Here's a little teaser....

[justapost]

Nobody found or got any new bench for the Barcellonas, guys?

Anandatech posted a few in an comparison with hapertown last week.

http://www.anandtech.com/IT/showdoc.aspx?i=3099&p=1

[JohannesRS]

Thanks! and that seems that Barcellona can fight direct against Hapertown clock per clock, isn't it?

Price, Watts for servers, and also OCs for desktops. Those are the only stuff that we are still awaiting to know how the game will go.

[PhilDoc]

Anandatech posted a few in an comparison with hapertown last week.

http://www.anandtech.com/IT/showdoc.aspx?i=3099&p=1

Thanks, nice article. Looks like they use a Tyan mb this time. The results lead me to the same conclusion as before. It all depends on how much AMD can ramp up Barcelona. We also don't know which stepping of Barcelona they used, which leaves things a little up in the air. Either way, if they can get up to 3G reasonably soon, things could get very interesting.

[justapost]

Update: For those that are looking for more details and wondering why certain other chips aren't included, at the time testing was conducted we did not have any of the faster 2.5GHz Barcelona chips (or the slower Harpertowns). That situation has been remedied in terms of AMD's CPUs, and we will have some update articles looking at how the faster Barcelona compares with other processors. Stay tuned....

Hope they will do a clock per clock comparison soon.

[PhilDoc]

Hope they will do a clock per clock comparison soon.

Sounds like it. Hopefully, within the next week or two.

[knightwolf654]

AMD has it down, less power consumption, less utilization to complete tasks.
harpertown will need one heck of a psu over 400 watts on a few of those tests!

[KTE]

Nobody found or got any new bench for the Barcellonas, guys?

Those 2 bench comparisons I posted compared to what we've seen here were both new from TechReport and 2CPU, as well as the AnandTech that justapost later posted. They're all pretty much the same in conclusion, although the absolute performance figure in all of them are different; clock-per-clock Barcelona server releasing Q4 07 looks better than Penryn server (1600FSB) releasing Q1 08 more often than not. Altogether, it even wins a few whilst with a 500MHz deficit and loses many due to raw MHz power.

I remember Henri saying this last year with regards to competition... "it's a race, of course it's a race."

[Blasar]

But how long will AMD take to get the Barcelona speeds to the 2.5-3.0 GHz?

And about Phenom´s launch date and speeds? Is there more information?

[informal]

K10 supply has improved greatly:
http://www.amdzone.com/modules.php?o...ticle&sid=8453

AMD Opteron Quad Core 2344 HE CPU OS2344PAL4BGE, 1.7GHz, Socket F (1207), 1000MHz HT, 2MB L2 Cache $236.00

[Thesavage]

K10 supply is has improved greatly:
http://www.amdzone.com/modules.php?o...ticle&sid=8453

Wonder if this means that Phenom will be a little early, as barcelona supplies look good?

[s7e9h3n]

Sorry I haven't been able to get back to this thread until now. At the moment, it seems that this is about as good as it's going to get with the current hw/sw support available. The bios on my Tyan s3993 mobo still has no OC options, no voltage options (including dpp), no HT3 options (yes, this board supposedly supports it), and no memory timing options. Add to that the fact that neither Broadcom, Tyan, nor Serverworks seems to have windows chipset drivers available for the Broadcom/Serverworks Ht2100 Chipset and the memory is stuck in single channel mode and you end up with this totally, utterly, stock-as-can-be dual Barcelona rig that I have here. Here's Cinebench and Everest's cpu-tests with the quads running @ the correct speeds:

















I'll see what I can figure out about the memory problem and run some in-depth benches later.......

[alayashu]

FP power looks weak... teh same B1 eh?:P

thanks Stephen for your patience with these CPUs!

[s7e9h3n]

FP power looks weak... teh same B1 eh?:P

thanks Stephen for your patience with these CPUs!

I noticed that as well, but also note that on all the FP tests, Everest gave that error message so something may be wrong there....

[informal]

Yep Steven,if Everest deals with unsupported HW,it gives wrong results in some tests(and warns with that message)

[Thesavage]

If you do a Sandra memory bandwidth test what are those scores looking like on single channel?, and thanks for all the hard work.

[PhilDoc]

Very nice, Thanks, S7. Looks like you finally got the bios work around or was there a differenet problem?

[s7e9h3n]

Better, but still crap

Very nice, Thanks, S7. Looks like you finally got the bios work around or was there a differenet problem?

It was a bug in the original bios that somehow didn't get caught. Basically pinned core0 @ max power-save mode. A bios revision corrected that problem

[PhilDoc]

Thanks S7 for the quick response. Sandra doesn't look too bad for single chanel. Hard to tell what the problems is there. I'm sure you've tried different ram modules. Could be the mb itself.

[Thesavage]

Wow yes thanks for the quick response, looks very good for single channel.

[JohannesRS]

great. So, now we have the cores at ptoper speeds, so.

But unfortunatelly memory still at single channel, and I would expect that to give a hit on opterons performance. Big one.

S7, do you think it worths trying a linux installation at this point, or it more wise to wait for a fully working bios?

And, of course, thanks a lot for sharing those results with us!

[BadNizze]

Sigel channel and DDR-667 @ 5-5-7-15 ain´t helping either.
Wonder how much 3-3-3-8 @ ddr800 dual would give in therms of sandra memory bandwith? As always thanks for sharing S7

[JohannesRS]

Those are the expectations for Phenoms in terms of memory. Don't know the timings of ECC memories, but 667 MHz are the specificated ones.

The main problem for server parts, for me, now looks like only the single channel problem still. Is it right?

[freeloader]

I really hope the corporate types aren't having the problems (ie...dual channel memory problems, cores not recognized properly) discussed in this thread. Even though it's not AMD's fault (nothing wrong with the chips that we know of), the motherboard manufacturers can hurt AMD just as bad with flaky bioses.

[s7e9h3n]

For reference from a dual-dual Opteron rig...Octal AMD should be MUCH, MUCH faster:



The problem seems to be with the chipset not allowing ccNuma to function properly along all 8 cores as the only two IMC's that get loaded during bw benchmarks are core0 of each physical proc.

[PhilDoc]

I wonder if its the chipset itself or a faulty mb. Tried to look the reviews that ran sandra, but they were running supermicro boards. I think the tech report got 11,500ish so they must have gotten dual channel to work. Probably no way to try a different mb.