First review of Opteron s939

[s7e9h3n]

You are asking for an unsual amount of trust considering AMD is not known for publishing false claims about their processors capabilities...you have any data?

No, I'm not saying at all that AMD is publishing false claims - just that some info isn't necessarily available in public documents. I'm gonna stop here before I say something that'll get me in trouble.....

[Mats]

What's it gonna be? VVJ's Opteron is a Venus, while some people say it's a San Diego. AFAIK, Opterons have never shared code name with A64's, though the difference may be small in real in real world.

Edit: It's a Venus of course!!! San Diego?? Duh!

...or, does the Venus got the D revision?? Is it San Diego? Maybe they don't have different names for D and E 1xx's, just like D and E Semprons??

[drcrawn]

Once again I point you all to this

http://www.amd.com/us-en/assets/cont...docs/26094.PDF

BTW, since when are these s939 Optys FX chips?

[saratoga]

I don't think I was too clear in my statement. You're speaking in terms of purely @ the hardware level while what I meant to address was the diferences in how the memory controllers are hard coded. Trust me, while the information that's available publicly may imply that the memory controllers in the Opterons and A64's are functionally similar, their differences lie far beyond the ability to hand registered dimms........

It was quite clear that you didn't understand how the memory controller is configured in the 64 or Opteron with your "the A64's use only a true 64-bit memory controller" remark. Yet I'm supposed to believe that you have inside information (and yet somehow couldn't be bothered to read the published spec)?

Sorry, but I know bull when I see it.

Once again I point you all to this

http://www.amd.com/us-en/assets/con..._docs/26094.PDF

I looked through page 109 and couldn't see what interested you so much. Could you explain?

[OC Detective]

No real advatage to buying a s939 Opteron then, IMO. Like I said, this is nothing more than AMD switching single cpu servers to cheaper unbuffered ram.

Registered memory is tremendously expensive still...

The advantage IMHO lies in their current pricing of the lower models e.g. the 144 on the AMD pricelist is USD125 so you basically get extra cache and same frequency for the same price as a Venice 3000+
When and whether they are readily available is another matter however.

[largon]

It's a Venus of course!!! San Diego?? Duh!

Okay. Venus is an Opteron core, they do not work with unbuffered ram, only registered.
If the core on S939 Opteron 144 is actually a Venus -> The review could not have been done. "True" Opterons don't even post if unbuffered ram is installed.

So, it's it must be something else:

Theory no. 1: Only the S939 Opteron core is San Diego, S940 Venus Opterons are something totally different -> If this is the case, VVJ's prog "AuthScanner" doesn't show the core code name correct. That I doubt because he appears to have some level of official collaboration with AMD.

Theory no. 2: All San Diegos are Opteron Venus cores with some "abilities" enabled/disabled (for example requirement for registered ram). Codenames differ from their "parent cores" (like crippled Venices aren't Venices, but Palermos).

I prefer the 2nd theory.
I doubt AMD would design one core for Opteron and another for FX-line if they could design one core (that is the Opteron core) and disable all
"unnecessary" functions for desktop use.

[ArcTan]

It was quite clear that you didn't understand how the memory controller is configured in the 64 or Opteron with your "the A64's use only a true 64-bit memory controller" remark. Yet I'm supposed to believe that you have inside information (and yet somehow couldn't be bothered to read the published spec)?

Sorry, but I know bull when I see it.

I do believe he was one of the first to have a venice in his hands well before they were sold on the market

[adz]

i think he knows more then we think

[Gogar]

No question about it. Opteron 175, 170, and 165 are the only dual core models for the 100 series. http://www.amd.com/us-en/Processors/...26_609,00.html

Too bad you can't use non-ecc on 939 Opterons.

Link? Where did you hear that you can't use non-ecc on 939 Opterons?

[Mats]

Okay. Venus is an Opteron core, they do not work with unbuffered ram, only registered.
If the core on S939 Opteron 144 is actually a Venus -> The review could not have been done. "True" Opterons don't even post if unbuffered ram is installed.

Not sure about that anymore. It used to be like that but now there are no S940 100's. Things have changed. Why? Well, as some already pointed out, there's probably no market for it. I think it's a true Venus. There are different cores for 100, 200 and 800 series.

[drcrawn]

It was quite clear that you didn't understand how the memory controller is configured in the 64 or Opteron with your "the A64's use only a true 64-bit memory controller" remark. Yet I'm supposed to believe that you have inside information (and yet somehow couldn't be bothered to read the published spec)?

Sorry, but I know bull when I see it.



I looked through page 109 and couldn't see what interested you so much. Could you explain?

p109 only states that the 2T bit should only be set with unbuffered dimms.

