Barcelona Opteron 2350(B1) arrived

[macci]

are you sure? then how CnQ will work with IMC?

CnQ lowers both CPU and IMC multi in idle, on the fly

that would be NB DID (divides the clock by 2). but the actual multiplier (NB FID) can not be changed (on the fly).

WHY did you run the Barchie with HIGHER fsb than it's default (200MHz)...?...

George,
I know your not an AMD guy () but please note that the HT Multiplier in this benchmark is at X4 mode. So the "FSB" is actually underclocked (960MHz vs 1000MHz which is default for that CPU).
The HT ref. clock (240MHz vs 200MHz) makes absolutely no difference in terms of performance.
HT Link (1GHz by default on this example) is the "FSB" of AMD world and it depends on HT Link multiplier and HT ref. clock. so 240x4 (960MHz) is slower than 200x5 (1GHz) (assuming that the Memory bus and CPU core clock would be at the same frequency).

[Morais]

very helpfull comment macci, thanks for posting

[ozzimark]

High fsb and high rams frequency helps a bit CPU score in 2k6....

surely you're joking. i know you've messed around with plenty of amd chips during K8's reign, and have seen first hand how the htt (or fsb if you want to call it that) does NOTHING for performance

edit:
should have read the rest of the thread before posting
good explaination macci, though i'd like to add something else to that: the 'HT Link' speed, or 'FSB" (i get frustrated by how things get multiple names that sometimes overlap) does very little for performance as well, except in multi-cpu situations when running multithreaded apps that require lots of communication between the cpus

[Lightman]

Thakn you Kyosen for answering on my question (about 0.5 multipliers)!

One thing which strikes me is Barcelona scaling under Win XP x64. It is more than 4x! How come?

Is this because of the way memory controller is working (unganged mode) or L3 Victim Cache is helping here? Any reasonable explanation??

[hipro5]

Well......I asked a friend (user) to our forums who owns a Q6600 and a 8800GT to bench 2k6 at 2400MHz and VGA 600/900......

There you go.....

He used "bios defaults"....Ram "by default" , etc.....As a "regular user" is going to buy a system like this and bench 2k6 without overclocking, playing with the rams, etc.....

Barchie 2400MHz (with higher fsb than default - edit: OK HT lower ) : 10926 marks...

Core 2 Duo QUAD Q6600 ALL DEFAULT : 12073 marks...

SO:

Barchie
SM2 = 4439
SM3 = 4672
CPU = 3553

Q6600
SM2 = 5052
SM3 = 5056
CPU = 3857

Now what?....

George,
I know your not an AMD guy () but please note that the HT Multiplier in this benchmark is at X4 mode. So the "FSB" is actually underclocked (960MHz vs 1000MHz which is default for that CPU).
The HT ref. clock (240MHz vs 200MHz) makes absolutely no difference in terms of performance.
HT Link (1GHz by default on this example) is the "FSB" of AMD world and it depends on HT Link multiplier and HT ref. clock. so 240x4 (960MHz) is slower than 200x5 (1GHz) (assuming that the Memory bus and CPU core clock would be at the same frequency).

I know........BUT I see "FSBs" all over the place.......

[KTE]

[B]Core 2 Duo QUAD Q6600 ALL DEFAULT : 12073 marks...

That's about right where it should be. Odd that CPU/MB made so much difference to SM2/SM3 scores though, 1000 points.

http://www.tweaktown.com/reviews/1210/5
http://www.bootdaily.com/index.php?o...1&limitstart=1
http://vr-zone.com/articles/Nvidia_G...w/5369-13.html

[hipro5]

That's about right where it should be. Odd that CPU/MB made so much difference to SM2/SM3 scores though, 1000 points.

http://www.tweaktown.com/reviews/1210/5
http://www.bootdaily.com/index.php?o...1&limitstart=1
http://vr-zone.com/articles/Nvidia_G...w/5369-13.html

It's not that odd......The other user used Windows Vista......

[tictac]

wow.. in 64bit windows barcelona speedup.. 4.0x.. thats 100% efficiency.. would be interesting to see other benchmark running on 64bit windows.. 3dmark ?

[savantu]

wow.. in 64bit windows barcelona speedup.. 4.0x.. thats 100% efficiency.. would be interesting to see other benchmark running on 64bit windows.. 3dmark ?

Nah , that's impossible.

The only explanation is IMO that something is wrong/suboptimal when running only one thread.Which is weird since one thread should have all the L3 for itself.
Moreover , one thread has both IMCs working for it.

In the multithreaded scenario , data BW from both L3 and memory is shared => welcome to Amdahl's law.

http://en.wikipedia.org/wiki/Amdahl's_law

[s7e9h3n]

Nah , that's impossible.

