@iscorefilm said:

I'm not trying to make fun of CM, or any other Intel fan on here..

i don't care aboput if it is intel or not, my favourite design would be a power-7 btw. if someone would support it ...

to my knowledge even the Magny Cours provide only 2 memory channels and i don't have any comparable tests related to its integer and floating point calculation units ... i'm not willing to spend money on a machine just to get confirmed it still doesn't work, especially MIR.

btw: AMD support for MIR is not an issue (iirc it actually runs on AMD with a few changes), it is the speed of filling calculation buffers and data throughput to/from memory and caches that counts.

@iscorefilm said:

I am running two AMD 8-core server CPU's (Magny Cours)

1) I have yet to meet an Intel computer as good as my own.

And I have yet to "meet" an AMD computer as good as my own. [;)]

DG

"iirc it actually runs on AMD with a few changes" - what changes? Changes to the software that have been done, or by users, settings, etc?

power-7 is cool, but like you said... I have only briefly read about the Tile-GX by Telera but I think that recent architecture developments are exciting anyway.

Magny Cours has 4 memory channels, just to clarify.

@Another User said:

Magny Cours has 4 memory channels, just to clarify.

right, but look here: one node (6 cores) has 2 DDR3 memory channels, the package (2 nodes, 12 cores) has then 4 DDR3 memory channels.

i7 has 3 DDR3 memory channels per node (4 or 6 cores without HT, 8 or 12 cores with HT) and this is what i was talking about:

it is critical to get data from harddisk to memory, from memory to processor and back again, many powerful cores (in fact calculation units) don't help if data transport is the bottleneck.

the change of the (QPI) architecture finally was what made MIR possible - before the i7 arrived we had 2 sockets of 4-core XEONS 3.2 GHz and dual channel DDR2 memory @ 1600 MHz, which was not enough (didn't matter if with or without HT). also the significant changes of HT architecture for the i7 helps much to get the calculation units filled with data in time.

i'm sorry, but AMD has always been a great number cruncher for integer, but not to be preferred for floating point calculations ... this may change of course as time goes by.

Thanks for the info, it's very useful.

I know this isn't exactly a new thread but I was reading about Hypertransport and came across something that now has me confused.

HT3.1 (2008) has the same bandwidth. If that is the case, then why is MIR not just as supported for AMD?

I'm not intending any kind of ignorant debate here... it's just that usually when Intel fanboys argue Intel superiority to me they usually give statistics that are either unreliable (intel-compiler, etc) or that are outdated, or compare the newest Intel specs to previous generation of AMD specs, or something along those lines.

I'm not accusing you of having a bad argument with what I just said... simply saying that because of a bad history of debate with people I simply want to understand exactly why MIR wouldn't be supported on AMD cpu's - Not to feel the need to defend some beloved company, but simply cause I feel that my AMD setup is perfectly fine and my own experience has been that several i7 setups I have worked on have been more than problematic. I am the first person to realize that the system board, audio interface, etc have just as much to do with any problems I might have had (if not more)... but simply that because 1) I usually hear crap arguments with intel/amd - 2) I have a great experience with AMD - 3) Every Intel argument provided to me has never stood on very solid ground - and 3) the QPI point you brought up doesn't seem to be an accurate cause of MIR incompatibilty with AMD based on the QPI and Hypertransport bandwidth point I brought up.

Lastly, again not accusing... simply saying from my personal experience and where I am looking- that doesn't seem right. And for a potential buyer of MIR or MIR Pro especially. I think it's a perfectly valid point to address. I certainly might be wrong, I just want to know for sure. The Bulldozer architecture, though specs haven't truly been released or tested, is being touted by AMD as having a "new memory controller with higher bandwidth". I'm wondering about Bulldozer and MIR as well but I guess I'll have to wait on that.

well, to my knowledge the bulldozer architecture was introduced resp. announced 2 weeks ago at computex and a few motherboard vendors actually presented boards for the A and FX series.

whereas HT has finally been expanded to 6.4 GT/s i still can see only 2 memory channels, I/O virtualization has been improved with chipsets 99x and such ... so lets see how that does in the real world aside from the gaming scene.

IIRC a german user ist running MIR with a not too big setup on an AMD computer but especially for MIR data throughput between caches, processor and memory is crucial for receiving oustanding results, so IMO it doesn't make too much sense to support configurations which allow expecting just an average score.

if the new processor generation becomes available and we have some money left we'll maybe testdrive the stuff .again ..

lol, thanks for the response! I know it might seem tedious of me to have asked, but I firmly believe in informed decision making... so thanks!

 My AMD six-core Sempron II 1090T seems to be working fine for me, except that the Realtec LAN that's onboard the ASUS Motherboard isn't cutting it. So I put an order in for a PCI-E INTEL Gigabyte CT Desktop Adapter today. I'm feeling confident that this will take care of my LAN issue. I understand that there have been issues with Realtek NIC's. :)

 Yes, I had problems with clicks, etc. with the Realtek onboard LAN on an Asus Rampage Gene II motherboard. Installed the Intel pci cards and all clicks went away.

Mahlon