With the Introduction of Intel’s Canterwood 875 and Springdale 865 chipsets, the memory landscape is changing rapidly. These mainstream chipsets bring many new features to the desktop and workstation PC, but memory support may be the biggest innovation as it takes a giant leap with standard support for Dual-Channel DDR400 memory.

It wasn’t long ago that we were talking about whether DDR400 would even become an official standard. Now, with the Intel 875/865 chipsets, we see DDR400 completely legitimized. More importantly, with the excellent performance of the Intel Pentium 4 800FSB chips, we are seeing enthusiasts actively searching for memory that can perform at DDR466/DDR500 and even higher speeds. These were memory speeds that we never thought we would be talking about on the way to DDRII, and apparently, memory manufacturers have also been caught by surprise. Introductions have been slow because chips are apparently unavailable, but we now see high-end memory manufacturers scrambling to bring out these extremely high-speed memory modules for which enthusiasts are asking.

Given this climate, we decided to take a closer look at the question of what is the best memory for the Canterwood and Springdale chipsets. In Part 1, we will try to determine the best Memory configuration for the 875/865. We were able to examine this and find some answers with Memory that we had on-hand. The answers will be important for many of you who are looking to buy memory for the new boards, so we decided to release the results of our memory configuration testing. Part 2 will investigate the performance of the new DDR466 and DDR500 modules that will be coming to market in the near future. Since we are still waiting to receive many of these new modules, or currently have only beta samples of some of the memory, it will be several weeks until Part 2 is posted.

The Best Memory Configuration for 865/875

One of the questions we are often asked is whether a particular motherboard can run with four DIMMs - or at the maximum number of memory slots for the board. Surprisingly, the answer is often ”no”, which is why AnandTech added the process of populating and testing all memory slots to the review procedures. However, with the Intel 875/865 Dual-Channel boards, we are realizing that additional questions need to be raised. Is there a performance difference in two DIMMs vs. four DIMMs? Do single-sided or double-sided DIMMs perform better on Canterwood/Springdale boards? What is the real performance difference in one DIMM, two DIMMs, and four DIMMs?

Answers to all of these questions will lead to determining the best-performing Memory configuration for Intel 875/865 boards. We set out to find the answer to this question, and what we discover may surprise you.

Test Design

AnandTech has had reservations about the usefulness of synthetic benchmarks to measure the performance of computers. However, testing memory is one area where it is hard to avoid synthetic benchmarks. We were particularly concerned about some of the memory benchmarks we have seen posted with the 875/865 chipset boards. Some of the more widely-used synthetic benchmarks are being used to “demonstrate” performance that we logically know – based on Intel 865/865 chipsets white papers – to be not the case.

MemTest86

MemTest86 has been around for many years, and it is the standard in Memory Testing for many computer enthusiasts. The program boots from CD-ROM or floppy using its own DOS-like operating system, and tests memory at its most basic level. To many enthusiasts, this program is the ultimate test to determine the stability of memory. If you wonder why we would use such an ancient benchmark, it is because we have not found anything that is really better to test memory performance and stability. Apparently, we are not alone, since MemTest86 is still widely used in the Memory industry to test the stability of memory modules.

As a baseline, MemTest86 was run on an 875 Canterwood system with a Pentium 4 3.0GHz CPU and 2x512MB double-sided high-speed DIMMs at timings of 2,5,2,2. This gives us a MemTest86 score at Default Settings of 2,722MB/sec.

More information on MemTest86 is available at their site. Floppy and ISO versions of the program are also available at this site for download.

UNBuffered Memory Performance in Windows XP

In search of a reliable memory benchmark, we find that SiSoft Sandra 2003 is one of the more widely used memory benchmarks. Looking closer, we did not feel the standard Sandra memory test, which relies heavily on synthetic memory buffering techniques, provided the kind of sensitivity we needed for these tests. It is true that Sandra uses the same buffering techniques as in the CPU and chipset tests, but these buffering techniques themselves can often mask poor memory performance and sometimes provide memory performance scores that do not correlate well with real-world performance.

While we don’t know exactly where it started, we have seen a number of enthusiast sites using a variation of the Sandra Memory, which turns off buffering techniques and comes closer to measuring raw memory performance. Macci was one of the first to use the UNBuffered benchmark – the famous Finn who is always at or near the top of the 3DMark scores. In addition, Tony Leach in England and I have used it in many reviews on other web sites.

