High Speed DDR2: Buffalo and Crucial Deliver 1000+
by Wesley Fink on July 7, 2006 12:05 AM EST- Posted in
- Memory
Buffalo PC2-8000 C5: Stock Memory Performance
The current Intel platform, based on NetBurst technology, does not really allow for easily testing different memory speeds at the same CPU speed. Instead most top motherboards provide a wide range of memory ratios that match available DDR2 memory. Most end-users select the memory ratio that matches their DDR2 memory speed. For those reasons, we first test all of the stock ratios at the fastest stable timings we can achieve at the given ratio. With ratios, CPU speed remains the same at 3.46GHz in our memory test bed, and memory speed is varied by selecting different ratios.
There are some downsides to this approach. With the memory controller in the chipset, instead of part of the processor as in AM2 systems, there is a small performance penalty for speeds other than a 1:1 ratio (DDR2-533 in this case). However the penalty is in reality very small and memory scales nicely through the various speed options.
The Buffalo PC2-8000 is rated at DDR2-1000, so the 1000 speed was also tested. To achieve that speed the 889 memory speed was selected at the standard 1066 FSB (266 setting quad pumped). The FSB was then overclocked to 300 (1200 FSB) resulting in a memory speed of DDR2-1000. The Buffalo FireStix clearly performed much better at DDR2-1000 than the rated timings of 5-5-5-15. Our testing found complete stability at DDR2-1000 at 4-4-5-12 timings. This is not quite as good as OCZ at 4-3-4 but it is still very fast for memory running at DDR2-1000.
The fastest possible timings with today's DDR2 are 3-2-2. If you look closely you will see the Buffalo FireStix work fine at the fastest timings that can be selected at both DDR2-400 and is nearly as fast at the stable 1:1 memory setting of DDR2-533 with 3-2-3 timings. Despite the fact that CPU multipliers cannot be dropped below 12X on the Intel platform, it is important to keep in mind that the first 5 table rows, from DDR2-400 to DDR2-1066 represent a constant CPU speed of 3.46GHz with memory running at faster speeds through ratios. 1:1 should always have the least overhead in the current Intel universe, but the DDR2 memory still scales well by speed over the full range. Memory bandwidth improves with memory speed, as does raw calculation performance. However, this does not always translate into linear gaming performance curves in Half Life 2: Lost Coast and Far Cry.
The Buffalo FireStix topped out at DDR2-1100, which is the same top speed achieved with our OCZ PC2-8000 EL. Timings were at the rated 5-5-5-15. Performance was excellent at DDR2-1100, but it was not the fastest performance found in our testing. However the FireStix matched the OCZ as being a memory that can reach the DDR2-1100 plateau.
The distinction for fastest overall performance goes to the rated DDR2 speed of 1000. At that speed the CPU is running at 13x300 or a 1200FSB, which is faster than the stock FSB of 266 or the 100 overclock from 1067 at 275 FSB. No doubt the DDR2-1000 performance was partly the result of faster memory timings and partly the result of the higher CPU frequency. 300x13 is 3.9GHz, which is certainly a good deal faster than the stock 3.46GHz.
There are other possible approaches to overclocking DDR2 memory on the Intel platform. Probably the most logical, given the memory controller resides on the Intel chipset rather than the processor, is to look at overclocking at a 1:1 ratio. For our test setup that means a DDR2-533 base setting. With the added ratio flexibility of Core 2 Duo Extreme we plan to adjust memory testing to better examine overclocking capabilities of DDR2 memory.
The current Intel platform, based on NetBurst technology, does not really allow for easily testing different memory speeds at the same CPU speed. Instead most top motherboards provide a wide range of memory ratios that match available DDR2 memory. Most end-users select the memory ratio that matches their DDR2 memory speed. For those reasons, we first test all of the stock ratios at the fastest stable timings we can achieve at the given ratio. With ratios, CPU speed remains the same at 3.46GHz in our memory test bed, and memory speed is varied by selecting different ratios.
