "contrast" (AKA Intel parts performance) these results only show that indeed the Shanghai cores are an overall improvement from the Barcelona parts, which is good and all, but to add perspective, it would be better to have Intel numbers as well (for both present and past generation, as well)

deanjo
01-02-2009, 05:23 PM
Interesting findings indeed, alas, without proprer "contrast" (AKA Intel parts performance) these results only show that indeed the Shanghai cores are an overall improvement from the Barcelona parts, which is good and all, but to add perspective, it would be better to have Intel numbers as well (for both present and past generation, as well)

On that note though doing a head to head on such systems should really have more tests relevant to what such a setup would be used for which would be typically server use. Right now PTS is a bit bare in that department. I really doubt for example mp3 encoding will be a big selling feature of server chips where as apache performance, many thread dbase queries, power usage, vitualization etc would be more a concern.

bridgman
01-02-2009, 06:09 PM
Yeah... right now the benchmarks all seem to say "AMD is faster on a bunch of things, and Intel is faster on a bunch of different things, and the better CPU depends on which things you think are most important".

Bah !! :D

Thetargos
01-03-2009, 01:36 AM
On that note though doing a head to head on such systems should really have more tests relevant to what such a setup would be used for which would be typically server use. Right now PTS is a bit bare in that department. I really doubt for example mp3 encoding will be a big selling feature of server chips where as apache performance, many thread dbase queries, power usage, vitualization etc would be more a concern.

For the current Opteron Shanghai cores, yes. But the desktop Phenom II CPUs will use virtually the same cores (with a few changes, like no ECC memory requirement and faster RAM). This, though, should give perspective on what could we expect the performance be like for the Desktop versions. Like Bridgman says, currently the tendency to determine he "best" CPU seems to be "what do you want to do with it?" question which might still be relevant for users who simply want a "general purpose" desktop PC.

deanjo
01-03-2009, 02:04 AM
For the current Opteron Shanghai cores, yes. But the desktop Phenom II CPUs will use virtually the same cores (with a few changes, like no ECC memory requirement and faster RAM). This, though, should give perspective on what could we expect the performance be like for the Desktop versions. Like Bridgman says, currently the tendency to determine he "best" CPU seems to be "what do you want to do with it?" question which might still be relevant for users who simply want a "general purpose" desktop PC.

In that case then you could probably go by the many Phenom II benchmarks already out there for those things or simply wait until the 8th.

BlueKoala
01-04-2009, 07:54 PM
I think it would have been more interesting to see a 2356 OC'd to 2.7ghz in this comparison so we could have a better perspective on what the clock for clock improvements are. In any case, I'm very enthusiastic about these new cores :)

d2kx
01-08-2009, 04:36 AM
So, my new AMD Phenom II X4 940 system using an AMD 790GX/SB750 motherboard and an AMD Radeon HD 4850 1GB is arriving soon :D Could have been here already if I didn't do a mistake, well it's coming early next week now :)

BlueKoala
01-08-2009, 07:41 AM
ooooooOOOOOOooooo I'm excited for you! :P
Let us know how it goes :)

DeepDayze
01-08-2009, 09:16 AM
So, my new AMD Phenom II X4 940 system using an AMD 790GX/SB750 motherboard and an AMD Radeon HD 4850 1GB is arriving soon :D Could have been here already if I didn't do a mistake, well it's coming early next week now :)

Now that sounds like a sweet machine :)

Load it up with as much RAM as it could accept, a kickass video card and some fast disk storage and you'd have a winner!

webdoctors
02-12-2009, 08:44 AM
Hi,

I'm curious, when testing the Shanghai Opteron CPUs in Linux, did cpufreq work?

Were you able to change the frequency of the cpu's through linux?

Or did the BIOS not support this.

Thanks,

d2kx
02-12-2009, 01:51 PM
I don't know about Shanghai, but cpufreq works great with my Phenom II X4 940 and supports all four CnQ modes (3000 MHz, 2300 MHz, 1800 MHz, 800 MHz).

webdoctors
02-12-2009, 02:48 PM
I don't know about Shanghai, but cpufreq works great with my Phenom II X4 940 and supports all four CnQ modes (3000 MHz, 2300 MHz, 1800 MHz, 800 MHz).

