Hello,
for next year i am planning a System-Upgrade to a Multi-Socket-System.

Now i think about, whats the better CPU-Brand-choice for CPDN?
The coming up Opteron 6200 Series (Bulldozer) has an increased Integer-Power (~220GFLOPS with 2x6276) cause AMD means this were the most used kind of CPU-Caltulatings.
So the Opteron 6200 has comparatively "bad" Floating-Point-Performance.
For FPU i think Intel is the better choice.

But what we do need for CPDN primarily?
thanks

The climate models use large amounts of floating point calcs, so a good fpu.
The latest lot of Sandy Bridge processors seem good.

There's several threads on these matters here in Computing efficiency & performance on our php forum. The latest thread is currently at the top.

---

Backups: Here

What Les said. Intel has been king in cpdn for 5 years, and especially since the Core iX series came out in 2009.

But I'd like to see what a high end Bulldozer would do in cpdn.

. . . and the recent Intel releases are six-core -- and expensive.

---

"We have met the enemy and he is us." -- Pogo
Greetings from coastal Washington state, the scenic US Pacific Northwest.

It would be interesting to see what the new 16 ish core beasts do. - Probably use up most of the power my pv is producing at this time of year.

I'm pleased so far with my new i7. I was expecting it to have four cores but was surprised to find eight.

I'd started listening to YouTube recordings of Teresa Berganza while doing various jobs on the computer but it's crunching such a lot and the eight models are using so much of my 20GB monthly bandwidth allowance that I've had to limit my listening.

http://www.youtube.com/watch?v=dWGc9IoxhAw Las tres hojas
http://www.youtube.com/watch?v=ZBEoTZvpHPg Sevillanas

It also has a usable GPU but there's not enough spare bandwidth to let the GPU have more than three or four hours' practice on Einstein each day.

---

Cpdn news

Sooo,

Let's have a look to the SPECs new Opteron 6200 Series:
The SPECfp results of the Opteron 6276 (16c/2,3Ghz) looks very very good:

http://www.tecchannel.de/server/prozessoren/2038251/cpu_test_amd_opteron_6262_he_und_6276_mit_16_core_bulldozer/
The 5th or 6th diagram "Specfp_rates"

and on spec.org u can see the details:
http://www.spec.org/cpu2006/results/res2011q4/cpu2006-20111025-18763.html
http://www.spec.org/cpu2006/results/res2011q4/cpu2006-20111025-18753.html
this is a 2-Socket-Machine
=> ~360GFlops!!!!

with old 2x6180SE Top-model:
http://www.spec.org/cpu2006/results/res2011q1/cpu2006-20110131-14298.html
320GFlops

Intel Xeon E7-4870 10Core 2-Socket:
http://www.spec.org/cpu2006/results/res2011q2/cpu2006-20110425-15899.html
370GFlops
BUT:
Price of E7-4870: ~4000Euro per CPU
Opteron 6276(2,3Ghz): 750Euro!!!!
Opteron 6272(2,1Ghz): 480Euro!

I think the new Opterons could be the best CPUs in power AND price for next months or years!
My next System could only be a Multi-Socket-Opteron :-)
...
then i discovered the SPEC-descriptions on: http://www.spec.org/cpu2006/CFP2006/

"481.wrf C,Fortran - Weather - Weather modeling from scales of meters to thousands of kilometers. The test case is from a 30km area over 2 days."

Could this the benchmark, with the most concordance to our Climate-Models?


PS: Sandy-Bridge-E I7-3960X reaches 140 GFlops - haaahaa :-)

My recommendation was based on what we have seen on cpdn. A Core i7 920 (hardly a high end processor) can beat a higher priced Phenom II X6 1100T in total throughput of models. Up until very recently, the 1100T was the best AMD desktop processor. Now if you're talking multiple sockets and high end server CPUs, most of us here don't have experience with that. The fact that these models were compiled with the Intel compiler, also plays some into why they are so fast on Intel.

I would need to win the lottery or something like that to get any two socket machine - so most of this thread is of purely academic interest to me. Still I can dream!

My recommendation was based on what we have seen on cpdn. A Core i7 920 (hardly a high end processor) can beat a higher priced Phenom II X6 1100T in total throughput of models.

Hmm, in my experience the i7-920 and X6-1090T performs similarly, but the X6-1090T has decidedly a big advantage then it comes to cost of buying. Just for the cpu the i7-920 was 33 % more expensive than the X6-1090T, and also other things like mainboard and memory was more expensive for i7-920 than for Amd.