There are some downsides to this approach. With the memory controller in the chipset, instead of part of the processor as in AM2 systems, there is a small performance penalty for speeds other than a 1:1 ratio (DDR2-533 in this case). However the penalty is in reality very small and memory scales nicely through the various speed options.
| Buffalo (Stock Memory Ratios) - 2x1GB Double-Bank | ||||||||
| CPU Ratio at 3.47 GHz |
Memory Speed |
Best Timings (Voltage) |
Far Cry (fps) |
Sandra Unbuffered |
Sandra Buffered |
SuperPI 2M Mod 1.5 (seconds) |
HL2 Lost Coast (fps) |
Quake 4 (fps) |
| (4:3) | 400 DDR2 | 3-2-2-5 1.8V |
62.1 | INT 2816 FLT 2847 |
INT 5546 FLT 5523 |
88.6 | 84.8 | 68.7 |
| (1:1) | 533 DDR2 | 3-2-3-8 2.1V |
63.8 | INT 3489 FLT 3508 |
INT 6476 FLT 6476 |
86.3 | 87.2 | 74.3 |
| (4:5) | 667 DDR2 | 3-3-3-9 2.1V |
64.2 | INT 3980 FLT 4014 |
INT 6589 FLT 6626 |
85.2 | 88.7 | 76 |
| (2:3) | 800 DDR2 | 3-3-4-10 2.1V |
64.6 | INT 4298 FLT 4368 |
INT 6744 FLT 6714 |
84.2 | 89.3 | 75.7 |
| (1:2) | 1067 DDR2 | 4-4-5-14 2.3V |
65.7 | INT 4559 FLT 4620 |
INT 6829 FLT 6826 |
83.5 | 90.9 | 78.9 |
| (3:5) OC 300x13 |
1000 DDR2* *Rated Speed |
4-4-5-12 2.2V |
69.9 | INT 4886 FLT 4937 |
INT 7651 FLT 7635 |
75.3 | 99.9 | 86.8 |
| Highest Mem Speed (1:2) OC 275x13 |
1100 DDR2 | 5-5-5-15 2.35V |
67.3 | INT 4614 FLT 4678 |
INT 7034 FLT 7031 |
81.3 | 92.8 | 81.2 |
The Buffalo PC2-8000 is rated at DDR2-1000, so the 1000 speed was also tested. To achieve that speed the 889 memory speed was selected at the standard 1066 FSB (266 setting quad pumped). The FSB was then overclocked to 300 (1200 FSB) resulting in a memory speed of DDR2-1000. The Buffalo FireStix clearly performed much better at DDR2-1000 than the rated timings of 5-5-5-15. Our testing found complete stability at DDR2-1000 at 4-4-5-12 timings. This is not quite as good as OCZ at 4-3-4 but it is still very fast for memory running at DDR2-1000.
The fastest possible timings with today's DDR2 are 3-2-2. If you look closely you will see the Buffalo FireStix work fine at the fastest timings that can be selected at both DDR2-400 and is nearly as fast at the stable 1:1 memory setting of DDR2-533 with 3-2-3 timings. Despite the fact that CPU multipliers cannot be dropped below 12X on the Intel platform, it is important to keep in mind that the first 5 table rows, from DDR2-400 to DDR2-1066 represent a constant CPU speed of 3.46GHz with memory running at faster speeds through ratios. 1:1 should always have the least overhead in the current Intel universe, but the DDR2 memory still scales well by speed over the full range. Memory bandwidth improves with memory speed, as does raw calculation performance. However, this does not always translate into linear gaming performance curves in Half Life 2: Lost Coast and Far Cry.
The Buffalo FireStix topped out at DDR2-1100, which is the same top speed achieved with our OCZ PC2-8000 EL. Timings were at the rated 5-5-5-15. Performance was excellent at DDR2-1100, but it was not the fastest performance found in our testing. However the FireStix matched the OCZ as being a memory that can reach the DDR2-1100 plateau.
The distinction for fastest overall performance goes to the rated DDR2 speed of 1000. At that speed the CPU is running at 13x300 or a 1200FSB, which is faster than the stock FSB of 266 or the 100 overclock from 1067 at 275 FSB. No doubt the DDR2-1000 performance was partly the result of faster memory timings and partly the result of the higher CPU frequency. 300x13 is 3.9GHz, which is certainly a good deal faster than the stock 3.46GHz.