The entire document, however, has lots of information regarding the difference, especially memory related, between s939 and s940.

[drcrawn]

Link? Where did you hear that you can't use non-ecc on 939 Opterons?

It's true. Ecc or non-ecc both work on s939 Optys.

[drcrawn]

Okay. Venus is an Opteron core, they do not work with unbuffered ram, only registered.
If the core on S939 Opteron 144 is actually a Venus -> The review could not have been done. "True" Opterons don't even post if unbuffered ram is installed.

So, it's it must be something else:

Theory no. 1: Only the S939 Opteron core is San Diego, S940 Venus Opterons are something totally different -> If this is the case, VVJ's prog "AuthScanner" doesn't show the core code name correct. That I doubt because he appears to have some level of official collaboration with AMD.

Theory no. 2: All San Diegos are Opteron Venus cores with some "abilities" enabled/disabled (for example requirement for registered ram). Codenames differ from their "parent cores" (like crippled Venices aren't Venices, but Palermos).

I prefer the 2nd theory.
I doubt AMD would design one core for Opteron and another for FX-line if they could design one core (that is the Opteron core) and disable all
"unnecessary" functions for desktop use.

Venus are relabled San Diego, no if ands or buts and they use unbuffered ram, NOT registered. I keep saying that AMD has moved the single cpu server market to s939 TO GET AWAY FROM EXPENSIVE REGISTERED RAM.

BTW, THERE IS NO VENUS CORE FOR s940 OPTERONS, only TROY (200 series) and ATHENS (800 series). VENUS is strictly for s939 100 series OPTERONS.

[ozzimark]

VENUS is strictly for s939 100 series OPTERONS.

so this means that the BN in the opn code shouldn't be interpreted as san diego like one would expenct?

[clayton]

I'm pretty sure the later versions of Opteron socket 939 will work with non ECC. What I heard was from the older 939 samples.
________
TOYOTA HILUX PLATFORM HISTORY

[dinos22]

I'm pretty sure the later versions of Opteron socket 939 will work with non ECC. What I heard was from the older 939 samples.

hey clayton

that is a super workstation.....why so much firepower can i ask?

now that would be a nice folding unit for our i4memory.com team

dinos22

[clayton]

Just received my new stuff today. Gotta say its too fast for myself too.
________
Ford fn10 platform history

[dinos22]

Just received my new stuff today. Gotta say its too fast for myself too.

what the quad box......so what do you reckon you install 6-7x folding consoles for us i4memorians

[iS.Terminator]

wich one would be better then? 3000+venice or a 1.8ghz Opty?(sckt939)

[OC Detective]

Both have similar architecture and same frequency but the Opteron has extra cache - not too difficult to figure out which is better.

[Petr]

You better read the review. Opteron s939 can use ECC as well as non-ECC unbuffered memory.

Venus vs. San Diego - the problem is this mess in names. There is nothing like San Diego core. Cores are named by letters and numbers. A64 3700+ s939 uses SH8 core in E4 revision. The name "San Diego" comes to say it is a s939 chip with 1 MB cache. That name is therefore a codename for a processor, not for a core. A year back a "Newcastle" was used to say it was an s754 chip with 512 kB cache... but no matter if it used DH7-CG (512 kB L2 cache) core or SH7-CG one (1 MB L2 cache with half of cache disabled).

Opterons for s939 use SH8-E4, exactly the same core as A64s w/ 1MB L2 use. And BTW, Opterons for s940 now also use SH8-E4 core. Only difference between s939 and s940 parts are that the memory controller is configured differently. But the core is the same.

[Esso]

Can't find the review that states that Opteron for Socket-939 supports non ECC memory, to blind obvioulsy.

But I came a cross this AMD document ...
http://www.amd.com/us-en/Processors/...,00.html#85258

AMD states again and again "supporting ECC unbuffered memory"
Nothing about non ECC memory support !
But the document might not be updated ?

Confused you will be.

[Petr]

Can't find the review that states that Opteron for Socket-939 supports non ECC memory, to blind obvioulsy.

But I came a cross this AMD document ...
http://www.amd.com/us-en/Processors/...,00.html#85258

AMD states again and again "supporting ECC unbuffered memory"
Nothing about non ECC memory support !
But the document might be not updated ?

Confused you will be.

See the RAMs I used - here

[Esso]

Thanks Petr, AMD seems to have problem with the documentation.
To say the least.

As soon as I locate what ever AMD calls it .... you can see for yourself ...

AMD64 Processor Compatible DDR Memory Modules
AMD Opteronâ„¢ Processor Recommended Motherboards.

[uOpt]

Since very few 939 mainboards support ECC it is near impossible that the Opteron for 939 requires ECC. It would limit their application way too much.