Now, atleast around here no-one has sold any i7-920 for the last year atleast, but let's look on the current entry-level offering from intel among the i7-cpu's, the i7-2600K. The i7-2600K is 44 % more expensive than an AMD's X6 1100T. No idea how fast the i7-2600K is, but would still guess it's less than 40 % faster...

A hex-core intel-cpu should be faster, but is also much more expensive. Also, running multiple memory-hungry CPDN-models will have an impact on performance, so it's not certain a hex-core gives much higher performance than a quad-core does.

This is exactly the Reason, why i go back to AMD.

For a long time Intel CPUs were the best in Price an Price/Money.
But when i look to the actually prices of XEON and I7, i think AMD is the most better choice.
If you only want power, u have to use a Intel but for what price?!
Some GFlops more will cost u some hundreds of Dollar/Euro more.

Next year AMD offers an Single-Socket Opteron 3000 up to 3Ghz on Socket G34!
And next yea AMD launches a new Server-Platform with "terramar" CPU up to 20 Cores. :-)
Intel is OUT and only for people, who have to much money.
PS: yes i also saw, that the power of 1100T and 1090T is very good for price.

Hmm, in my experience the i7-920 and X6-1090T performs similarly, but the X6-1090T has decidedly a big advantage then it comes to cost of buying. Just for the cpu the i7-920 was 33 % more expensive than the X6-1090T, and also other things like mainboard and memory was more expensive for i7-920 than for Amd.

Hmm, I was going from memory (which may be faulty) of performance on slab. But now running 8 hadcm3ns on the i7 920 and 6 hadcm3ns on the my 1090T, it appears the 1090T has about 8% greater total throughput.

I bought the i7 920 for $199 U.S. 2.5 years ago. The motherboard was about $200. I bought the 1090T for $225 1.5 years ago and the associated motherboard for $180. I could have gone cheaper on the motherboard but figured if I was going to run distributed computing nearly continuously on this PC, the motherboard probably should be a quality one.

Nowadays, there is no i7 anywhere near equivalent in price to the AMD hex-cores. Of course an upper end Sandy Bridge quad core might be about as fast, but they are pretty expensive yet as well.

So, for total throughput for the price, AMD hex-cores do look very good.

I'm pleased so far with my new i7. I was expecting it to have four cores but was surprised to find eight.

I'd started listening to YouTube recordings of Teresa Berganza while doing various jobs on the computer but it's crunching such a lot and the eight models are using so much of my 20GB monthly bandwidth allowance that I've had to limit my listening.

http://www.youtube.com/watch?v=dWGc9IoxhAw Las tres hojas
http://www.youtube.com/watch?v=ZBEoTZvpHPg Sevillanas

It also has a usable GPU but there's not enough spare bandwidth to let the GPU have more than three or four hours' practice on Einstein each day.

I think you will find that you 4 physical cores (Quad core) but with hyperthreading turned on Windows reports that you have 8 cores, 4 physical and 4 virtual (Intel does not have a Core i7 with 8 physical cores just yet but are working on it).

But as far as you are concerned you can have 8 things running at once just like an 8 core computer.

Enjoy
Conan

---

mo.v wrote:

I'm pleased so far with my new i7. I was expecting it to have four cores but was surprised to find eight.

<snip>

Hi mo.v, etal,

After looking at your setup something shocking was staring me in the face...I'm using the same processor in an iMac and my Integer benchmarks are nowhere near as high as yours. I'm curious whether you are using just the four actual cores only on CPDN or are you using the virtual cores as well...i.e. 4 WUs at a time or 8 WUs?

My iMac i7

Your i7
Measured floating point speed 3282.91 million ops/sec
Measured integer speed 12205.76 million ops/sec

My i7
Measured floating point speed 3760.46 million ops/sec
Measured integer speed 6692.85 million ops/sec

I run 8 simultaneous CPDN WUs. I'm thinking these measurements are based on per-core performance, perhaps someone can shed some light on this for me?

Thanks! :)

You are correct in that the numbers are per core, real cores as I understand it. Using the four virtual cores boosts throughput 15-20% but individual tasks take longer in wall-time than if running four at a time.

For CPDN, only floating-point matters.

---

"We have met the enemy and he is us." -- Pogo
Greetings from coastal Washington state, the scenic US Pacific Northwest.

