Hi Folks,

just looked through some statistics on BOINCstats.com - namely, the per-CPU listings for SETI@Home.

One thing stood out -

AMD Athlon(tm)/Opteron(tm) CPU-model unknown

with a per-CPU RAC of 1,457.64, which equates to ~2915 RAC for the whole dual CPU system.

Now, why is this significant? Contrast this ratio with the currently highest-ranked dual-core systems in the top computers list (all data taken from "Top Computers" pages, not BOINCstats).

This is top host #51 (Mark Sattler's), pulling an RAC of 3007. Checking out this host's details, you will see it runs a sky-high OC - default would be 2.93 GHz, but it runs at 3.784 GHz instead.

This host delivers only 100 RAC extra compared to this new AMD Opteron system.

Now contrast it to the highest-ranked dualcore AMD system in the top 1000 hosts.

Also belonging to msattler (kudos Mark ;), this is an AMD FX-60 dualcore system pulling 1511 RAC (or ~755 per core) at spot #497. Knowing Mark, this system doesn't exactly run stock (2.6 GHz), either - instead, it runs at 2.94 GHz.

So, a quick comparison of these dual CPU systems:

New AMD Opteron system (unknown clock speed, estimated below 3 GHz)
2 x 1457 -> 2915 RAC

Core 2 X6800 system (3.784 GHz)
2 x 1503 -> 3007 RAC

A64 FX-60 system (2.94 GHz)
2 x 755 -> 1511 RAC

Can anyone spot a trend yet?
No? Let's calculate a per-MHz efficiency for them, then.

Let the simple formula -

RAC(core) / MHz(core) = Efficiency(core)

determine this.

New Opteron (for the sake of argument, let's say it runs 3 GHz, though to my knowledge, it's around 2.6)
1457 / 3000 = 0.4856

X6800
1503 / 3784 = 0.3971

A64 FX-60
755 / 2940 = 0.2568

Now, the trend should definitely be visible. This new AMD CPU is so far ahead in efficiency it simply cannot be the same architecture. I can only surmise that dual-core Barcelona versions have already made their way to partners (and onto SETI@Home!).

I must say, I'm flabberghasted. I did expect a performance increase, but this really reminds me of the first Opteron/A64 vs. Athlon XP/Pentium 4 benchmarks.

Out of this world.

Opinions?

Kind regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

link please, by me at Top host 51

http://setiathome.berkeley.edu/top_hosts.php?sort_by=expavg_credit&offset=40

Greetings from Germany NRW
Ulli

Well, I never figured AMD was going to just rollover and die. :-)

I would guess they have a few more tricks up their sleeve for a bit further down the road. ;-)

Alinator

Maybe we could have Berkley give it a drug test? :-)

---

Hi Folks,

just looked through some statistics on BOINCstats.com - namely, the per-CPU listings for SETI@Home.

One thing stood out -

AMD Athlon(tm)/Opteron(tm) CPU-model unknown

with a per-CPU RAC of 1,457.64, which equates to ~2915 RAC for the whole dual CPU system.

Now, why is this significant? Contrast this ratio with the currently highest-ranked dual-core systems in the top computers list (all data taken from "Top Computers" pages, not BOINCstats).

This is top host #51 (Mark Sattler's), pulling an RAC of 3007. Checking out this host's details, you will see it runs a sky-high OC - default would be 2.93 GHz, but it runs at 3.784 GHz instead.

This host delivers only 100 RAC extra compared to this new AMD Opteron system.

Now contrast it to the highest-ranked dualcore AMD system in the top 1000 hosts.

Also belonging to msattler (kudos Mark ;), this is an AMD FX-60 dualcore system pulling 1511 RAC (or ~755 per core) at spot #497. Knowing Mark, this system doesn't exactly run stock (2.6 GHz), either - instead, it runs at 2.94 GHz.