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.

The UNBuffered Memory Benchmarks are quite different from what you may be accustomed to seeing. As an example, these are Buffered or Standard Memory benchmarks with two double-sided DIMMs on a 875 Canterwood board at stock 3.0GHz at DDR400 at memory timings of 2-5-2-2.

Using the same settings for the UNBuffered Memory Benchmark (3.0GHz, DDR400, 2-5-2-2, 2x512MB DIMMs), we achieved much lower numbers.

Please look carefully at the screen captures above, and notice that THE MEMORY BAR STATES “BUFF” IN THE DESRIPTION ON A STANDARD TEST. “BUFF” IS NOT USED IN THE DESCRIPTION IN THE UNBUFFERED MEMORY TEST

While the score of the UNBuffered benchmark is much lower than the standard Memory Test, the UNBuffered benchmarks are very comparable to MemTest86 results. This excellent correlation of MemTest86 and SiSoft Sandra UNBuffered Memory Test results seems to be true in most situations. Benchmarks are only useful in the end in how they can compare different configurations and hardware. We have found standard Dual-Channel DDR400 Performance on Intel 875/865 chipset motherboards, as you see above, in the 2300 to 3000 MB/sec range as measured with the Sandra UNBuffered Memory Benchmark. MemTest86 gives similar results.

We will post a few more sets of Buffered (Standard) and UNBuffered Sandra 2003 benchmarks in this review to demonstrate the validity of the UNBuffered benchmark as a Memory Performance testing tool, and then use only UNBuffered benchmarks for the rest of the review.

Performance Test Configuration

Performance Test Configuration
Processor(s):	Intel Pentium 4 3.00GHz (800MHz FSB) Intel Pentium 4 2.6GHz (800MHz FSB)
RAM:	4 x 256MB OCZ3700 GOLD (DS) 2 x 512MB Winbond BH5 PC3500 (DS) 2 x 256MB Crucial DDR400 (Samsung chips) (SS) 4 x 256MB OCZ4000 Beta (SS)
Bus Master Drivers:	875P: Intel INF Update v5.00.1012, RAID IAA drivers not installed on 875P board for consistency in test results
Video Card(s):	ATI 9800 PRO 128MB ATI 9700 PRO 128MB
Video Drivers:	ATI Catalyst 3.5
Operation System(s):	Windows XP Professional SP1
Motherboards:	DFI 875PRO LAN Party (875P) with 6/25/03 release BIOS Asus P4P800 Deluxe (865PE) with 1.009.004 BIOS

Two motherboards tested at Anandtech were used in tests to determine the best memory configuration for 875/865 chipsets. All testing at DDR400 and DDR466 were with a 3.0Ghz 800FSB CPU on the DFI 875PRO LanParty. The DFI benchmarks used either one, two, or four OCZ3700 GOLD 256MB DS modules, or 2 OCZ3700 GOLD DS and 2 Crucial 400 SS 256Mb/Winbond BH5 DS 512MB for mixed mode testing. All Benchmarks at DDR500 were run with a 2.6GHz 800FSB CPU on the Asus P4P800 Deluxe, which we reviewed in our 865PE/875P Motherboard Roundup June 2003. We used one, two, or four OCZ4000 GOLD beta modules, which are single-sided.

Performance Tests

Double-Sided Memory

DDR400, 800FSB (200x4) Intel 875 Chipset, 2-7-3-3 Memory Timings OCZ3700 GOLD 256MB DS
Number of DOUBLE-Sided DIMMS	MemTest86 MB/Second	UNBuffered Sandra 2003 Memory Test (MB/Second)	Buffered (Standard) Sandra 2003 Memory Test
1	1671	1790 INT 1837 FLT	2974 INT 2972 FLT
2	2544	2600 INT 2660 FLT	4769 INT 4710 FLT
4	2489	2861 INT 2848 FLT	4714 INT 4693 FLT

DDR466, 932FSB (233x4) Intel 875 Chipset, 2-7-3-3 Memory Timings OCZ3700 GOLD 256MB DS
Number of DOUBLE-Sided DIMMS	MemTest86 MB/Second	UNBuffered Sandra 2003 Memory Test (MB/Second)	Buffered (Standard) Sandra 2003 Memory Test
1	1947	2046 INT 2145 FLT	3447 INT 3456 FLT
2	2964	3192 INT 3247 FLT	5614 INT 5584 FLT
4	2900	3421 INT 3455 FLT	5604 INT 5609 FLT

