Mushkin XP2 PC2-5300 DDR2 – Xtreme Performance Memory
by Steve Carmel & Wesley Fink on March 28, 2006 12:01 AM EST- Posted in
- Memory
Mushkin XP2 PC2-5300 DDR2: Overclocking Performance
Using the stock 13X multiplier of the 955 EE Presler, we were able to hit 350 FSB easily at the lowest timings of 3-3-3-10, at a 1:1 ratio, as specified by Mushkin. We could scale as high as 360 FSB when we used the 12X multiplier at 1:1 or an effective DDR-721 rate at the lowest timings.
First is a look at the improvement in bandwidth provided at 350 x 13, using the 1-to-1 memory strap. In averaging the INT and Float, there was an improvement of 2221.5 Mb/s in the buffered SiSoft Sandra bandwidth results, while unbuffered results from default to overclocked at 350 x 13 showed an improvement of 1464 Mb/s. You can also note the improvement (as expected when ramping up the front side bus speeds) of approximately 2718 Mb/s for the 16MB Memory Read scores from PCMark05, and approximately 1713 Mb/s in the 16MB Memory Write scores over that from the default values.
If you are looking for good bandwidth gains, the XP2 PC2-5300 DDR2 modules certainly offer the enthusiast that opportunity. Here are just a few of the 360 x 12 results using a 1:1 memory ratio (which was the highest that we could reach with tight timings).
What happens when you relax the timings and push higher? The Mushkin modules are very flexible; once timings were relaxed, this memory really scaled quite well.
Using a 3:2 memory strap, we reached DDR2-1048, using an FSB of 349 x 13 to obtain DDR2-1048 speeds.
Relaxing the memory timings to 4-4-4-10 allowed us in our testing to reach a front side bus speed of 372, using a 12 multiplier. Here are a couple of Everest results from that escapade. It also allowed buffered SiSoft Sandra Professional 2005 bandwidth result of 8.9 Mb/s and unbuffered bandwidth of ~ 4.52 Mb/s.
Using the stock 13X multiplier of the 955 EE Presler, we were able to hit 350 FSB easily at the lowest timings of 3-3-3-10, at a 1:1 ratio, as specified by Mushkin. We could scale as high as 360 FSB when we used the 12X multiplier at 1:1 or an effective DDR-721 rate at the lowest timings.
First is a look at the improvement in bandwidth provided at 350 x 13, using the 1-to-1 memory strap. In averaging the INT and Float, there was an improvement of 2221.5 Mb/s in the buffered SiSoft Sandra bandwidth results, while unbuffered results from default to overclocked at 350 x 13 showed an improvement of 1464 Mb/s. You can also note the improvement (as expected when ramping up the front side bus speeds) of approximately 2718 Mb/s for the 16MB Memory Read scores from PCMark05, and approximately 1713 Mb/s in the 16MB Memory Write scores over that from the default values.
If you are looking for good bandwidth gains, the XP2 PC2-5300 DDR2 modules certainly offer the enthusiast that opportunity. Here are just a few of the 360 x 12 results using a 1:1 memory ratio (which was the highest that we could reach with tight timings).
What happens when you relax the timings and push higher? The Mushkin modules are very flexible; once timings were relaxed, this memory really scaled quite well.
Using a 3:2 memory strap, we reached DDR2-1048, using an FSB of 349 x 13 to obtain DDR2-1048 speeds.
Relaxing the memory timings to 4-4-4-10 allowed us in our testing to reach a front side bus speed of 372, using a 12 multiplier. Here are a couple of Everest results from that escapade. It also allowed buffered SiSoft Sandra Professional 2005 bandwidth result of 8.9 Mb/s and unbuffered bandwidth of ~ 4.52 Mb/s.
