Conventional wisdom in the memory industry in recent years has been that CAS latency was the most important specification in memory performance. This has spawned a whole group of high-end DDR products with acronyms like "LL" for Low-Latency, "EL" for Enhanced Latency, and other labels with the message that Latency was the lowest possible and performance was the highest that you could buy. This whole thinking reached a frenzy in the DDR market around DDR333 to DDR400 where the message became so pervasive that if a memory was not CAS 2, then it was not the best performance that you could get.

As DDR memory frequencies have soared beyond DDR400 to DDR500, DDR550, and even higher, we have seen memory modules appear with timings of 3-4-4 at these higher memory speeds. Since customers were now firmly indoctrinated into the CAS 2 camp, many Enthusiasts complained that the new high-speed memory was only half there because what they really wanted was DDR500 at 2-2-2 timings. Variations on this theme were the "proof" that DDR400 run at 5:4 ratio at CAS 2 could normally meet or beat performance of higher speed DDR500 memory, which used slower CAS 3 timings. The point that was forgotten in this argument was that with all things being equal, DDR500 would always perform faster than DDR400, but technology was not to a point where DDR500 CAS 2 memory could be manufactured. This further reinforced for some the almighty importance of CAS latency.

Several recent developments have caused memory manufacturers to re-examine the importance of CAS latency to performance. First, manufacturers have been unable to produce a DDR500 memory that could run at CAS 2. In fact, the DDR500 modules have not even been able to run also at DDR400 at 2-2-2 timings. Second, Winbond announced their exit from the market, removing DDR400 2-2-2 and even 2-2-3 timings from currently available memory. Third, the development of DDR2 memory, which at DDR533 is being launched with 4-4-4 to 5-5-5 timings. An interesting development with DDR2 is that while memory timings are even slower than the worst DDR timings, the performance can sometimes keep up with DDR because of other internal memory timings. This is not praise for DDR2, but all these developments bring to the forefront that CAS Latency is not the only thing that matters in memory performance, and in fact, it may not even be the most important specification as we move to faster memory.

Into this climate, OCZ has introduced their latest EB, or Enhanced Bandwidth, memory with the claim that CAS 2.5 or 3 can often perform just as well as CAS 2. According to Dr. Michael Schuette of OCZ, "Enhanced Bandwidth technology is an exciting improvement that, at 2.5-2-3 (CL-tRP-tRCD) latencies, allows most applications to use data bandwidth that is the same as that delivered by CL-2 modules." Many of you already know Dr. Michael Schuette from his well-respected Lost Circuits technical website. What you may not know is that Michael is now Director of Technology Development for OCZ.

Enhanced Bandwidth Technology

So, what does OCZ mean by Enhanced Bandwidth? How does it work?

Basically, Enhanced Bandwidth (EB) is a means of increasing memory bandwidth by optimizing memory latencies for the best interaction between the memory and the chipset and memory controller. By analyzing memory traffic and benchmark results under varying memory latency settings in the CMOS setup of the BIOS, OCZ claims to have pinpointed the bottlenecks that most affect performance. They further claim that conventional wisdom regarding some memory latencies, namely that the lowest CAS latency yields the best performance, were found to no longer hold true in the higher speed memory of today.

Click to enlarge.

Timing diagram for two modules, one running at TRCD-4, CL-2.5, tRP-4 (bottom) and the second with tRCD-3, CL-2.5, tRP-2 (top) showing two consecutive bursts of 8 from two different pages - one of the most common scenarios in real world applications. The effective bandwidth is the ratio between data transfers (black diamonds):NoOps (red arrows). In the case of EB this ratio is 8:7, while without EB it is 8:10, meaning that every transfer of 16 bits is penalized with either 7 or 10 subsequent bus idle cycles. (Clk: clock; Act: row activate command; Rd: read command; Pr: Precharge command, NoOp: No Operation)

OCZ further explains that their EB series uses low tRP (RAS Precharge, which has a value of 2 in EB) and tRCD (RAS-to-Cas Delay, with a value of 3 in EB) latencies to enhance performance. This is then combined with a Variable Early Read Command to allow for the highest possible effective data bandwidth. Using these optimizations, OCZ says that in most applications, the 2.5-2-3 (CL-tRP-tRCD) will deliver bandwidth that is indistinguishable from CL-2 modules.

Click to enlarge.