Interesting... What motherboard do you have?
I have a phenom 9500, and it only supports 2 CnQ modes (1100 MHz, 2200 MHz) under Linux, using a Gigabyte 770 AM2+ board.

i had read earlier that cpufreq wasnt working when phoronix reviewed the opteron 2356

The only disturbance we ran into with the Opteron 2356 processors were PowerNow technology not working with Ubuntu 8.04 Beta
with the Linux 2.6.24 kernel. We suspect, however, that this issue should be addressed in the near future with cpufreq.

not sure if its because of the bios or linux kernel...
it'd be great to know before I buy one :)

d2kx
02-12-2009, 03:12 PM
MSI DKA790GX with BIOS v1.6

It isn't even one of the newer boards... it's shipped with 790GX/SB750!

deanjo
02-12-2009, 05:41 PM
Interesting... What motherboard do you have?
I have a phenom 9500, and it only supports 2 CnQ modes (1100 MHz, 2200 MHz) under Linux, using a Gigabyte 770 AM2+ board.

i had read earlier that cpufreq wasnt working when phoronix reviewed the opteron 2356

not sure if its because of the bios or linux kernel...
it'd be great to know before I buy one :)


Your 9500 shows only 2 values because there are actually only two values on the processor. I actually filed a bug report with that on the kernel and after further investigation and confirmation from AMD that the Phenoms only had two values and it was actually the windows C'n'Q driver that was misreporting the additional values.

betabandido
04-30-2009, 09:13 AM
Hello,

does anyone know whether it is possible to perform DVFS on a per-core basis? That would mean that it is possible to set one core at frequency F1 and another core (in the same chip) at frequency F2 (F1!=F2).

Have you performed any test like this one? For example the command 'cpufreq-info' can provide that information.

Thank you for any information you can provide us on this.

Best Regards,
Victor

frantaylor
07-08-2009, 09:38 AM
Hi,

I'm curious, when testing the Shanghai Opteron CPUs in Linux, did cpufreq work?

Were you able to change the frequency of the cpu's through linux?

Or did the BIOS not support this.

Thanks,

I have a SuperMicro H8DME-2 with two Shanghai Opterons on it. I'm running Fedora 10. CPUfreq just works by default, I didn't have to do anything. At idle it runs 800 MHz. The steps are: 800 MHz, 1.3 GHz, 1.8 GHz, 2.5 GHz.

jidacoadway
08-26-2009, 05:53 PM
I seem to have difficulty finding reviews relating to the AMD Dual Core Opteron 185. I believe its the last Socket 939 upgrade. The only thing I know about this cpu is from people from other forums saying how good it was. Any suggestions and comments as to where I can read a review about it, or you may have your own opinion or experience of it? How long has this cpu been around?

deanjo
08-26-2009, 06:18 PM
I seem to have difficulty finding reviews relating to the AMD Dual Core Opteron 185. I believe its the last Socket 939 upgrade. The only thing I know about this cpu is from people from other forums saying how good it was. Any suggestions and comments as to where I can read a review about it, or you may have your own opinion or experience of it? How long has this cpu been around?

Good chip, overclockable, somewhat old being around since the summer of 2006.

bridgman
08-26-2009, 07:13 PM
Looks like the Opteron 185 launched in March 2006. I don't have any experience with it but all the posts I saw seemed to think well of it. A number of posts commented that it was very similar to an FX-60 but with a locked multiplier and aimed at 24/7 server business. Seems to be around the -5000 level of performance; not sure what CPU you're running now.

It apparently sold for a bit over $1000 originally, hope the price has come down :D

highlandsun
08-26-2009, 11:13 PM
I still have one running in my Asus A8V-Deluxe, with 4GB of registered ECC DDR400. No problems with the chip itself, although the Asus BIOS didn't recognize it and therefore didn't enable C'n'Q for it. But I copied some ACPI tables from an Opteron 280 and tweaked them to get that enabled. It's actually still running OpenSUSE 10.2 with cherrypicked patches from 10.3; I've been loathe to update it because it's basically just humming along. Looking at my logs, "last reboot" was November 17 2007.

It was around the $300 mark when I got it, replacing an X2 3000+ which I gave to a friend.

energyman
08-27-2009, 12:53 AM
Hello,

does anyone know whether it is possible to perform DVFS on a per-core basis? That would mean that it is possible to set one core at frequency F1 and another core (in the same chip) at frequency F2 (F1!=F2).

Have you performed any test like this one? For example the command 'cpufreq-info' can provide that information.

Thank you for any information you can provide us on this.

Best Regards,
Victor

four month late, but yes. at least with phenom/phenom2