Original Link: https://www.anandtech.com/show/1135



Some of the Memory received for our recent “Searching for the Memory Holy Grail — Part 2” article could not be fairly reviewed in that roundup. Since the base requirement was that memory perform at DDR500 or faster, fast memory that did not meet that requirement was eliminated from the review.

Mushkin has taken a different approach to the fast question with their Level II PC3500 2-2-2 Memory modules. Mushkin has aimed for the absolute fastest timings possible with memory more likely to be used with the 2.8 to 3.2GHz P4 processors, or AMD Bartons that will not likely see Front Side Bus Speeds greater than about DDR433. Another unique approach is Adata DDR450, a very unusual DDR rating to be sure. While Adata 450 could not quite reach our DDR500 cutoff in testing, it did turn in some of the most aggressive timings at DDR400 that we have seen, Because of these different approaches to performance, we decided to take a closer look at both the Mushkin and Adata memory.

Intel legitimized DDR400 with the 875/865 chipsets, and that is now an official JEDEC standard. In fact, the fastest memory that the Intel 875/865 and AMD Athlon/Barton are designed to run is DDR400. Anything faster than DDR400 is overclocking the memory or the system in one form or another. So, with the Mushkin and Adata, we will try to determine which memory is the best performing memory at DDR400. For those of you who do not overclock, this installment of our memory series was written for you. For those who have, or plan to get, one of the higher speed P4 CPUs or an AMD Barton, this will also be good information to help you buy the best performing memory for your system. We will also go back and pick up DDR400 performance from the 7 memories tested in Part 2 of the “Search for the Memory Holy Grail”.

So we are asking the question again — what is fast memory? We’ve seen that raw FSB speed definitely has a tremendous impact on memory performance in games and applications. But the other side of that coin is that memory timings can also greatly affect performance. So which is better: DDR500 running 3-4-4-8 or DDR400 running at 2-2-2-4?




Test Design

For tests of the Mushkin Level II PC3500 and Adata PC450, we are continuing the memory test suite established in our article, “Search for the Memory Holy Grail — Part 2”.

1) SiSoft Sandra Max3 UNBuffered Memory Test

The Sandra UNBuffered Memory Test, as we demonstrated in Part 1, turns off Memory Buffering schemes in an attempt to better measure raw memory bandwidth. As a result, it also correlates well with bandwidths reported with Memtest86, an industry-standard memory testing tool.

The idea of the UNBuffered Memory Benchmark is very simple — you merely turn-off all memory buffering techniques. Sandra makes this very easy to do. Select “Memory Benchmark”, right-click “Module Options”, and uncheck the 9 boxes that have to do with buffering as you see here.


Click image to view a larger picture.


2) SiSoft Sandra Max3 Standard Memory Test

The UNBuffered Memory Benchmarks are quite different form what you may be accustomed to seeing in memory testing with SiSoft Sandra. For reference, we are again including the Sandra Max3 standard Memory Test, sometimes called the Buffered Memory Test.

3) Super PI

Pure number crunching benchmarks are very useful for measuring system bandwidth. Some of the more popular number-crunchers are the MPEG/DIVX encoding tests, such as we used in our standard motherboard testing, and Super PI. MPEG/DIVX tests are very useful for a single motherboard benchmark and in cross-platform testing (Athlon vs. Pentuium4, for example). However, they are often very sensitive to the test environment or system configuration. In addition, it can be difficult to use reliably in a testing environment of a large number of conditions with the same test, such as we will be doing here in our memory testing. Super PI, on the other hand, is very simple to use and has been shown to be less sensitive to the operating system environment. In other words, we don’t have to reinstall the operating system on a clean hard drive each time we run a benchmark just to get reliable numbers.

Super PI for Windows 1.1 is a freeware program developed by the Super Computer Consortium at the University of Tokyo. The concept of Super PI is very simple. It calculates the value of pi to “x” number of places, and reports the time this calculation requires. We chose to use 2 million places in our tests. Super PI is measuring total system bandwidth, and memory is only part of that bandwidth, since the CPU has a very significant impact on results. We therefore would expect to see smaller changes in Super PI relative to larger changes in memory-only benchmark tests like Sandra.