The effect of issuing an early Read Command on back-to-back transactions of consecutively requested data blocks within the same page. Following one Row Activate Command, three Read commands are given at a CAS Latency of either 2, 2.5 or 3. The colored diamonds are the data transfers that belong to the Read Command in the same color. The graph shows that the net effect of increasing the CAS latency is a single cycle delay within a string of (in this case) 12 consecutive transfers, but no degradation of bandwidth. The double-arrows indicate the CAS latency which is amended by moving the read command further to the left (relative to the end of the previous burst). (Clk: clock; Act: row activate command; Rd: read command; Pr: Precharge command, CL: CAS Latency)

OCZ says that they have redesigned their memory products using these findings to deliver the highest possible bandwidth to any computer system, and they call these new products EB or Enhanced Bandwidth.

A more technical explanation is beyond the scope of this memory review, but those of you who would like more information on the theories behind Enhanced Bandwidth should take a look at the OCZ Enhanced Bandwidth white paper.

Is OCZ correct: can memory timings of 2.5-2-3 perform just as well as CAS 2 in many applications? Our goal is to find out.

OCZ 3500EB

Test samples of OCZ EB memory were a matched pair of 512MB 3500EB modules. This is just one of the offerings, however, since EB is offered in both DDR400 (PC3200) and DDR433 (PC3500) speeds. EB is also available in a wide range of capacities, with both 3200EB and 3500EB available in Dual-Channel kits of 512MB (256MBx2), 1GB (512MBx2) and 2GB (1024MBx2).

The OCZ 3500EB DIMMs have the Platinum heatspreader that OCZ uses to identify their top-performing Platinum series memory.

As is often the case with memory manufacturers, OCZ buys the blanks from a memory chip maker, then sorts according to their internal specs, and marks them with their own logo. OCZ tells us that the EB chips are made for them by several chip makers, and they preferred not to reveal the manufacturers.

OCZ 3500EB Specifications

OCZ 3500EB Memory Specifications
Number of DIMMs & Banks	2 DS
DIMM Size Total Memory	512MB 1 GB
Rated Timings	2.5-2-3 at DDR433
Rated Voltage	2.8V

We confirmed the SPD timings to be set at 2.5-2-3-8. OCZ tells us that this is to provide the best compatibility with the widest range of motherboards. We found the best performance on Intel 865/875 chipsets to be at 2.5-2-3-5 and on AMD nForce boards to be 2.5-2-3-9.

The 3500EB modules are rated at 2.8V, and are said to handle higher voltages without a problem. Most boards can supply 2.8V vDIMM, but some boards top out at 2.7V or lower, so please check your motherboard specs before purchasing. We found that the OCZ 3500EB performed fine at 2.5V at DDR400 and 2.65V at DDR433, but higher overclocks did require voltages of 2.8V or higher.

Performance Test Configuration

The Memory testbed for evaluating OCZ 3500EB is the same used in our earlier reviews of DDR400 and faster Memory.

Mushkin PC3200 2-2-2 Special: Last of a Legend
PMI DDR533: A New Name in High-Performance Memory
Samsung PC3700: DDR466 Memory for the Masses
Kingmax Hardcore Memory: Tiny BGA Reaches For Top Speed
New Memory Highs: Corsair and OCZ Introduce DDR550
OCZ PC3700 Gold Rev. 2: The Universal Soldier
OCZ 4200EL: Tops in Memory Performance
Mushkin PC4000 High Performance: DDR500 PLUS
Corsair TwinX1024-4000 PRO: Improving DDR500 Performance
Mushkin & Adata: 2 for the Fast-Timings Lane
Searching for the Memory Holy Grail - Part 2

All test conditions were as close as possible to those used in our earlier memory reviews.

INTEL 875P Performance Test Configuration
Processor(s):	Intel Pentium 4 2.4GHz (800MHz FSB)
RAM:	2 x 512MB OCZ 3500EB (DS) 2 x 512MB Mushkin 2-2-2 Special (DS) 2 x 512MB PMI4200 Gold (DDR533 DS) 4 x 256MB Samsung PC3700 (DDR466 SS) 2 x 512MB Kingmax DDR500 Hardcore Series (DS) 2 x 512MB Kingmax DDR466 Hardcore Series (DS) 2 x 512MB Corsair XMS4400v1.1 TwinX (DS) 2 x 512MB OCZ PC4400 DC Kit (DS) 2 x 512MB OCZ PC3700 Gold Rev. 2 (DS) 2 x 512MB OCZ 4200EL(DS) 2 x 512MB Mushkin PC4000 High Performance (DS) 2 x 512MB Corsair TwinX4000 PRO (DS) 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 4.1
Power Supply:	Vantec Stealth 470 Watt Aluminum
Operating System(s):	Windows XP Professional SP1
Motherboards:	Asus P4C800-E (875) with 1015 Release BIOS

