N00b's Guide to Conroe Based Performance Computing and OC Results

[Gautam]

Ross, you might want to re-evaluate that statement after reading this comparison between a Conroe at 370MHz FSB vs a Merom at 228MHz FSB. They are nearly identical in everything but raw memory bandwidth.

 

[Gautam]

Yes. That is correct. For example, the older AMD XP-M's were much loved chips. The way I see it, DTR's hold great overclocking potential. In my experience, the chips which achieve the max OC were the ones which had the lowest startup VID. There is no reason to posit otherwise. However, its the motherboard department which may cause problems. This may or may not be fixable with a BIOS mod. I was discussing this exact same issue with Gautam. My initial choice was to get the DTR version of the Conroe (Mermon), but he alerted me to the problems he was facing with his Dothan setup. So, a DTR may not always be the best option.


Excellent info Ross! I've added it to the main thread.

There are no DTR's. Merom is a true mobile part. There are a lot of problems with it and when its all said and done Conroe, the desktop part is the way to go.

Sorry to keep the responses short...lets just say the mailman brought some interesting goods that are keeping me occupied today...

 

[Ross]

Ross, you might want to re-evaluate that statement after reading this comparison between a Conroe at 370MHz FSB vs a Merom at 228MHz FSB. They are nearly identical in everything but raw memory bandwidth.

I *always* reserve the right to be wrong, but I did see that already and am sticking with what I said The last time checked, mem bandwidth is important. I was a little surprised the Merom hung that well, but you can easily see the difference in more mem-heavy tests like SPi 32M and encoding...Conroe kills it with the extra FSB. There are a couple of things that remain to be seen: full 3Ds (Merom does nicely on Cinebench though) and just what kind of a max OC Merom has compared to Conroe.

OK, I'll caveat the statement that Merom would need to clock "significantly better" than Conroe....that would really depend what you're doing with it. For "overall performance" (and probably max clocks?), Conroe is still clearly the way to go.

What did you get Gautam? I am already tired of running a billion gigahertz and getting stomped by Yonahs and Meroms at 1/2-1/3 the speed, LOL

 

[RedDragonXXX]

What's the deal with Nforce590?

As I'm a big fan of Nvidia's chipsets I'm thinking this may be the way to go for me, unless you guys give me some input on why I shouldn't.


Anyways Nade this is an awesome thread as I was looking for something like this for awhile now since it will be gold for us hardcore AMD users that are switching to Intel for the first time.

This should be sticked right away!!!!

 

[ziggo0]

You and me both Super Nade, I completely agree with you. Usually me making a huge platform switch solely based on speculation will never be found in my morals, but DAYUM, Conroe looks 31337! Chances are I'll never make it to Dual Core with my AMD setup due to the prices of the chips so Conroe is going to forfil the pleasures I never got !

 

[Jame Gumb]

I've always understood all Nforce chipsets to be junk for OCing Intel's. Intel ocing mobos are like grits, you can't serve it up plain, you got to put a little salt on it.

 

[Ross]

As I'm a big fan of Nvidia's chipsets I'm thinking this may be the way to go for me, unless you guys give me some input on why I shouldn't.

if (($proc == 'Core') && ($chipset == 'Intel')) {
  $error = false;
  $ati_cf = true;
  $nv_single = true;
  $sli = false;  // Because NV is lame
  $oc++;
  $oc = bcpow($oc, 10);  // raise $oc to the 10th power
} elseif (($proc == 'Core') && ($chipset == 'NVIDIA')) {
   if ($current_owner == 'AMD') {
     echo 'This is not AMD so NV chipsets are better suited for garbage disposals.';
   }
    $error = true;
    $bank--;
    $sli = true;
    $oc = NULL;
    $oc--;
    $oc = ($oc-1000000);
}

Sorry, I dunno why I felt like doing it like that Yeah, don't get an NV chipset board for Intel proc...trust us. You'd be better off giving up SLI and running a single NV card than trying to run SLI in an NV board with an Intel proc.

 

[Evilsizer]

o where did that link go to a 975 with sli nv cards... all we need are hacked drivers....