4) Quake3 Demo FOUR.dm_66

Quake 3 Demo FOUR is one of our standard game benchmarks. You will likely be surprised how sensitive Quake3 can actually be in testing wide variations in Memory Speed. We run the benchmark 3 times, check for score consistency, repeat if we see any wide variation in individual scores, and average the 3 scores for the reported Frames Per Second (FPS) value.

5) Unreal Tournament 2003 Demo

We ran UT2003 as a further test to check correlation of our benchmarks. We are reporting UT2003 results from our Mushkin and Adata tests. UT2003, as shown in our past memory testing, mirrors relative results from Quake3.



Performance Test Configuration

 INTEL 875P Performance Test Configuration
Processor(s): Intel Pentium 4 2.4GHz (800MHz FSB)
RAM: 2 x 512MB Mushkin Level II PC3500 (DS)
2 x 256MB Adata DDR450 (SS)
2 x 512MB Adata PC4000 (DS)
2 x 512MB Corsair PC4000 (DS)
2 x 512MB Geil PC4000 (DS)
4 x 256MB Kingston PC4000 (SS)
2 x 256MB Kingston PC4000 (SS)
2 x 512MB OCZ PC4000 (DS)
4 x 256MB OCZ PC3700 GOLD (DS)
Hard Drives 2 Western Digital Raptor Serial ATA 36.7GB 10,000 rpm drives in an Intel ICH5R RAID configuration
PCI/AGP Speed Fixed at 33/66
Bus Master Drivers: 875P Intel INF Update v5.00.1012, SATA RAID drivers installed, but IAA not installed
Video Card(s): ATI 9800 PRO 128MB, 128MB aperture, 1024x768x32
Video Drivers: ATI Catalyst 3.6
Operating System(s): Windows XP Professional SP1
Motherboards: Asus P4C800-E (875) with 1.010 Release BIOS


The Memory testbed is the same used in “Search for the memory Holy Grail — Part 2” All test conditions were as close as possible to those in our earlier memory review.

While DDR500 is not currently a real option on an AMD board, we have seen recent nForce2 Ultra 400 motherboards performing in the DDR450 range (225 FSB). Thus, really fast DDR400 memory with range to DDR450 or so looks like an ideal match to the nForce2 motherboards and chips that are capable of running the new 200FSB operating speed of top Athlon processors. In addition to Intel 875 tests, we evaluated performance of the Mushkin PC3500 and Adata 450 on nForce 2 Ultra 400 with a Barton 2500+ proven to have a range to 2500 MHz on the better AMD boards.

 AMD nForce2 Ultra 400 Performance Test Configuration
Processor(s): AMD Barton 2500+
RAM: 2 x 512MB Mushkin Level II PC3500 (DS)
2 x 256MB Adata DDR450 (SS)
Hard Drives Maxtor 80GB 7200RPM ATA133 with 8MB Cache
PCI/AGP Speed Fixed at 33/66
Bus Master Drivers: nVidia nForce unified driver 2.45.
Release Date 7/29/03
Video Card(s): ATI 9800 PRO 128MB, 128MB aperture, 1024x768x32
Video Drivers: ATI Catalyst 3.6
Operating System(s): Windows XP Professional SP1
Motherboards: Albatron KX18D Pro II nForce 2 Ultra 400
Release BIOS 1.01



The Candidates

With what we now know about 875/875 memory configurations, the ideal test with greatest performance would be 4 Double-Bank modules. To be fair in our memory comparison, we therefore asked memory vendors, at the very minimum, to supply 2 DS dimms or 4 SS modules of their fastest memory for evaluation. Mushkin supplied our requested 2 x 512MB of Double-Bank PC3500 Level II. Adata supplied just 2 SS 256MB dimms, so please keep this in mind when comparing performance. 2 SS dimms on the Intel 875 platform tests will perform slightly worse than 2 DS dimms. 4 Single-Bank dimms on the 875 are required to equal performance of 2 Double-Bank dimms.


