For our test, we will use this 512MB module in our P4 rig which consists of:
- Intel P4 2.4B Confidential (multiplier unlocked)
- Asus P4B533 Motherboard
- Abit GF4 Ti4200 64MB
- 2 X Maxtor 80GB ata/133 7200RPM HD, RAID 0
- Windows XP Professional SP1
That is correct, we have a multiplier unlocked P4:
It allows us to lower the multiplier and raise the FSB, much like our AMD friends. For people with the Asus P4B533, you can obtain the 3:4 divider by following this hack.
All tests will be conducted using the manufacturer rated speeds. This is accomplished simply by setting our SPD setting to enable in our motherboard BIOS. This allows the motherboard to grab all memory data like module size, speed, voltage, etc. directly off the EEPROM chip on the DDR module itself. Often times, there are arguments about the performance increase in setting your memory to aggressive settings. This basically means you are lowering the latency and precharge time within your motherboards BIOS settings for your ram. I normally set the settings to SPD and let the manufacturer specs take over. We are here to see if this module can really reach PC3500 ratings and perform reliably, not beat it senseless.
Our test will use Sisoft Sandra and PCMark2002. A lot of times people do use 3DMark to bench ram, but that relies heavily on your video card, which I thought was not what this review is aimed at.
We compared this module to a 512MB Samsung PC2700 (DDR333), which maxed out at DDR 412 (206MHz X 2). We compared both with our CPU running at 18 X 155.
- RAM INT: 3096
- RAM FLOAT: 3092
- RAM INT: 3091
- RAM FLOAT: 3091
As you can see, both scores are nearly identical, with Samsung actually nudging ahead a few points. Here are the PCMark2k2 scores:
- CPU: 6835
- Memory: 6650
- HD: 1738
- CPU: 6839
- Memory: 6675
- HD: 1779
Again, very close and nearly identical. With this initial test, it is noted that there isn't any extra bandwidth or speed at 155MHz FSB. Now the fun part is that, for the Samsung module, anything higher than 155 caused a BSOD.
Swapped out the Samsung and installed the BuffaloTech module again. Raised the FSB to 162 max and it was stable. Anything above that and it caused WinXP errors during or after booting. Does that mean the ram can't handle the stated speeds? No, it just means our CPU can't go any higher at that combination. So we dropped the multiplier down to 17 and managed to raise our FSB to 168, which is DDR 446 [ ((168 X 4) / 3 = 223) X 2 ]. Here are some benches:
From the above Sisoft screenshot, we actually operate faster than PC1066 RDRAM, at a fraction of the cost.
So where does this lead us to? With the added benefit of the larger memory bandwidth, our P4 system has never felt a larger increase in performance. Compared to our previous Samsung module, this BuffaloTech sample open up the gates a tad larger for us.
There are many brands and speeds available on the market, do some research and find out which brand is reliable. Even if you can obtain insane CPU speeds, a flaky ram module can crash your system without a sweat.
If you run across BuffaloTech units on your ram quest, give it a shot. They are currently negotiating with resellers in regards to pricing. Once we confirm that, we will update this review.
Some readers wanted aggressive timing benches, here is a Sisoft Sandra bench at 2.5,3,3,6 @ DDR 436.
Here's another shot at 2,2,2,6:
At more aggressive timings, the chip still surpassed the DDR433 stock rating. At this speed, it does not perform better than previous test at SPD.
BuffaloTech memory is now available at Newegg.com. Prices are quite low, best to grab some before they go out of stock :)