As you can see from the benchmarks above, SiSoft Sandra UNBuffered is a more sensitive benchmark for determining differences in memory performance than the Standard Buffered test – which can sometimes conceal differences in memory performance with buffering techniques. It also appears to be more sensitive than MemTest86 in determining memory bandwidth. This is particularly clear in the benchmarks for two and four double-sided DIMMs in Dual-Channel mode. Standard buffered benchmarks and MemTest86 show the performances of two and four double-sided Dimms as fairly equal, while UNBuffered benchmarks show that four double-sided DIMMs are about 7 to 10% higher in performance – a significant increase.

Later in the review, you will read confirmation from a respected source that the UNBuffered Memory Test results are a better reflection of what is really happening in memory performance on the 875/865 chipsets when comparing two versus four DIMMs. Therefore, for the remainder of tests, we will only use Sandra UNBuffered Memory Test and MemTest86 for benchmarking.

DDR500, 1000FSB (250x4) Intel 875 Chipset, 2.5-7-4-4 Memory Timings OCZ3700 GOLD 256MB DS
Number of DOUBLE-Sided DIMMS	MemTest86 MB/Second	UNBuffered Sandra 2003 Memory Test
1	1953	2193 INT 2203 FLT
2	2821	3051 INT 3129 FLT
4	2821	3318 INT 3467FLT

Single-Sided Memory

DDR500, 1000FSB (250x4) Intel 875 Chipset, 2.5-7-4-4 Memory Timings OCZ4000 GOLD 256MB SS
Number of SINGLE-Sided DIMMS	MemTest86 MB/Second	UNBuffered Sandra 2003 Memory Test
1	1924	2076 INT 2090 FLT
2	2551	2725 INT 2762 FLT
4	2821	3254 INT 3194FLT

Single-sided modules appear to perform very differently on the Intel 875/865 chipsets than double-sided modules. Both MemTest86 and Sandra UNBuffered Memory Test show that four single-sided DIMMs perform much better than two. The UNBuffered Sandra benchmark shows an improvement of almost 20% in performance when using four Dimms instead of two. This is more than double the UNBuffered Performance improvement in going from two DS DIMMs to four DS DIMMs on the 875/865 chipsets

UNBuffered Memory Performance Summary

Having established that the SiSoft Sandra UNBuffered Memory Test is a sensitive benchmark for measuring performance differences in memory configuration, it is now useful to compare UNBuffered Memory Benchmarks for one, two, and four double-sided and single-sided DIMM modules.

UNBuffered Memory Benchmark 875/865
Number of DIMMS	DDR400/800FSB DOUBLE-SIDED	DDR466/932FSB DOUBLE-SIDED	DDR500/1000FSB DOUBLE-SIDED	DDR500/1000FSB SINGLE-SIDED
1	1790 INT 1837 FLT	2046 INT 2145 FLT	2193 INT 2203 FLT	2076 INT 2090 FLT
2	2600 INT 2660 FLT	3192 INT 3247 FLT	3051 INT 3129 FLT	2725 INT 2762 FLT
4	2861 INT 2848 FLT	3421 INT 3455 FLT	3318 INT 3467FLT	3254 INT 3194FLT

In all cases, on both 865 and 875 chipsets, with both single-sided and double-sided DIMMs, four DIMMs is the best performing memory configuration at DDR400(1:1) or higher speed. This may come as a surprise to many of you looking for memory for your 865/875 motherboard. As expected, two DIMMs in dual-channel memory configuration performed better than a single DIMM in all cases. Beyond this, we see that two double-sided DIMMs perform much better than two single-sided DIMMs, but that four single-sided DIMMs perform almost as well as four double-sided DIMMs.

Intel White Papers Confirm Results

We have seen that SiSoft Sandra UNBuffered Memory benchmarks show four DIMMs to perform better than two in all configurations at DDR400 or higher base speeds on the 875/865 chipsets. However, MemTest86 Bandwidth and Sandra Standard Memory benchmark do not show improvement in going from two to four double-sided DIMMs. So why do we believe Sandra UNBuffered Memory benchmarks?