Memory Specifications
Adata DDR450 Mushkin PC3500 Level II
Number of Dimms & Banks 2 SS 2 DS
Dimm Size
Total Memory
256 MB
512 MB
512 MB
1 GB
Rated Speed DDR450 DDR433
Rated Timings 2.5-4-4-8 2-2-2-?
SPD Timings 2.5-4-4-8 2.5-4-4-8
Rated Voltage ? 2.75V

In the Intel test, with each set of memory, we tested at the following settings:

1. 800FSB/DDR400 — the highest stock speed supported on 875/865 motherboards.
2. Rated Memory Speed — the specified rating of the memory modules we were testing.
3. Highest Stable Overclock — the highest settings we could achieve with the memory being tested.

These are essentially the same memory tests performed on the 7 sets of memory in “Searching for the Memory Holy Grail — Part 2”.

In the added AMD test, with each set of memory, we tested the following settings:

1. 333FSB/DDR333 — the highest stock speed officially supported by the Barton 2500+ CPU.
2. 400FSB/DDR400 — the highest stock speed officially supported by the nForce 2 Ultra 400 chipset and AMD Athlon processors.
3. Rated Memory Speed — the specified rating of the memory modules we were testing. We made attempts to reach the rated speed of the memory modules being tested.

To test overclocked stability on both the Intel and AMD testbeds, we used a run of the very demanding Gun Metal 2 — Benchmark 2, which pushes systems with its demanding DX9 routines. To be considered stable for test purposes, Gun Metal, our Quake3 benchmark, UT2003 Demo, and Super PI had to complete without incident. Any of these 4, and in particular Super PI and Gun Metal, will crash a less-than stable memory configuration.




Adata DDR450

Adata DDR450 is a very reasonably-priced memory, but somewhat difficult to find in the US market. It is readily available and widely used in many Asian markets. The Adata 450 we tested came as 2 separate Single-Bank dimms and was not supplied as a Dual-Channel kit. The dimms are packaged without heatspreaders and are assembled from blanks that are carrying an Adata ID.


Click image to view a larger picture.


The Adata is conservatively programmed in SPD at 2.5-4-4-8 timings, but the memory performs at much better timings, particularly at lower speeds. The Adata does not operate at anything less than a Ras-to-Cas setting of 3 at any speed we tested. This is not an issue with just Adata. We are finding this to be typical of most high-speed modules these days, including those advertised as low latency. Even our Corsair 3200LL rev.1.2 now has an SPD program of 2-2-3-6, with Ras-to-Cas at 3. It appears one trade off with recent memory chips is poorer Ras-to-Cas performance.


Adata DDR450 — 2 x 256MB SS Dimms
Intel 875P Performance
Memory Speed, Voltage, Timings Quake3
fps
UT2003
Flyby fps
UT2003
Botmatch fps
Sandra UNBuffered Sandra Standard
Buffered
Super PI 2M places
(time in sec)
[email protected]
2-2-3-7
311.00 192.6 68.00 INT 2590
FLT 2667
INT 4776
FLT 4744
132s
[email protected]
2.5-3-4-7
334.57 205.94 74.10 INT 2587
FLT 2611
INT 5209
FLT 5203
121s
[email protected]
2.5-4-4-8
356.90 216.42 79.55 INT 2679
FLT 2685
INT 5630
FLT 5584
113s

Since Adata is only available as SS modules and Adata only supplied 2 Single-Bank modules, performance cannot be fairly compared to 2 Double-Bank modules on the Intel 875 platform. In the larger DDR400 comparison that follows, Adata DDR450 can be compared to Kingston PC4000 2 x 256 SS, and then to 4 x 256 SS to get a rough idea of the performance improvement available with 4 SS dimms or 2 DS dimms. For best performance on an Intel 875/865 platform, we strongly suggest that you use FOUR Adata DDR450 modules.