 

[Gautam]

I *always* reserve the right to be wrong, but I did see that already and am sticking with what I said The last time checked, mem bandwidth is important. I was a little surprised the Merom hung that well, but you can easily see the difference in more mem-heavy tests like SPi 32M and encoding...Conroe kills it with the extra FSB. There are a couple of things that remain to be seen: full 3Ds (Merom does nicely on Cinebench though) and just what kind of a max OC Merom has compared to Conroe.

OK, I'll caveat the statement that Merom would need to clock "significantly better" than Conroe....that would really depend what you're doing with it. For "overall performance" (and probably max clocks?), Conroe is still clearly the way to go.

What did you get Gautam? I am already tired of running a billion gigahertz and getting stomped by Yonahs and Meroms at 1/2-1/3 the speed, LOL

You might be right, just like last time...where you might have been, still not sure though, didn't get to use netburst for long enough.

Though TBH encoding and Sandra mem are probably the two least useful things for me. I'm not sure why it falls so heavily in 32M...he might have set something up wrong, dunno. I think the Merom should be half a min quicker than what he put up at least.

But even in the encoding test, the Conroe was definitely faster, but it didn't totally crush the Merom or anything. That's how it is across the board it seems.

Anyways I've said time and time again that Conroe is the obvious choice for our purposes. Merom ain't bad though.

 

[Ross]

o where did that link go to a 975 with sli nv cards... all we need are hacked drivers....

Yeah, like they're just floating around everywhere That's my main problem with NV. SLI was doable on Intel chipsets when SLI was first released. NV specifically and purposely removed Intel chipset support from their drivers along with all the other "bugs" I guess you can't blame them since they want to sell their chipsets too, but their chipsets for Intels are a$$, yet the AMD versions work flawlessly.

If enough AMD+SLI users jump ship for Conroe and start pawning off SLI setups (or even 1 card out of them) and/or don't buy NV boards because their Intel chipsets stink, NV will be paying for it. If NV was smart, they'd add back driver support for Intel chipsets on their own right now (or by June) or REALLY step up their Intel chipset so people might even consider buying their Intel boards. I guess the only people it would matter for are those with SLI that want to switch since single NV cards seem to rock-n-roll on Intel chipsets just fine, right Fishy?

Anyways I've said time and time again that Conroe is the obvious choice for our purposes. Merom ain't bad though.

TBH myself, I am still a little bothered Merom is that close to it, LOL. Maybe the more efficient core or the huge/shareable L2 is reducing the impact of substantial FSB increases? On P4, I guarantee you it's night and day with/without high FSB.

I don't think he mucked up the Merom 32M. There's a lot more bandwidth on Conroe, but since nothing else really changed (mem speed/timings), all increasing the bandwidth did was drop the efficiency (75%->49%). If he had that b/w increase and was able to maintain efficiency with faster/tighter mem, it would be a whole different story. In any event, look what the FSB does for the random access latency (Merom: 78ns -> Conroe 60ns). On 32M with a lot more going through mem, that's vitally important and it shows in 32M.

Agreed though, Merom does **extremely** well. Conroe does just as well and better on somethings.

Now, what did you get or is it a secret? Which section should I be looking in for posts? Mem? 3Ds? SPi? Extreme cooling?

 

[NODES]

o where did that link go to a 975 with sli nv cards... all we need are hacked drivers....

they are out there, even for the 955X but no one wants to give em out

 

[trueplaya4ever8]

Nice start SuperNade, Good lite reading on conroe. So many people are going conroe, this should prove for quite a fun summer.

 

[gone_fishin]

I read an interesting tidbit somewhere some time ago by someone who is using a Conroe setup.
You can set the processor in bios to only run one core but that one core will use all of the l2 cache and not have to share it with the second core. Will be handy for gaming where most games today are not optimized for dual core.

 

[Ross]

That's done on the fly, no BIOS settings involved. Either core can use 100% of the L2 cache and both can access any portion/amount of it at any time.

they are out there, even for the 955X but no one wants to give em out

That's what I was talking about. Early SLI worked on 955 boards long before Crossfire was even out....then NV changed their drivers.