The answer, surprisingly, is in Intel 875P Chipset Memory Configuration Guide White Paper and the Intel 865P Chipset Memory Configuration Guide White Paper. The tables below are taken from Page 13 of the 875P White Paper, but the tables are identical in both the 865 and 875 documents, except that the first and second positions are reversed in the DDR400 chart for the 865. The Intel performance rankings exactly match the rankings we have measured using SiSoft Sandra UNBuffered Memory Test. It is interesting that UNBuffered Sandra was able to distinguish the performance differences between #1 and #2 –- four double-sided vs. two double-sided. This performance distinction was not apparent in either MemTest86 Bandwidth or SiSoft Sandra Standard Memory Test.

DDR400 Performance Configurations on Intel 875/865 from Intel White Papers
DDR Speed	Number of DIMMS	Sides per DIMM	Mode	SC or DC	Performance
400 MHz	4	2	Dynamic	Dual Channel	1 (2 for 865)
400 MHz	2	2	Dynamic	Dual Channel	2 (1 for 865)
400 MHz	4	1	Dynamic	Dual Channel	2
400 MHz	2	1	Dynamic	Dual Channel	3
400 MHz	4	any	Normal	Dual Channel	4
400 MHz	any	2	Dynamic	Single Channel	5
400 MHz	any	1	Dynamic	Single Channel	6
400 MHz	any	any	Normal	Single Channel	7

DDR266/333 Performance Configurations on Intel 875/865 from Intel White Papers
DDR Speed	Number of DIMMS	Sides per DIMM	Mode	SC or DC	Performance
233/333 MHz	2	2	Dynamic	Dual Channel	1
233/333 MHz	4	1	Dynamic	Dual Channel	1
233/333 MHz	2	1	Dynamic	Dual Channel	2
233/333 MHz	4	2	Dynamic	Dual Channel	3
233/333 MHz	any	any	Normal	Dual Channel	4
233/333 MHz	2 (1 in each channel)	any	Dynamic	Single Channel	5
233/333 MHz	1	2	Dynamic	Single Channel	5
233/333 MHz	any	1	Dynamic	Single Channel	6
233/333 MHz	any	any	Normal	Single Channel	7

While we didn’t test Asynchronous 5:4 or 3:2 performance, or Memory Performance when using a 533FSB CPU, the DDR266/333 table should be useful for those situations. When running DDR333/266 as a base speed, two double-sided DIMMS or four single-sided DIMMS perform the fastest. Two single-sided DIMMs are second in speed performance, and four double-sided DIMMs – the fastest performer at DDR400 and higher – drops to third place.

Mixed Configurations

In general, Intel’s White Papers assume that you will be using two or four matched DIMMs. The reality for many users, however, is that while they may have a pair of matched DIMMs, they do not likely have four matched DIMMs at their disposal to use in their new 865/875 boards. Intel’s White Papers address mixed memory configurations only to say that they will work, but they will default to the slowest speed and SPD timings of the mixed DIMMs.

To get a better idea of what happens to performance when mixing DIMMs at default performance at DDR400/800FSB, refer to the statistics below.

DIMM Configuration	Best Memory Timing	UNBuffered Sandra 2003 Memory Test (MB/Second)	% Changed from Matched DIMM Performance
2x256MB DS + 2x512MB DS	2-7-3-3	2094 INT 2148 FLT	-25%
2x256MB DS + 2x256MB SS	2.5-7-3-3	2064 INT 2132 FLT	-22%
2x512MB DS + 2x256MB SS	2.5-7-3-3	2097 INT 2150 FLT	-23%
4x256MB Matched DS	2-7-3-3	2861 INT 2848 FLT	----

The performance declines significantly from mixing different pairs of memory, even with two pairs of matched dual-channel. A 22% to 25% drop in memory performance compared to four matched double-sided DIMMs is certainly nothing to sneeze at.

Undoubtedly, the memory modules themselves are having an impact on this drop in performance. However, the four matched DIMMs run at 2-7-2-2, which is certainly not the best score I have seen at DDR400. In fact, the 2x512MB DS pair used in the mixed tests run as a dual-channel pair at DDR400 with 2-5-2-2 timings. The SS pair of 256MB DIMMs do require CAS 2.5 for best performance, and run fine as a pair at 2.5-6-3-2 at DDR400. Yet, these slower SS DIMMs perform better mixed with the DS GOLD modules than the faster DIMMs.