So, a quick comparison of these dual CPU systems:

New AMD Opteron system (unknown clock speed, estimated below 3 GHz)
2 x 1457 -> 2915 RAC

Core 2 X6800 system (3.784 GHz)
2 x 1503 -> 3007 RAC

A64 FX-60 system (2.94 GHz)
2 x 755 -> 1511 RAC

Can anyone spot a trend yet?
No? Let's calculate a per-MHz efficiency for them, then.

Let the simple formula -

RAC(core) / MHz(core) = Efficiency(core)

determine this.

New Opteron (for the sake of argument, let's say it runs 3 GHz, though to my knowledge, it's around 2.6)
1457 / 3000 = 0.4856

X6800
1503 / 3784 = 0.3971

A64 FX-60
755 / 2940 = 0.2568

Now, the trend should definitely be visible. This new AMD CPU is so far ahead in efficiency it simply cannot be the same architecture. I can only surmise that dual-core Barcelona versions have already made their way to partners (and onto SETI@Home!).

I must say, I'm flabberghasted. I did expect a performance increase, but this really reminds me of the first Opteron/A64 vs. Athlon XP/Pentium 4 benchmarks.

Out of this world.

Opinions?

Kind regards,
Simon.

Hmmm, and AMD's stock is down low right now...Might just be a good time to buy...

---

Official Abuser of Boinc Buttons...
And no good credit hound!

Opinions?

Blame Misfit.

---

me@rescam.org

I thought Barcelona was a quad-core design, not dual-core.

---

I thought Barcelona was a quad-core design, not dual-core.

My first thought as well...though it IS very interesting to see that there is some life in AMD still.

It just didn't feel right to buy Intel when the Core2 came out, but the performance was too good to resist.

/Anton

I thought Barcelona was a quad-core design, not dual-core.

It is... here a link to the AMD news room.

ISSCC, SAN FRANCISCO -- 12th February 2007 --

AMD (NYSE: AMD) today unveiled more industry-defining architectural features for its upcoming native Quad-Core AMD Opteron™ processors (codenamed “Barcelona”) during a presentation to the International Solid State Circuits Conference (ISSCC). In addition to the estimated 40-percent performance advantage native Quad-Core AMD Opteron™ processors can offer over the competition in benchmark testing, AMD’s redesigned microarchitecture will enable new power- and thermal-management techniques, strengthening the industry-leading performance-per-watt AMD Opteron processors currently deliver today. Among the new features are enhancements to AMD PowerNow!™ technology for dynamic adjustment of individual core frequencies........

;-)

---

Hi Folks,

TheInquirer.net linked this thread, only they got their source wrong - seems someone emailed Charlie Demerjian alongside me emailing Mike Magee (the editor). In any case, 'twas me who researched it ;o)

http://www.theinquirer.net/default.aspx?article=37907

As for Barcelona being the first true quad-core on a single chip design, that's certainly true. However, it's not just about that - obviously, there will be variants with less cores as well.

The real architectural leap isn't just putting 4 cores on one piece of silicon; rather, it's massively improved FP performance as well as better power saving features.

In short:

• shared L3 cache (AMD first)
  
• massively improved FP performance (AMD specify 1.8x current Opteron performance core-by-core)
  
• higher IPC (Instructions per Clock) than any released architecture
  

Check this page for a breakdown of these features.

They also specify that Barcelona chips will run at lower frequencies than current dual-core ones; so the numbers I quoted in the first post may be even better.

Now for my second point - Barcelona is not only a quad-core design! Check out this Inquirer link and this CPU-Z image.

There will be dualcore variants at launch.

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Wait a minute, it just dawned on me you can't use the BoincStat CPU breakdown page for this kind of comparison.

It doesn't represent a single machine, as you are implying (the Inquirer too for that matter), it shows the aggregate numbers for all the hosts with that CPUID tag. Also IIRC, there has to be at least two hosts with the same CPUID tag to show up on the CPU breakdown at BoincStats.

To verify that, you can't find a host that has the total credit shown on the CPU breakdown page which matches the tag.

I guess we'll have to wait until someone actually spots one of them as a WU partner before we'll know for sure what the story is, but with only two of them out there at this point it'll be almost like finding the signal from ET. ;-)

Alinator

Usually, yes; however, in this instance there are a total of 2 such CPUs listed. It stands to reason that they are indeed part of a dual-core system.