Adata DDR450 — 2 x 256MB SS Dimms
AMD nForce2 Ultra 400 Performance
Memory Speed, Voltage, Timings Quake3
fps
UT2003
Flyby fps
UT2003
Botmatch fps
Sandra UNBuffered Sandra Standard
Buffered
Super PI 2M places
(time in sec)
[email protected]
2-2-3-4
289.8 199.82 73.27 INT 1202
FLT 1294
INT 2566
FLT 2415
124s
[email protected]
2-2-3-4
344.6 233.61 87.53 INT 1402
FLT 1503
INT 3060
FLT 2880
104s
[email protected]
2.5-3-4-6
376.43 251.44 96.26 INT 1529
FLT 1617
INT 3411
FLT 3180
94s

Performance on the Athlon test platform was generally outstanding, especially considering the price of Adata PC450 memory. We were a little surprised that we could reach a stable DDR450 on our nForce2 Ultra 400 testbed, and the increased speed did provide improved performance across the board.

These test setups were designed to test memory performance and should not be used as a direct comparison of Athlon Barton vs. Pentium4 performance. Because of the differences in architecture, our fastest P4 speed in this test is 2.73Ghz (244x12) and our fastest Barton in this test is 2475Ghz or about 3700+ on a rating scale. If you wish to compare Barton vs. P4, please compare these best Barton scores to the best P4 scores running at about 3.6Ghz.



Mushkin PC 3500 Level II

Mushkin has built an enviable reputation for dependability as a memory supplier. However, they have not been a particularly active player in the recent increases in DDR memory speeds. Mushkin PC3500 Level II 2-2-2 is the fastest memory currently available from Mushkin. We have been told that this is about to change since Mushkin will soon be introducing DDR500 memory based on a unique design.


Click image to view a larger picture.


Mushkin uses Winbond BH5 chips in PC3500 Level II. BH5 is an almost legendary memory chip that has recently been discontinued, but Mushkin assures us they have sufficient stock to continue manufacture of Level II modules for many months. It was the chip used in many of the very best low-latency modules in recent months, and replacement chips have not been able to achieve the same aggressive timings at DDR400 we have seen in the past.

There were surprises on two fronts with PC3500 Level II. First, with an advertised rating of 2-2-2 at DDR433, we wondered why the SPD was programmed at 2-4-4-8. Mushkin explained that this was done for maximum compatibility. Since we have also seen the same type of SPD programming on competing memory modules, we will accept this explanation. However, this Mushkin is not a memory that can be used with SPD timings for best performance. You must set manual SPD timings to get the most from this memory.

Second, we found the 2-2-2 rating at DDR433 a bit optimistic on the Intel 875 test platform. Mushkin explained that 2-2-2 at 433 was based on a single module in the nForce2 platforms, and we can confirm that 2-2-2 is a valid rating on the nForce 2 platform at DDR433. In fact, 2-2-2 is a fair rating in Dual-Channel mode on the nForce2 Ultra 400. In Dual-Channel mode on 875/865 motherboards, we find that 2-2-3 is the true speed rating. Mushkin has indicated they will be making changes to their web descriptions to describe better the performance of the PC3500 Level II.

Despite these concerns, Mushkin does achieve the fastest memory timings we have seen from current memory at DDR400 to DDR433. In its performance range, Mushkin PC3500 Level II is one of the fastest memories you can put in your machine.

Mushkin PC3500 Level II (DDR433) — 2 x 512MB DS Dimms
Intel 875P Performance
Memory Speed, Voltage, Timings Quake3
fps
UT2003
Flyby fps
UT2003
Botmatch fps
Sandra UNBuffered Sandra Standard
Buffered
Super PI 2M places
(time in sec)
[email protected]
2-2-2-6
328.3 198.34 70.25 INT 3091
FLT 3145
INT 4942
FLT 4901
126s
[email protected]
2-2-3-6
335.5 202.63 72.68 INT 3042
FLT 3097
INT 5111
FLT 5118
122s
[email protected]
2.5-2-3-7
350.6 211.91 77.0 INT 3159
FLT 3154
INT 5405
FLT 5424
115s