There will be great variation in mixed DIMM performance, with some close DIMMs running much better than these results. However, you can clearly see from the results that the very best performance at 1:1 requires four matched DIMMs. The performance loss from running two different matched pairs of DIMMs can be enormous.

FSB Overclocking with 1, 2, and 4 DIMMs

Since we have established that the best performance at DDR400 or above (1:1) is with four DIMMs, it is natural to ask whether there is a downside to running four DIMMs instead of two or even one DIMM.

Front Side Bus Overclocking Test Setup
Processor:	Intel Pentium4 3.0C 800FSB Hyperthreading	Intel Pentium4 2.6C 800FSB Hyperthreading
Motherboard:	DFI 875PRO Lan Party(875)	Asus P4P800 Deluxe (865)
CPU Vcore:	1.60v	1.650V
Memory:	OCZ3700 GOLD DDR466	OCZ3700 GOLD DDR466
vDIMM:	1.70V (MB limited)	1.85V (MB Limited)
Cooling:	Thermalright SLK-900U	CoolerMaster HeatPipe
Power Supply:	Vantec 520W	Powmax 400W

	Pentium4 3.0C 800FSB CPU			Pentium4 2.6C 800FSB CPU
# of DS DIMMs:	1	2	4	1	2	4
MAXIMUM Stable FSB:	992 (4x248) CPU Limited	992 (4x248) CPU Limited	968 (4x242)	1048 (4x262)	1032 (4x258)	1024 (4x256)
SPD or Manual:	SPD	SPD	SPD	SPD	SPD	SPD
CAS Latency:	2.5	2.5	2.5	2.5	2.5	2.5
RAS to CAS Delay:	3	3	3	4	4	4
RAS Precharge:	7	7	7	7	7	7
Precharge Delay:	3	3	3	4	4	4

Despite the slightly lower overclock possible with four DIMMs, the best 1:1 performance is still with four DIMMs. For example, DDR516 using four DIMMs out-performs DDR530 with two DIMMs – since the four DIMMs are about 7% to 10% higher in UNBuffered memory performance.

Conclusion

If you plan to run DDR400 as your base memory speed with an 800FSB processor, your best memory performance will clearly be with four matched double-sided DIMMs. Our tests, confirmed by Intel White Papers, show these configurations ranked from fastest to slowest performance on an Intel 875 Canterwood or 865 Springledale chipset motherboard.

DDR400 (1:1) Performance	DIMM Configuration	Single-Channel or Dual-Channel
1	4 DS	Dual Channel
2	2 DS or 4 SS	Dual Channel
3	2 SS	Dual Channel
4	4 SS/DS Mixed Matched Pairs	Dual Channel
5	Any DS	Single Channel
6	Any SS	Single Channel

You can run mixed matched pairs of DIMMs. That is, two of one type and size and two of another type and size, but the performance can be significantly lowered. In our tests of mixed pairs, performance decreased 22% to 25% compared to four matched double-sided DIMMs.

If you do plan to use single-sided DIMMs, then make sure you run four matched DIMMs. A pair of single-sided DIMMs ranked third in our performance tests, and was almost as slow as mixed pairs.

The situation changes if you plan to run a 533FSB P4 Processor, or if you intend to run your memory at 333 or 266 Base (5:4 or 3:2 ratio). With a 533 CPU or asynchronous 5:4 or 3:2 memory ratio, the best performing memory configuration is two double-sided DIMMs. Based on Intel White Papers, memory performance with a 533CPU or 5:4/3:2 (333/266) is ranked from best to worst.

DDR333/266 Performance	DIMM Configuration	Single-Channel or Dual-Channel
1	2 DS or 4 SS	Dual Channel
2	2 SS	Dual Channel
3	4 DS	Dual Channel
4	4 SS/DS Mixed Matched Pairs	Dual Channel
5	Any 1 or 2 DS or SS (1 DIMM or 1 in each Channel)	Single Channel
6	Any 1 SS	Single Channel

While four DIMMs do not overclock quite as well as two DIMMs on an 875/865 board, we confirmed that the added memory bandwidth more than makes up for the slightly lower overclock with four double-sided DIMMs.

Now that we have determined the best Memory Configurations for 875/865 motherboards, we will look at the best performing high-speed memory in Part 2 of Searching for the Memory Holy Grail. Since we are still waiting for several new releases of DDR466 and DDR500 Memory, it will be several weeks until we can publish a reasonable comparison of the latest in high-speed memory. Stay tuned!