Madshrimps Forum Madness

Madshrimps Forum Madness (https://www.madshrimps.be/vbulletin/)
-   Articles & Howto's (https://www.madshrimps.be/vbulletin/f6/)
-   -   Intel Core 2: Is high speed memory worth its price? (https://www.madshrimps.be/vbulletin/f6/intel-core-2-high-speed-memory-worth-its-price-25607/)

jmke 1st August 2006 17:15

Intel Core 2: Is high speed memory worth its price?
 
Does the Intel Core 2 need fast DDR2 memory to perform at its best? In this article we compare different memory speed and timings to provide you with the answer. Read on to find if cheap memory is enough to feed the Conroe

http://www.madshrimps.be/gotoartik.php?articID=472

Rutar 1st August 2006 18:42

What about overclocking? Are there dividers for Core 2 865 boards as well so we can still clock our Allendales through the roof with cheapskate ram?

jmke 1st August 2006 18:47

I don't know for other boards, I975 allows for quite an adjustment range for divider settings http://www.madshrimps.be/articles/In...otke-12981.jpg

Memory Frequency; 333 / 400 / 533 / 667 / 800 Mhz

so with cheap 533 / PC4200 you run run it at 333 and get quite a bit of headroom for overclocking without the memory running out of specifications.

for the newer P965 and older 865 I can only hope there are similar settings, don't know, google ? :)

jmke 1st August 2006 19:45

I've updated the last page with info on the decrease of performance with PC5300 vs PC4200, thanks to Sintel for the clear explanation:

The Core 2 has a front side speed bus (FSB) of 266Mhz x 4 (Quadruple) “1066Mhz”, the ram is running at 266Mhz x2 (Dual Channel) x2 (DDR) = “1066Mhz”, so with PC4200 memory and FSB are running synchronized. When you use PC5300 you are no longer running synchronous with the FSB and a factor of x1.25 has to be used. The older Athlon XP from AMD also displayed this decrease in performance when running memory asynchronous, where memory latency is more important then memory bandwidth; the Core 2 technology is similar in this aspect. With the Pentium 4 the pipe lines were longer and the effect of running asynchronous which increased latency was masked.

By running the memory synchronous to the FSB you have the least amount of latency and thus performance is at its best. Why is PC6400, which is also running asynchronous faster then? Because the memory latency and speed is now that much higher that it compensates for the loss of running asynchronous and overall performance does increase. PC5300 worked great with Pentium 4 but it should be avoided when running Core 2.

jmke 1st August 2006 20:59

it's still a bit more complicated it seems:

http://www.bleedinedge.com/forum/sho...016#post180016

Quote:

Now we need to look at memory ratios.

In 800 strap we have the following ram ratios.

400,533,667,800 or as some quote them 1:1, 3:4, 3:5, 1:2

In 1066 strap we see the following.

400, 533, 667, 800 but the actual ratios have altered, 400 mode is now a down clock so the following applies.

400=4:3, 533=1:1, 667=4:5, 800=4:6

Moving to 1333 strap we see another change.

400= 5:3, 533=5:4, 667=1:1, 800=5:6
as the SYNC changes if you set a higher strap

impar 2nd August 2006 12:18

Greetings!

How can you say that the reason is the synchronicity between the RAM and the FSB, if you only tested one FSB?

jmke 2nd August 2006 12:34

if you read through the complete text of the forum post linked above, you'll see that by changing the CPU native FSB and thereby also the north bridge strap, you shift the 1:1 sync from 400 to 533 to 667; but even at 667/1333(FSB) the performance is lower, when running in sync, as the NB is running on higher latency to obtain this speed, thereby decreasing performance.

if you CAN change the NB strap, you can alter the NB strap setting, forcing it to run either slower/faster to give you more headroom for overclocking, but that something for a different article. :)

Xploited Titan 3rd August 2006 00:18

mmm, I ordered my Core 2 Duo which will arrive second half of August.

I ordered 2GB dual channel G.Skill memory DDR2-800. I took that due to the timings offered (4-4-4-12) at that speed AND it being the lowest priced (near 180€).