Mushkin PC3500 Level II (DDR433) — 2 x 512MB DS Dimms
AMD nForce2 Ultra 400 Performance
Memory Speed, Voltage, Timings Quake3
fps
UT2003
Flyby fps
UT2003
Botmatch fps
Sandra UNBuffered Sandra Standard
Buffered
Super PI 2M places
(time in sec)
[email protected]
2-2-2-4
297.87 203.99 74.83 INT 1260
FLT 1363
INT 2572
FLT 2437
120s
[email protected]
2-2-2-4
353.03 238.18 89.55 INT 1475
FLT 1589
INT 3073
FLT 2905
101s
[email protected]
2-2-2-5
379.5 252.31 96.51 INT 1593
FLT 1714
INT 3315
FLT 3136
94s




Performance Tests

Mushkin PC3500 Level II, Adata 450, and all of the Memory in our Part 2 roundup was compared at DDR400. Results are compared for Quake 3, Sandra UNBufferred Memory Test, Sandra Buffered Memory Test, and Super PI. The Sandra Memory Test reports 2 results for each memory test — an Integer value and a Float value. We considered adding the results together for our tests, but felt it might be difficult to compare to other Sandra UNBuffered test results.

Intel 875P DDR400 — 800FSB Memory Performance
Memory Memory Timings
& Voltage
Quake3
fps
Sandra UNBuffered Sandra Standard
Buffered
Super PI 2M places
(time in sec)
Mushkin PC3500 Level II 2-2-2
— 2 x 512MB DS
2-2-2-6
2.55v
328.30 INT 3091
FLT 3145
INT 4942
FLT 4901
126
OCZ 3700 Gold
— 2 x 256MB DS
2-2-3-6
2.55v
328.07 INT 2840
FLT 2852
INT 4821
FLT 4789
130
Kingston HyperX 4000
— 4 x 256MB SS
2.5-3-3-6
2.55v
324.30 INT 2514
FLT 2500
INT 4735
FLT 4757
131
Corsair XMS4000
— 2 x 512MB DS
2.5-3-4-5
2.55v
319.63 INT 2636
FLT 2647
INT 4702
FLT 4740
132
Kingston HyperX 4000
— 2 x 256MB SS
2.5-3-3-6
2.55v
320.13 INT 2589
FLT 2671
INT 4808
FLT 4745
132
OCZ 4000 Copper
— 2 x 512MB DS
2.5-3-4-5
2.55v
319.67 INT 2620
FLT 2657
INT 4701
FLT 4742
132
Adata DDR450
2 x 256MB SS
2-2-3-7
2.65v
311.0 INT 2590
FLT 2667
INT 4776
FLT 4744
132
Adata PC4000
— 2 x 512MB DS
2.5-3-3-6
2.55v
326.47 INT 2655
FLT 2624
INT 4704
FLT 4745
133
Geil Platinum 4000
— 2 x 512MB DS
2.5-3-4-5
2.55v
321.53 INT 2612
FLT 2646
INT 4701
FLT 4710
133

Results reported in our charts are the result of averaging the INT and FLOAT scores, which are normally close in value. INT and FLOAT scores were added and divided by 2 for our reported score. Since this is basically a continuation of our “Searching for the Memory Holy Grail” series, which examines the best performing configuration and memory for the 875/865 chipset, we have only included Intel 875 Performance Data in our charts.

Since current AMD systems really can’t use memory with DDR500 performance, and don’t require DDR500 for best performance, they were not included in earlier parts of this survey. However, DDR400 performance is very important on AMD boards, so we have included AMD memory performance for Mushkin and Adata.