We only tested the OCZ EB memory in our Intel testbed, but OCZ tests for 2.5-2-3 compatibility on nVidia nForce2/nF3, VIA K8T800, Intel 865/875, and SiS 655TX chipsets. While we have found the fastest performance on Intel 865/875 to be achieved at Cycle Time or tRAS of 5, we have found that the best performance on nForce family chipsets is often achieved with a tRAS of 8 or even as high as 11. In brief testing with nForce3 and nForce2 boards, we found that the best performance with OCZ3500EB was achieved at 2.5-2-3-9 settings, where the discontinued BH5 is usually best on nF2/nF3 at tRAS of 10 or 11.

Test Settings

We ran our standard suite of memory performance benchmarks. The following settings were tested with OCZ 3500EB:

1. 800FSB/DDR400 - the highest stock speed supported on 875/865 and K8T800/nF3/SiS755 motherboards.
2. 866FSB/DDR433 - the rated speed of OCZ 3500EB.
3. 933FSB/DDR466 - a common rating and overclock for high-speed memory.
4. 1000FSB/DDR500 - a standard speed rating that we have used in testing the highest speed memory available.
5. Highest Stable Overclock - the highest settings we could achieve with this memory and other memory that we have tested.

These are the same general settings used in benchmarking other memory in the above list of memory tests. DDR400, DDR500 and Highest Memory Speed have been used for all benchmarking. DDR433 and DDR466 have only been tested in the past for modules with this speed rating, and are included where available.

Test Results: OCZ 3500EB

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

OCZ 3500EB (DDR433) - 2 x 512MB Double-Bank
Speed	Memory Timings & Voltage	Quake3 fps	Sandra UNBuffered	Sandra Standard Buffered	Super PI 2M places (time in sec)
400DDR 800FSB	2.5-2-3-5 2.5V	322.2	INT 2735 FLT 2782	INT 4456 FLT 4459	131
433DDR 866FSB	2.5-2-3-5 2.65V	350.4	INT 3009 FLT 3018	INT 4856 FLT 4867	121
466DDR 933FSB	3-2-3-5 2.65V	376.5	INT 3216 FLT 3241	INT 5204 FLT 5215	113
500DDR 1000FSB	3-2-3-5 2.85V	403.3	INT 3457 FLT 3501	INT 5580 FLT 5586	104
510DDR 1020FSB	3-2-3-6 2.85V	408.1	INT 3516 FLT 3557	INT 5715 FLT 5718	102

At DDR433, OCZ 3500EB clearly meets its specifications. We were able to run our complete memory test suite at the SPD timings with no problem. We did achieve the best performance on the Intel test bed, however, with the lowest Cycle Time (tRAS) that we could run. Therfore, the best performance on an Intel chipset was with the lowest possible tRAS, which is reported in the chart.

It is also worth mentioning that OCZ 3500EB was completely stable at 2.5-2-3 timings to DDR460. Above 460, we needed to raise our CAS setting to CAS 3 for most stable performance. However, across the DDR400 to DDR510 range we were able to maintain a tRP (RAS Precharge) of 2, and a tRCD (RAS-to-Cas Delay) value of 3.

Performance Comparisons

Performance of the OCZ 3500EB was compared to all of the memory recently tested in:

Mushkin PC3200 2-2-2 Special: Last of a Legend
PMI DDR533: A New Name in High-Performance Memory
Samsung PC3700: DDR466 Memory for the Masses
Kingmax Hardcore Memory: Tiny BGA Reaches For Top Speed
New Memory Highs: Corsair and OCZ Introduce DDR550
OCZ PC3700 Gold Rev. 2: The Universal Soldier
OCZ 4200EL: Tops in Memory Performance
Mushkin PC4000 High Performance: DDR500 PLUS
Corsair TwinX1024-4000 PRO: Improving DDR500 Performance
Mushkin & Adata: 2 for the Fast-Timings Lane
Searching for the Memory Holy Grail - Part 2

Memory performance was compared at DDR400, DDR500 and the highest stable overclock we could achieve that would run Gun Metal 2 Benchmark 2, Quake 3, UT2003, and Super PI to 2MM places. Where the data was available, comparisons were also made at DDR433 and DDR466.

