AMD FX-8320E Performance On Linux

[drSeehas]

... the FP pipeline ... The integer pipeline can't function without it. ...

You have to explain this to me.
There is code without a single FP instruction.

[drSeehas]

... ECC-capable DRAM controller. While not all MBs advertise this feature, in fact most AM3+ MBs would actually work with unbuffered ECC RAM modules. ...

AFAIK for non-server MBs only ASUS support ECC.

[duby229]

You have to explain this to me.
There is code without a single FP instruction.

That's true, you could craft some code that doesn't use the FP pipeline. Then you'd have to make sure the compiler you choose doesn't generate any instructions for any extension that uses that pipeline.

I doubt it would be easy, but I'm sure it could be done with enough effort.

[duby229]

AFAIK for non-server MBs only ASUS support ECC.

I'm pretty sure since Socket939 any AMD board supports ECC. My Biostar 939 board did. My ASUS AM2+ did. My Gigabyte AM3+ does.

EDIT: You're probably thinking about registered RAM, but you do need a Server board for that.

[duby229]

The HT clock has no relation to the southbridge.

Although he called it wrong, he was talking about the 200mhz reference clock.

[profoundWHALE]

In case no one's heard, more AMD stuff has been leaked.

• Built on 14nm
• Utilize DDR4
• Up to 95W
• Opt for Simultaneous multithreading as opposed to Cluster Multithreading

And on top of that, the slim PS4 (or what it probably looks like has been leaked)

[drSeehas]

I'm pretty sure since Socket939 any AMD board supports ECC. ... My Gigabyte AM3+ does.

Which Gigabyte AM3+?
How did you test ECC?

EDIT: You're probably thinking about registered RAM, but you do need a Server board for that.

No, I wrote "non-server MBs"!

[Michael_S]

In case no one's heard, more AMD stuff has been leaked.

• Built on 14nm
• Utilize DDR4
• Up to 95W
• Opt for Simultaneous multithreading as opposed to Cluster Multithreading

And on top of that, the slim PS4 (or what it probably looks like has been leaked)

All of that would be great news, but remember:

1. Intel isn't standing still. It does no good to hit market in 2016 with something that can match Broadwell just in time for Cannonlake (Intel's planned Broadwell successor, on 10nm) to arrive and maintain Intel's lead.

2. Aren't most of the other manufacturers having a high defect rate with their sub 32nm parts? A jump right to 14nm makes a ton of sense for AMD, the question is whether GlobalFoundries or whoever they're using for manufacturing can actually pull it off.

[duby229]

In case no one's heard, more AMD stuff has been leaked.

• Built on 14nm
• Utilize DDR4
• Up to 95W
• Opt for Simultaneous multithreading as opposed to Cluster Multithreading

[IMG]---[/IMG]

And on top of that, the slim PS4 (or what it probably looks like has been leaked)

That graph is so misleading though. Don't let it fool you into thinking that SMT doesn't have serious drawbacks. On pipeline flushes, both threads get flushed. Stalls are worse on SMT architectures. AMD's architectures have historically been terrible at recovering from flushes.

[duby229]

Which Gigabyte AM3+?
How did you test ECC?


No, I wrote "non-server MBs"!

Google is your friend man.

The board I'm using now with ECC RAM is a GA-785GMH-UDT.

If you have more than like 256MB of RAM it should be ECC. I don't know why ECC isn't already part of the standard, it should be. It should hav been a long time ago. It has to do with the density of the fabrication process that makes single bit errors unavoidable. Once that threshold was hit, ECC became necessary. That was a long time ago.