Was that a good move? Or was there something better to be bought? If I could clock down the timings to something like 3-3-3-8, would it be interesting enough to try it, or just a waste of time?

jmke 3rd August 2006 09:17

if you are planning to overclock, the headroom of the DDR2-800 will allow you to run 1:1 while you increase the FSB

impar 3rd August 2006 11:11

Greetings!

But then, how to explain the weird behaviour of Quake4?
The performance at 800*600 rises with faster memory, regardless of being 1:1 or not.


From the article:
Quote:

The older Athlon XP from AMD also displayed this decrease in performance when running memory asynchronous...
The synchronicity feature on Athlon XPs applied only to nForce2, not VIAs chipsets.
So, other C2D chipsets may show a different behaviour, no?

jmke 3rd August 2006 11:16

even with VIA there was an noticeable performance increase if both CPU:MEM were running in sync :)

Quote:

The results obtained by the memory subsystem in SiSoft Sandra 2002 benchmark prove our conclusions once again. DDR333 turns out much more efficient if not only the DDR SDRAM frequency grows but also the processor bus frequency. In standard VIA KT333 mode, however, that is when the FSB frequency is 133MHz, the use of faster DDR333 memory instead of DDR266 hardly tells on the performance. This observation lets us conclude that it would make much more sense to use DDR333 in Athlon systems supporting synchronously working 166MHz CPU bus.
http://www.xbitlabs.com/articles/cpu...lonxp-166.html

the Quake 4 results mirror the synthetic Sisoft Sandra benchmark where raw memory BANDWIDTH wins over lower LATENCIES :)

you can compare 667 5-5-5-15 vs 533 3-3-3-8 and will see the result difference remains consistent in all benchmarks.

RichBa5tard 3rd August 2006 11:17

Quote:

Originally posted by onetimeposter
Greetings!
....

Quote:

onetimeposter. posts 2 (0.00 posts per day)

liar! ;)

impar 3rd August 2006 11:40

Greetings!
Quote:

Originally posted by jmke
[b]even with VIA there was an noticeable performance increase if both CPU:MEM were running in sync :)
Check this article:
http://www.xbitlabs.com/articles/chi...cketa-400.html
Particularly, the FSB333 tests:
http://www.xbitlabs.com/articles/chi...eta-400_7.html
Where you see the KT600-FSB333-DDR400 faster than KT600-FSB333-DDR333.

The nForce2 chipset was built to achieve better performance at 1:1. Will look for the PDF.
Quote:

Originally posted by RichBa5tard
liar! ;)
Not really. The forum no longer recognized my username and when I tried to re-register, it didnt allowed me to use my e-mail, as it said it was in use. I asked for it to send the username/password to the e-mail and got this new username.

jmke 3rd August 2006 11:48

what was your old username?

article I linked seemed to be a bit dated, your link is better, thanks! And for the Core2 the same still applies, as DDR2 800 is faster then DDR2 533 , although DDR2 800 is not running in sync:)

raw bandwith overcomes the lower latencies, but if you check PC4200 with very tight timings (3-3-3-8) you see it's not trailing far behind PC6400.

impar 3rd August 2006 12:10

Greetings!
Quote:

Originally posted by jmke
what was your old username?
Should have been the usual one, "impar". If you have access to the member database, could you see to which username is my e-mail attributed to?
Quote:

Originally posted by jmke
And for the Core2 the same still applies, as DDR2 800 is faster then DDR2 533 , although DDR2 800 is not running in sync:)
Yeah... so the comparison to Athlon XP shouldnt be made on the article. As you can see on the XbitLabs article, the nForce2 was what made the Athlon XP work better at 1:1.
Granted, every enthusiast had a nF2, not a KT600 motherboard.
If it had to be compared to any CPU it would be the A64, for its love for tight timings.
Per this article: ;)
http://www.madshrimps.be/?action=getarticle&articID=325



Found the nForce2 PDF:
http://www.nvidia.com/object/LO_20021105_7263.html
Check at the bottom of page 2.