We have included memory based on Winbond BH5 chips, Mushkin 3500 Level II and OCZ 3500 Platinum Ltd, for reference in these comparisons, but neither are available for purchase in the marketplace. All discontinued products will be removed from future benchmark comparisons. Since Mushkin claims availability of Mushkin 3200 2-2-2 Special for the next 2 to 3 months, it will be included in memory benchmark comparisons until it is no longer available for purchase.

Results are compared for Quake 3, Sandra UNBufferred Memory Test, and Super PI. SiSoft Sandra 2004 reports 2 results for each memory test - an Integer value and a Float value. Results reported in our charts are the result of averaging the INT and FLOAT scores, which are normally close in value. In other words, INT and FLOAT scores were added and divided by 2 for our reported score.

We have also added additional Memory Timing and Memory Bandwidth comparisons in this review of OCZ 3500EB. Please refer to the additional test data for the 6 fastest currently AVAILABLE memories that we have tested in the performance range of 3500EB. This can be found on Page 12.

DDR400 Charts

DDR433 Charts

DDR466 Charts

DDR500 Charts

Highest Memory Speed Charts

Test Results: Memory Timings & Bandwidth Comparison

Our standard memory tests allowed comparison of OCZ 3500EB to other memory tested at AnandTech, but the charts do not clearly show what is happening with timings and bandwidth with EB memory. Using the same data in a different way illustrates better the impact of Enhanced Bandwidth on performance.

We took a closer look at the test results from six currently available memories that have been top performers in our benchmarks. If you look closely at the memory timings, you will see that while OCZ 3500EB was at CAS 2.5 at DDR400, all the other 5 memories managed CAS 2. By DDR433, the 3500EB was still at CAS 2.5 and the Mushkin BH6, OCZ 3700 Gold Rev.2, and Kingmax DDR500 were still at CAS 2.

Memory Timings
Speed	OCZ 3500EB	Mushkin 3200 2-2-2 Special	OCZ 3700 Rev. 2	Kingmax DDR500	Corsair XMS4000 PRO	OCZ 4200EL
400DDR 800FSB	2.5-2-3-5 2.5V	2-2-2-5 2.65V	2-2-3-5 2.65V	2-2-2-5 2.5V	2-3-3-5 2.65V	2-3-3-5 2.55V
433DDR 866FSB	2.5-2-3-5 2.65V	2-2-3-5 2.75V	2-2-3-6 2.65V	2-2-3-6 2.5V	Not Tested	Not Tested
466DDR 933FSB	3-2-3-5 2.65V	NA	2.5-3-3-6 2.65V	2.5-2-3-7 2.5V	Not Tested	Not Tested
500DDR 1000FSB	3-2-3-5 2.85V	NA	2.5-3-3-7 2.75V	2.5-3-3-7 2.5V	2.5-3-4-6 2.75V	2.5-3-4-6 2.65V
Highest Speed	3-2-3-6 510 - 2.85V	2-2-3-7 454 - 2.85V	3-3-4-8 520 - 2.85V	3-3-4-8 528 - 2.85V	3-3-4-7 539 - 2.85V	3-3-4-7 560 - 2.85V

Yet, if we look at the bandwidth results from SiSoft Sandra Unbuffered Memory Test, the OCZ 3500EB is neck-and-neck with the CAS 2 Mushkin 3200 and OCZ3700 Gold Rev.2 by the time we reach DDR433. At DDR466 and DDR500, the EB is at CAS 3, while all 5 of the competing top memories are at CAS 2.5. Yet the EB outperforms ALL of the other 5 memories at DDR500 despite requiring the slower CAS 3 timing. It is worth mentioning that RAS Precharge remains at 2, and Ras-to-Cas delay remains at 3, throughout the stable bandwidth of OCZ 3500EB.