There are other possible approaches to overclocking DDR2 memory on the Intel platform. Probably the most logical, given the memory controller resides on the Intel chipset rather than the processor, is to look at overclocking at a 1:1 ratio. For our test setup that means a DDR2-533 base setting. With the added ratio flexibility of Core 2 Duo Extreme we plan to adjust memory testing to better examine overclocking capabilities of DDR2 memory.
Facebook
Twitter
RSS
24 Comments
View All Comments
Wesley Fink - Monday, July 10, 2006 - link
Crucial has advised AnandTech that "all of Crucial's memory products come with a lifetime warranty". We have updated the review to reflect this information on the Crucial warranty.MacGuffin - Saturday, July 8, 2006 - link
It should read DDR2 1067 (1:2) Performance.
This needs to be fixed on Page 10, along with the link on Page 9 that points to page 10, and the Article Index drop-down list.
Excellent Review, nonetheless.
JarredWalton - Saturday, July 8, 2006 - link
Fixed, thanks. :)PLaYaHaTeD - Saturday, July 8, 2006 - link
I thought since the front side bus of the 965 is 1066, it would be the 'Holy Grail' to have the memory running at 1066 as well. Wouldnt this make it synchronous again? What am i missing?MacGuffin - Saturday, July 8, 2006 - link
Synchronous Operation (meaning FSB:DRAM Ratio at 1:1)266MHz FSB -> 266MHz RAM Speed -> 533MHz DDR2
The 1:2 Divider (which isn't synchronous) yields 1066MHz
266MHz FSB -> 533MHz RAM Speed -> 1066MHz DDR2
Am I right or have I gotten it wrong? I haven't used Intel since I got this Socket 754 I am typing on.
poohbear - Friday, July 7, 2006 - link
hello, just wanna clarify if the a64 can actually use any of the extra bandwidth provided by ddr2 800+? is it only for bragging rights or is the a64 actually saturated for memory bandwidth & therefore this higher bandwidth provides performance improvements? thanks in advance.Wesley Fink - Friday, July 7, 2006 - link
The A64 does exhibit tremendous DDR2 bandwidth with the on-chip DDR2 memory controller, and memory bandwidth continues to improve as speed goes up. However, as we found in our testing of the AM2 in the DDR2 vs. DDR article, the AM2 design is not memory bandwidth starved, and the extra memory bandwidth makes almost no difference in real-world performance on the current AM2 platform. The improved memory bandwidth may make more of a difference in future AM2 designs.lopri - Friday, July 7, 2006 - link
I thought this issue was mentioned in the article but I couldn't find it when I re-read it. I know on intel system the memory running slower than 1:1 will result in small penalty, but how about memory running faster than FSB? I vaguely remember that I've heard somewhere it's better than 1:1 cause that way memory "pushes" or "rushes to" FSB. Another theory I've heard is that faster memory can make up for possible performance loss on FSB subsystem, leading to less CPU idle time. According to this review, regardless the ratio, the performance seems to increase linearly to memory speed increase.So the questions being:
1. Is 1:1 the most ideal ratio without "waste"?
2. Or a slightly higher memory speed than FSB (such as 4:5) better than 1:1, preventing possible CPU idle time and "pushing" the data at the same time?
3. Or under the same CPU/FSB speed, the faster the memory the better the performance - indefinitely, taking advantage of faster memory speed?
I would think No.3 doesn't make sense because of the very FSB. In the end the FSB has been what's limiting both CPU and memory on Intel system. How could the performance get benefit from 3:5? In an ideal world there should be waste of 2. (5 - 3 = 2) Is the performance even better with 1:2? I can't imagine the FSB system being only 50% efficient, but is that the case?
Gary Key - Saturday, July 8, 2006 - link
Lopri,Please email me about this subject. Short story is 1:1 or 4:5 are your best ratios for the Intel platform at this time although this will change depending your choice of Conroe model. We will go over this in more detail shortly and I will respond here further once I complete some article testing.
Thanks,
Gary
Locust - Friday, July 7, 2006 - link
Very good article, but I have a question. How come you guys did not review Corsair's PC8500 memory modules. I have been using 2GB kit(2x1GB) for over a month and getting timings comparable to OCZ's. DDR2 800 runs at 3-3-3-5 memory settings on same mobo.Best si DDR 1000 @ 4-4-3-8 @ 2.2 recommended voltage.
Good to see more vendors offering these memory speeds, now let's just hope prices will get under $400 :-)