jmke 3rd August 2006 12:23

"No users found matching those criteria." for Impar ... weird can you find your older username in the memberlist? http://www.madshrimps.be/forums/memberlist.php?s=
I could just change your current nickname to the one you want too:)

well.. tight timings AND running in sync, as even with tight timings the 667 does not do better than 533 :)

impar 3rd August 2006 13:00

Greetings!
Quote:

Originally posted by jmke I could just change your current nickname to the one you want too:)
Then, please do change it for "impar".
Quote:

Originally posted by jmke
well.. tight timings AND running in sync, as even with tight timings the 667 does not do better than 533 :)
Ok, then.
I still think that the comparison to Athlon XP shouldnt be made and that the article should have been made using other FSBs than the default (lets face it, Madshrimps readers wont run theirs C2D at default).

Also, and this is a picky one, when I read PC3200 I still think DDR-400, not DDR2-400. PC2-3200 is the acronym I relate to DDR2-400.

jmke 3rd August 2006 13:41

Quote:

Originally posted by onetimeposter
lets face it, Madshrimps readers wont run theirs C2D at default
I disagree with you here; I run my main system at default speed.

we'll have an overclocking Core 2 / DDR2 article pending, this one is for mainstream and does help people a lot; just today a friend send me his new PC config, he doesnt overclock, he mixxed his E6600 with DDR2-800 which sky rocketted the price (€180 for 512Mb :o) so .. yes this article is for our readers, which (hopefully) doesn't solely include overclockers but PC enthusiasts and mainstream also

impar 3rd August 2006 14:04

Greetings!
Quote:

Originally posted by jmke ... yes this article is for our readers, which (hopefully) doesn't solely include overclockers but PC enthusiasts and mainstream also
That would make an interesting poll for the Home page. ;)

If you can, on the upcoming C2D overclocking article, test different RAM speeds/timings. I know its a lot of work...

Regarding username, thanks. :ws:

jmke 3rd August 2006 14:12

well, the ram timings/speed have been mostly tested here, what we'll try is to see if a higher FSB on CPU helps performance (in combination with lower multiplier to keep CPU speed the same) with Athlon XP this FSB increase paid of well, it worked good with P4 too, less with Athlon 64 (if not at all) so it'll be interesting to see how the Core 2 will handle this;

here is a test by Hardware France (translated) which looks at timings/DDR2 speed and FSB speed: http://www.behardware.com/articles/6...-duo-test.html

Quote:

Originally posted by impar
That would make an interesting poll for the Home page. ;)
mainstream users would come from other sites, google searches, forums, etc; they won't see the main site, and even if they did, spend time to cast a vote. They want information served fast and clearly; spending time on reading a large page and clicking on links which "polls" is not high on their priority.

If I started a poll and the "mainstream user" got 0 votes, it would actually prove that point :p

HitenMitsurugi 3rd August 2006 14:15

The sync thing is mainly on how the memory controller is built, different memory controllers have different charasteristics. Yes you can argue that VIA had memory controllers which showed a very marginal performance improvement when running async with faster memory, however, if you compare the scores vs nforce sync, you will see that for example FSB333/RAM333 on nforce will perform faster than FSB333/RAM400 on VIA on about every benchmark that isn't bottlenecked by some other chipset factor.

The comparison with the AthlonXP is in my opinion very valid because it was the only 'short pipeline' cpu where running memory async was an option and which had a FSB/memcontroller architecture. The A64 is on an inherently different, more modern platform and does NOT have the same issues, yes it does have a short pipeline and a dependence on memory timings, but the memory controller runs at a factor of the full processorspeed, the hypertransport link to the chipset can not be compared to a FSB and doesn't get used at all in cpu/memory communication. The FSB is entirely eliminated from the equation so you don't have async issues, in essence on A64 the memory always runs "in sync"..

impar 22nd August 2006 13:18

Greetings!

Well, nForce and C2D look much better at DDR2-800 than at slower RAM speeds:
http://www.anandtech.com/printarticle.aspx?i=2820


All times are GMT +1. The time now is 14:26.

Powered by vBulletin® - Copyright ©2000 - 2024, Jelsoft Enterprises Ltd.
Content Relevant URLs by vBSEO