 

[El<(')>Maxi]

I'm going to guess and say I think Gautam will get much higher clocks out of it than the previous owner, whatever it is

 

[Super Nade]

Ross, you mentioned huge shareable cache being a possible candidate for the close spacing between Conroe and Merom benchmarks. Do you think the wider pipeline also has a role to play in even-ing the playing field?

I'm still a bit puzzled by the memory access latency tests:
Conroe fall behind in the linear access tests but wins heavily in the random access tests.

With encoding, Conroe pulls ahead easily. So, encoding is really FSB dependent right?

With 3dm06, it was expected to follow cinebench although I still don't know why. I did not find any numbers for CPU cache latency in there (maybe I didn't look hard enough). It would be interesting to see if both CPU's have different cache latencies? I'm willing to bet that Merom has a lower latency which enables it to hang in there with the Conroe.

Final question, to my limited knowledge Laptops are NOT LGA775 sockets, right? Therefore, Merom must be the older skt 478(?), right? That is what Gautam mentioned when he alluded to it NOT being a DTR?

 

[Shelnutt2]

Final question, to my limited knowledge Laptops are NOT LGA775 sockets, right? Therefore, Merom must be the older skt 478(?), right? That is what Gautam mentioned when he alluded to it NOT being a DTR?

That is correct mobiles are not socket 775, they are now (correct em if I am wrong) socket 479 (480? or was that only for the SMP version of Yohan was 480). They have 478 pins, but to keep the confusion between the old standard 478 (Northwood/Prescott), they renamed the socket 479 for the new mobile chips starting with Banis.

 

[greenmaji]

That is correct mobiles are not socket 775, they are now (correct em if I am wrong) socket 479 (480? or was that only for the SMP version of Yohan was 480). They have 478 pins, but to keep the confusion between the old standard 478 (Northwood/Prescott), they renamed the socket 479 for the new mobile chips starting with Banis.

SMP Yohan???? whats that?

Yohan and memron have a switched pin but otherwise would fit in the old pentium-M socket SN, the swiched pin is why they don't, some say they would work I kinda doubt it from what I've read.

 

[BWR]

I would like to request SuperNade include information about the two chipsets that will launch with Conroe, 965 and 975X. There are obviously differences, although I haven't found concrete information, but using information that I have gathered (and some common sense) the 975X will out perform the 965...

Right?

---

*looks at Gautum's sig*

*looks at Gautum's avatar*

 

[Ross]

Ross, you mentioned huge shareable cache being a possible candidate for the close spacing between Conroe and Merom benchmarks. Do you think the wider pipeline also has a role to play in even-ing the playing field?

I'm still a bit puzzled by the memory access latency tests:
Conroe fall behind in the linear access tests but wins heavily in the random access tests.

With encoding, Conroe pulls ahead easily. So, encoding is really FSB dependent right?

With 3dm06, it was expected to follow cinebench although I still don't know why. I did not find any numbers for CPU cache latency in there (maybe I didn't look hard enough). It would be interesting to see if both CPU's have different cache latencies? I'm willing to bet that Merom has a lower latency which enables it to hang in there with the Conroe.

Final question, to my limited knowledge Laptops are NOT LGA775 sockets, right? Therefore, Merom must be the older skt 478(?), right? That is what Gautam mentioned when he alluded to it NOT being a DTR?

Possibly, the entire proc is ridiculously more efficient than P4. I just noticed that he's running different mem speeds/timings between the two tests, duh. That could be part of it on the linear mem tests. My 6400s ran real tight, but actually yielded higher bench #s at slightly looser timings. Encoding is heavily mem dependent and in turn FSB dependent, much like a big Pi.

If he ran a ScienceMark, that would show any differences in cache latencies and mem in general between the two. I can't particularly see how Merom caches would be faster, but that certainly would explain a lot. The other thing I noticed is Merom is Stepping 4 and the Conroe is Stepping 1....could be a lot read into that.

Yes, mobiles are actually S479 and the pins are on the proc. Desktops are 775 and the pins are on the mobo. You can scroll up to that link and take a look at the physical procs to see the differences.