The only explanation is IMO that something is wrong/suboptimal when running only one thread.Which is weird since one thread should have all the L3 for itself.
Moreover , one thread has both IMCs working for it.

In the multithreaded scenario , data BW from both L3 and memory is shared => welcome to Amdahl's law.

http://en.wikipedia.org/wiki/Amdahl's_law

Hate to say it, but you're right......here's my Cinebench on Server 2k3 x64. Kyosen-san's single threaded score is unusually low....



*BTW, K10 has only ONE IMC, not two

[tictac]

i guess 3.1ghz phenom b2 could break 12k on cinebench under 64bit mode...

[KTE]

welcome to Amdahl's law.

http://en.wikipedia.org/wiki/Amdahl's_law

A theoretical prediction used only as an estimate of expected performance which has been disproven in many computing scenarios well known among scientists, especially when you integrate caches and variate sub-components in a given processor, and even more so with multi-core technologies. Microprocessors depend on how well they are utilized by a given application and we do not know how to utilize 100% perfect working power and resources of a CPU in all scenarios since the application coding is not CPU specific, but rather wildly generic. Hence, if you could make the coding very optimized and processor specific, you are bound to see a speedup even at the same frequency. It's more about processor resource efficiency.

[JohannesRS]

Disagree. Amdahl is an estimate, but it's maximum performance estimate. if that thing would go 4 times faster in 4 cores, that would need to mean perfect scalling, coding, and even worst, parallel from the first to last line of code.

I really don't see it happening.

[savantu]

..
*BTW, K10 has only ONE IMC, not two[/B]

Umh...there are actually 2 64 bit channels which put together act as a 128 one.This improves performance when multiple threads access memory.
This is something Intel chipsets have done since the I865.

[savantu]

A theoretical prediction used only as an estimate of expected performance which has been disproven in many computing scenarios well known among scientists, especially when you integrate caches and variate sub-components in a given processor, and even more so with multi-core technologies. Microprocessors depend on how well they are utilized by a given application and we do not know how to utilize 100% perfect working power and resources of a CPU in all scenarios since the application coding is not CPU specific, but rather wildly generic. Hence, if you could make the coding very optimized and processor specific, you are bound to see a speedup even at the same frequency. It's more about processor resource efficiency.

Give 2 examples which respect the conditions imposed by the law.

[siyah]

@s7e9h3n

hi my Quad 6600 B3 is faster your barci

[Dagalidis]

@s7e9h3n

hi my Quad 6600 B3 is faster your barci

Dear friend....
I will not ask about your age...
But DO YOU REALLY know what means COMPARE?

Try again with more equal speeds & FSB and post again....
Don’t fire if you don’t know how to use a gun....(you can kill yourself and our sense of humor)

[Lastviking]

Hmmm is it becuse the lower speed for the K10 thats make it to have Multiprocessor speedup 6.79x instead of low 3.46x like the Q6600 have? or do i see big improvement with more cores on K10?

[accord99]

Hmmm is it becuse the lower speed for the K10 thats make it to have Multiprocessor speedup 6.79x instead of low 3.46x like the Q6600 have? or do i see big improvement with more cores on K10?

6.79x is with 8 cores.

[Lastviking]

6.79x is with 8 cores.

ah missed that , to bad..

[mstp2009]

6.79x is with 8 cores.

Hold on a sec here. Am I reading this right: in C10 a 2.27GHz C2Q is faster than TWO Quads K10 at 2.0GHz?

:scratches head:

[nemrod]

Dear friend....
I will not ask about your age...
But DO YOU REALLY know what means COMPARE?

Try again with more equal speeds & FSB and post again....
Don’t fire if you don’t know how to use a gun....(you can kill yourself and our sense of humor)

Why?
Speed why not but, Q6600 is entry level quad, it's not like using the lowest speed barcelona vs the highest intel offer...
FSB, should he add an IMC inside his Q6600? Why not increase the fsb on the max level with stock cooler on both platforms...
4 cores vs 8 cores...
When I buy a cpu, I'd like to now how it performs for a given price, not how it performs if it would be worst...

[JohannesRS]

you are right when it comes to finding which one is the best performance/cost ratio plataform.

But you are wrong if you consider which is the best plataform. then, you need to look to how far can it get in clock (and it will take some time for us to know in any situation), and how well they perform at given clocks.

[accord99]

Hold on a sec here. Am I reading this right: in C10 a 2.27GHz C2Q is faster than TWO Quads K10 at 2.0GHz?

:scratches head:

It was overclocked to 3.4GHz, EIST probably kicked in while the cpu-z screenshot was taken.

[tictac]

with 8core it will need to go by ht link so speed up not as fast as native 4core speed up. in unganged mode it use 2imc. in ganged mode it use 1imc ...i guess.. hehe?