Sandra 2004 Unbuffered Bandwidth
Speed	OCZ 3500EB	Mushkin 3200 2-2-2 Special	OCZ 3700 Rev. 2	Kingmax DDR500	Corsair XMS4000 PRO	OCZ 4200EL
400DDR	2759	2907	2898	2911	2840	2834
433DDR	3014	3013	3091	3042	-	-
454	-	3280	-	-	-	-
466DDR	3229	-	3301	3267	-	-
500DDR	3484	-	3443	3365	3297	3371
510	3537	-	-	-	-	-
520	-	-	3493	-	-	-
528	-	-	-	3501	-	-
539	-	-	-	-	3551	-
560	-	-	-	-	-	3729

Looking at the memory bandwidth patterns, OCZ 3500EB is competitive at DDR400 with faster CAS memory, but it is still outperformed. However, by the time we reach DDR433 to DDR466 range, EB is neck-and-neck with the best memory that we tested. By DDR500, EB is the fastest memory that we have yet tested. It appears that EB gets better and better, the higher the bandwidth. It is also interesting that at its highest speed of DDR510, 3500EB has a greater bandwidth than OCZ 3700 R2 at DDR520, Kingmax DDR500 at DDR528, and is about the same as Corsair XMS4000 PRO at DDR539. It appears, at DDR510, that OCZ 3500EB is showing the same bandwidth we find with the best memory we have tested at about DDR540.

OCZ 3500EB was also designed for maximum compatibility with the wide variety of motherboards and chipsets available for Intel and AMD processors. The board used in our memory test bed is known to be one of the most compatible boards with a wide variety of memory, which is one of the reasons why we use the Asus P4C800-E. However, not all boards are as accommodating of different memory, as we often see in our motherboard memory test section. OCZ tells us that one of the major goals in the design of 3500EB was to provide the widest compatibility with the widest range of motherboards. This includes current Athlon 64 chipsets, some of which have proven to have problems with memory timings on current high-end memory.

Conclusion

OCZ is breaking new ground with their latest Enhanced Bandwidth series. Many readers will likely find it disturbing that their long-held beliefs that the lowest CAS is always the best performer is no longer completely true. Based on the results we see with OCZ 3500EB, you will now need to look deeper than CAS timings to evaluate memory performance.

In general, 3500EB performed exactly as OCZ claimed it would in our benchmarks. It is not as fast as some current CAS 2 memory at DDR400, but its performance is very close to the best, even though EB is rated at CAS 2.5. However, by the time we reached the DDR433 to DDR466 range, 3500EB performed neck-and-neck with the best memory that we have tested, even though the CAS latency was slower. By the time we reached DDR500, OCZ 3500EB was reaching the highest bandwidth we have yet seen at DDR500, even though it ran at CAS 3 and the other fast memories were at CAS 2.5. It appears that the more you overclock 3500EB, the greater the impact of the EB optimizations.

The other surprising finding is that this DDR433 module has no problem at all running at DDR500 and providing the best performance we have measured at that frequency. In fact, we were able to reach a stable DDR510 with a memory rated at DDR433. Perhaps OCZ should use EB to also stand for Extended Bandwidth because we have never seen a DDR433 memory capable of performance at these DDR500+ speeds.

There are other memories that perform well in the same DDR400-510 range as OCZ 3500EB. At the DDR400-454 range, Mushkin 3200 2-2-2 Special performs well. Across the entire DDR400-510 range, OCZ 3700 Gold Rev.2 and Kingmax DDR500 are neck-and-neck. Also, Corsair XMS4000 PRO, Mushkin 4000 High Performance, and OCZ 4200EL perform well across the same range and reach even higher frequencies than OCZ 3500EB. However, all-in-all, there is no memory we have tested that can deliver the kind of extended and enhanced bandwidth we have found in our tests of 3500EB.

Certainly the six memories we listed are the best we have tested so far that you can still buy. OCZ 3500EB belongs in that list, if your motherboard can support the required memory voltage of about 2.8V. You should look at these 6 carefully to determine what best meets your needs. However, all 6 are also among the most expensive memories available from each memory manufacturer - except for OCZ 3500EB. OCZ has priced all EB memory at about 10% less than the other top OCZ memory products. 3500EB is also designed to be more compatible with fast performance timings on nForce2/3 and VIA AMD chipsets, so you should definitely look closely at OCZ 3500EB for an Athlon or Athlon 64 motherboard.

EB is an interesting and successful technology for OCZ. You should definitely add it to your short list if you are looking for DDR memory. For the broad range of DDR400 to DDR510, 3500EB is a remarkable performer.