AMD nForce2 Ultra 400 DDR400 — 400FSB Memory Performance
Memory Memory Timings
& Voltage
Quake3
fps
Sandra UNBuffered Sandra Standard
Buffered
Super PI 2M places
(time in sec)
Mushkin PC3500 Level II 2-2-2
2 x 512MB DS
2-2-2-4
2.55v
353.03 INT 1475
FLT 1589
INT 3073
FLT 2905
101
Adata DDR450
2 x 256MB SS
2-2-3-4
2.65v
344.6 INT 1402
FLT 1503
INT 3060
FLT 2880
104

Intel and AMD Performance numbers from these tests should not be compared. These tests were designed to benchmark memory performance. Due to differences in architecture, the AMD Barton testbed is running at 3200+ in these tests while the Pentium 4 2.4C is running at 2.4Ghz.




DDR400 Test Results






DDR400 Test Results (continued)






Final Words

In its performance range of DDR333 to DDR458, Mushkin PC3500 Level II is the best performing memory we have tested on the Intel 875. It is the only current memory we have tested to pass our benchmark tests on the Intel 875 in Dual-Channel mode at 2-2-2 timings, which is even more remarkable considering our tested modules were 512MB Double-Bank. Corsair 3200LL Rev. 1.1 also completed 2-2-2 tests at DDR400, but that Rev. has been replaced with Rev. 1.2, whose best performance at DDR400 is 2-2-3.

On the nForce2 Ultra 400 platform, the Mushkin Level II is as fast as you will find. And, it performs well in the complete range of FSB speed you are likely to reach on an Athlon platform with a 200FSB CPU. The Mushkin PC3500 Level II 2-2-2 is a close to perfect High-Performance memory for Athlon platforms.

The Adata DDR450 is an excellent performer at DDR400 and covers a broad range to DDR488. It is comparable to any current memory and is a very good value. For best performance, we recommend that you use four 256MB modules on the Intel 875/865 platform. On nForce2 Ultra 400, the Adata 450 is a very good performer, but with the requirement that Ras-to-Cas must be set to 3 or above. It is out-performed by other modules capable of 2-2-2 operation on nForce2 boards.

Looking at the larger picture of DDR400 performance, we find Mushkin PC3500 Level II and OCZ 3700 GOLD at the top of all the performance categories. Either of these will provide the best possible performance at DDR400. Kingston HyperX 4000, Adata DDR450, and Adata PC4000 are also other particularly good performers in the DDR400 range.

If you do not plan to overclock, there is no doubt that Mushkin PC3500 Level II or OCZ 3700 GOLD offer the best performance at DDR400 on the Intel 875/865 platform. Corsair 3200LL Rev. 1.1 — if you can still find it — is certainly in this same league and won Evan Lieb’s DDR400 shootout a few months ago.

If you have a 2.8 to 3.2GHz CPU and plan to overclock, then it is likely to be limited to an overclock somewhere in the 240 to 245 FSB range. For mild overclocks to about 228FSB, the Mushkin PC3500 LII is a great choice. But to cover the full overclocking range at 1:1, you would require OCZ 3700 GOLD, Adata DDR450, or any of the DDR500 modules tested in “Searching for the Memory Holy Grail — Part 2”.

For a 2.4C or 2.6C Intel Pentium 4 processor, you may achieve overclocks up to the 290 FSB range. The winners of our DDR500 shootout — OCZ PC4000 Copper and Geil Platinum 4000 — will let you reach the highest 1:1 or synchronous ratios possible. Be aware, however, that there is a downside to the very fastest DDR500 modules. We have found that many of them do not perform at all in asynchronous mode — ratios like 5:4 or 3:2. We have yet to find a pattern to this issue, and it almost seems a roll-of-the-dice with current DDR500 memory and Intel 875/865 motherboards as to which combo will work at 5:4 ratio.

For an Athlon on an nForce2 Ultra 400, Mushkin PC3500 Level II seems almost an ideal choice, since it is able to provide 2-2-2 timings at any FSB that we could reach in our tests. Another good choice for best performance would be Corsair 3200LL Rev. 1.1.

In the end, there is not just one answer to “What is the best memory for my computer?” It depends on your motherboard, your CPU, the applications you will run, and how you will run your computer.

Log in

Don't have an account? Sign up now