Well, astroWX, therein lays my problem in understanding this...I would have assumed that since mo.v and I are using identical processors, and that my Integer benchmark is pretty much half of his, that CPDN was Integer-intensive and not Floating Point-intensive...or possibly that the reason my benchmarks were showing up as half was because the ALU had to split time, cycles and circuitry between the work of the actual core and the associated virtual core...that is, if he was only running 4 WUs and me, 8WUs. As it is both our Floating Point benchmarks are in the same ball park.

Thanks for your help, but I'm thinking that maybe I'm still missing some other obvious point here about the discrepancy in our Integer benchmark differences. Any thoughts anyone?

The benchmarks are (or, perhaps, were) used by some projects for the allocation of credit. The benchmarks are ignored by CPDN, which awards credit based on science-work completed (i.e. 'trickles') rather than effort - the boulders have to be pushed to the top of the hill and CPDN rewards boulders not sweat.

The most likely reason for the difference in reported benchmark performance is the difference in operating systems. I don't mean by that that one operating system is faster, but mean instead that the benchmarks are meaningless in differing ways on the two platforms.

And the benchmarks are a 'BOINC thing', (which creates them), where the formula varies now and then between different BOINC versions.

---

Backups: Here

"the benchmarks are meaningless in differing ways on the two platforms." That makes sense to me, I remember when looking at some crashes, being asked if I had my box which runs linux significantly overclocked because the benchmarks were high for that particular processor.

Dave

Thank you, everyone, for your feedback here, it's certainly helped me with my understandings of a few things going on here. :)

However, my main question still remained unanswered, Does using CPDN using 4 actual cores vs. 4 actual cores + 4 virtual cores cause the Integer benchmarks to be halved, or otherwise affected? Thank you, Les, for having me consider that this might be a BOINC thing instead of a CPDN thing...you prompted me to do an experiment I was loathe to do earlier because of the fragility of BOINC Manager 6.12.35 (x86). (I found that messing with the preferences was causing random appearances of the dreaded Waiting for GPU Memory WU status glitch!)

So I decided to brave messing with my preferences to try and determine whether this Integer Benchmark-halving effect was being caused by actual and virtual core usage. Here are my results...

BOINC Manager set to 50% CPU usage, no projects running...

Thu Dec 22 08:05:37 2011 | | Number of CPUs: 4
Thu Dec 22 08:05:37 2011 | | 4367 floating point MIPS (Whetstone) per CPU
Thu Dec 22 08:05:37 2011 | | 11954 integer MIPS (Dhrystone) per CPU

BOINC Manager set to 50% CPU usage, 4 CPDN WUs running...

Thu Dec 22 08:06:38 2011 | | Number of CPUs: 4
Thu Dec 22 08:06:38 2011 | | 4360 floating point MIPS (Whetstone) per CPU
Thu Dec 22 08:06:38 2011 | | 11868 integer MIPS (Dhrystone) per CPU

BOINC Manager set to 100% CPU usgae, 4 CPDN WUs running...

Thu Dec 22 08:08:18 2011 | | Number of CPUs: 8
Thu Dec 22 08:08:18 2011 | | 3762 floating point MIPS (Whetstone) per CPU
Thu Dec 22 08:08:18 2011 | | 6683 integer MIPS (Dhrystone) per CPU

BOINC Manager set to 100% CPU usage, 8 CPDN WUs running...

Thu Dec 22 08:12:21 2011 | | Number of CPUs: 8
Thu Dec 22 08:12:21 2011 | | 3767 floating point MIPS (Whetstone) per CPU
Thu Dec 22 08:12:21 2011 | | 6697 integer MIPS (Dhrystone) per CPU

And, finally,
BOINC Manager set to 100% CPU usage, 0 CPDN WUs running...

Thu Dec 22 08:37:29 2011 | | Number of CPUs: 8
Thu Dec 22 08:37:29 2011 | | 3764 floating point MIPS (Whetstone) per CPU
Thu Dec 22 08:37:29 2011 | | 6696 integer MIPS (Dhrystone) per CPU


It appears to me from these results that both Floating Point and Integer performance are directly affected by BOINC Manager CPU settings regardless of project WU load or (as these numbers now match more closely those of mo.v's identical CPU) OS platform...

Floating Point went from 4-core @ 4367 to 8-core (4-core + 4-virtual core) @ 3767 (Whetstone) an ~16% performance hit per core.

Integer went from 4-core @ ~11,900 to 8-core (4-core + 4-virtual core) @ 6696 (Dhrystone), an ~44% performance hit per core.

Any and all �feedback welcomed, along with pointing out any errors in my methods or results!

Hope this all adds to the original discussion on FPU or Integer Power?

:)