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Usually, yes; however, in this instance there are a total of 2 such CPUs listed. It stands to reason that they are indeed part of a dual-core system.

Regards,
Simon.

No. They are two separate machines. Both quad core. Not very impressed with the stats or benchmarks.

Edit: And they were added back in October and November of 2006!


http://www.boincstats.com/stats/host_graph.php?pr=sah&id=2877462

http://www.boincstats.com/stats/host_graph.php?pr=sah&id=2858830

Or from SETI:

http://setiathome.berkeley.edu/show_host_detail.php?hostid=2877462

http://setiathome.berkeley.edu/show_host_detail.php?hostid=2858830

---

Dublin, California
Team: SETI.USA

Wait a minute, it just dawned on me you can't use the BoincStat CPU breakdown page for this kind of comparison.

Right. It takes the RAC average of all the chips. So the more numerous chips will have a lot of part-time crunchers, multi-project crunchers, people who have left and RAC approaching 0.

It doesn't represent a single machine, as you are implying (the Inquirer too for that matter), it shows the aggregate numbers for all the hosts with that CPUID tag. Also IIRC, there has to be at least two hosts with the same CPUID tag to show up on the CPU breakdown at BoincStats.

Right.

---

Dublin, California
Team: SETI.USA

Usually, yes; however, in this instance there are a total of 2 such CPUs listed. It stands to reason that they are indeed part of a dual-core system.

Regards,
Simon.

Found them:

2877462

2858830

Alinator

<edit> I see others were hunting them as well. :-)

Don't get me wrong though, I'd love to see AMD push it's way back up to the top of the heap again. ;-)

Alinator

Usually, yes; however, in this instance there are a total of 2 such CPUs listed. It stands to reason that they are indeed part of a dual-core system.

Regards,
Simon.

No. They are two separate machines. Both quad core. Not very impressed with the stats or benchmarks.

Edit: And they were added back in October and November of 2006!


http://www.boincstats.com/stats/host_graph.php?pr=sah&id=2877462

http://www.boincstats.com/stats/host_graph.php?pr=sah&id=2858830

Or from SETI:

http://setiathome.berkeley.edu/show_host_detail.php?hostid=2877462

http://setiathome.berkeley.edu/show_host_detail.php?hostid=2858830

Keep in mind that if these are engineering samples, they most likely are clocked slower than the production Opty 800 series. It looked like they were doing better than half the RAC of the best 8 core 800's I scanned by, although there were a few 200 series with better RAC as well.

Alinator

For those that are interested, AnandTech have an article on what AMD have done with their architecture to give us Barcelona.

---

Grant
Darwin NT

For those that are interested, AnandTech have an article on what AMD have done with their architecture to give us Barcelona.

Good article, and a good summary of the history and trade-off between Intel and AMD.

The opening comment is very apt and rather amusing: "Over the past several years, Intel has followed an odd path of microprocessor design..."

AMD have quite a fight for their design teams, yet they do keep coming up with some good moves.

Happy crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Don't get me wrong though, I'd love to see AMD push it's way back up to the top of the heap again. ;-)

Alinator

I'm hoping my dual FX-70 will see an AMD back in the top 100. It's been on SETI 100% for about a week and moving up nicely, can't wait to see the quads hit the market:-)


Fish

The measurement of processor speed/efficiency that I like to use is how many CPU seconds does it take to crunch a credit at the 62.4 credit WU size. I learned this from hiamps and it helped me to determine that my AMD X2 wasn't crunching as efficiently as it should have been doing for it's speed.

Some numbers:

My OC'd AMD X2 @ 2600 MHz = 110 CPU seconds / credit @ the 62.4 credit WU size

Fish's FX-70 @ 2972 MHz = 94 CPU seconds / credit @ the 62.4 credit WU size

Msattler's Core 2 Quad @ 3100 MHz = 56 CPU seconds / credit for 62.4 WUs.

What is the value for this mystery machine? Can we find the results?

Brock

---