11 Comments
View All Comments
rallyhard - Friday, June 23, 2006 - link
NOTE:Under "Anandtech Deals" at the top of each page of the review, there is a link called " Mushkin 2 x 1 GB PC2-5300 DDR2 RAM". This link will take you to Mushkin part #991382 memory, which that page will tell you has 4-4-4 timings. Newegg will tell you this RAM has 5-5-5 timings. In any case, the RAM this link refers to is not the same stuff that they reviewed. The reviewed RAM is 3-3-3 timings, which I'm assuming must be Mushkin part #991512.
artifex - Tuesday, March 28, 2006 - link
Wow, last time I had a warranty return at Mushkin, I just emailed them their serials and they shipped me replacements overnight with my credit card as collateral for the return. Of course, I bought from their website, not from a retail store... but there was no talk of packaging. What packaging? :)Has Mushkin changed owners?
bldckstark - Tuesday, March 28, 2006 - link
On page 7 you show the Mushkin memory at the top of the graph with 65.1 and all other memories have lower times. To be consistent with all of your other graphs the rankings are apparently reversed since the super pi results are "lower is better", and the Mushkin should have finished last, not first and be on the bottom of the graph, not the top. Unless I forgot to take my medicine today, again, and I am reading the article upside down, again.Nice article. I'll keep waiting for the latencies to come down so that we might see some life out of AM2 against Conroe.
Wesley Fink - Tuesday, March 28, 2006 - link
The Super Pi chart is now in Ascending order as it should be. Thanks for bringing this to our attention.Pirks - Tuesday, March 28, 2006 - link
I recall buying some Mushkin stuff a long time ago. A HUGE price AFAIR and then one DDR stick died and they replaced it no questions asked. But now, after checking out their prices and seeing noname 2 GB DDR sticks for $200 CAN TOTAL (!) in my local store... well, guess which way I went. Now 2GB with the same price as Mushkin's EXTRADUPERFAST 512MB is soo sweet, no swap in XP and stuff... not saying Mushkin is bad, it's just so nice to spend $$$ on the AMOUNT of memory, not the speed. Sometime I'll get 4GB of noname sticks and make myself a hefty RAM disk and outrun those Raptor boasting kids, hehe :))artifex - Tuesday, March 28, 2006 - link
Pirks, spend a little more for name brand, even if not high-performance. That $20 or so you save up front isn't going to look so good if you ever have a memory glitch that causes you to lose data.I buy Mushkin, but only for machines that I'm going to be building at the margins of their rated specs, intend to overclock, etc. I feel like Mushkin sticks might be just a little more solid for that.
I don't overclock all my stuff, though, and for just regular performance that is solid, like in my Mac Mini, I use Crucial. I wouldn't go with a no-name. And I never, ever, would buy loose sticks from a place like Fry's. :)
Pirks - Wednesday, March 29, 2006 - link
That's because you don't know what www.memtest.org is :P
Inkjammer - Tuesday, March 28, 2006 - link
I've not seen enough of a performance boost from the "elite" memory to justify the price. Now, I say this, and I have 2GB of Kingston HyperX memory in my system. I swapped it out for 2GB of regular Kingston Value RAM and the performance difference was minor at best. If I weren't running a benchmark on it via SANDRA I'd never have even noticed.It seems to be geared to overclockers primarily, or people who just wanna pimp their rig. I think you could take the extra $200 for 2GB and invest in a beefier CPU or graphics card and really, REALLY see a performance difference.
WxChaser - Tuesday, March 28, 2006 - link
Mushkin was the first memory company to produce enthusiast memory using the Elpida chips, allowing them to overclock very well at tight timings for DDR2. These modules will provide excellent performance for Intel DDR2 motherboards, and migrate quite well to the AMD socket AM2 when those platforms are released later this year.Please note that Mushkin has revamped their main web presence recently, and markets to 3 main target groups now. These are enhanced performance, high performance, and extreme performance. The XP2 PC2-5300 modules we tested were the extreme performance category - in other words aimed at the overclocking and gaming niche markets primarily.
Inkjammer - Tuesday, March 28, 2006 - link
I'd still love to see a comparison on RAM between standard and enthusiast and stock speeds. I know that the built in spreaders are fantastic for heat dissapation, but how much added performance is gained for the extra money?Granted, I have enthusiast RAM, so... don't think of me as a hypocrit, but I still ponder just how much better this RAM is if you're not overclocking to the MAX.