Memory Module Analysis. Part 12: GeIL DDR2-667 Modules

We proceed with the series of articles, dedicated to the low level analysis of the most important characteristics of DDR2 memory modules using our RightMark Memory Analyzer test package. In this article we are going to review a pair of 512MB DDR2-667 modules, manufactured by Golden Emperor International Ltd. (GeIL is an abbreviation).

Manufacturer Information

Module manufacturer: Golden Emperor International Ltd. (GeIL)
Manufacturer of module chips: Golden Emperor International Ltd. (GeIL)
Web site of the module manufacturer: http://www.geil.com.tw/portal/product.php

Module Appearance

Photo of the memory module

Module Part Number

There is no Part Number expansion on the product page (http://www.geil.com.tw/portal/product_DDR2-667DC.php). The manufacturer provides only brief module characteristics:

Performance characteristics: PC5300 (PC2-5300 would be more correct), 667 MHz, 4-4-4 timings
Chips: 64x8 DDR2 in FBGA-package
Module parameters: 240-pin, non-ECC, unbuffered
Aluminium heatsink
6-layered screened PCB with a very low interference level
Voltage: 1.8 — 1.9V

SPD module chip data

Description of the general SPD standard:

JEDEC Standard No. 21-C, 4.1.2 - SERIAL PRESENCE DETECT STANDARD, General Standard

Description of the specific SPD standard for DDR2:

Parameter	Byte	Value	Expansion
Fundamental Memory Type	2	08h	DDR2 SDRAM
Number of Row Addresses on this assembly	3	0Eh	14 (RA0-RA13)
Number of Column Addresses on this assembly	4	0Ah	10 (CA0-CA9)
Number of DIMM Banks	5	60h	1 physical bank
Data Width of this assembly	6	40h	64 bit
Voltage Interface Level of this assembly	8	05h	SSTL 1.8V
SDRAM Cycle time (t_CK) at maximum supported CAS# latency (CL X)	9	30h	3.00 ns (333.3 MHz)
DIMM configuration type	11	00h	Non-ECC
Refresh Rate/Type	12	82h	7.8125 ms — 0.5x reduced self-refresh
Primary SDRAM Width (organization type) of the memory module chips	13	08h	x8
Error Checking SDRAM Width (organization type) of the memory chips in the ECC module	14	00h	Not defined
Burst Lengths Supported (BL)	16	0Ch	BL = 4, 8
Number of Banks on SDRAM Device	17	04h	4
CAS Latency (CL)	18	38h	CL = 5, 4, 3
Minimum clock cycle (t_CK) at reduced CAS# latency (CL X-1)	23	3Dh	3.75 ns (266.7 MHz)
Minimum clock cycle (t_CK) at reduced CAS# latency (CL X-2)	25	50h	5.00 ns (200.0 MHz)
Minimum Row Precharge Time (t_RP)	27	30h	4, CL = 5 3 (3.2), CL = 4 2 (2.4), CL = 3
Minimum Row Active to Row Active delay (t_RRD)	28	1Eh	2.5, CL = 5 2.0, CL = 4 1.5, CL = 3
Minimum RAS to CAS delay (t_RCD)	29	30h	12.0 ns 4, CL = 5 3 (3.2), CL = 4 2 (2.4), CL = 3
Minimum Active to Precharge Time (t_RAS)	30	2Dh	45.0 ns 15, CL = 5 12, CL = 4 9, CL = 3
Module Bank Density	31	80h	512 MB
Write recovery time (t_WR)	36	3Ch	15.0 ns 5, CL = 5 4, CL = 4 3, CL = 3
Internal write to read command delay (t_WTR)	37	1Eh	7.5 ns 2.5, CL = 5 2.0, CL = 4 1.5, CL = 3
Internal read to precharge command delay (t_RTP)	38	1Eh	7.5 ns 2.5, CL = 5 2.0, CL = 4 1.5, CL = 3
SDRAM Device Minimum Active to Active/Auto Refresh Time (t_RC)	41, 40	3Ch, 00h	60.0 ns 20, CL = 5 16, CL = 4 12, CL = 3
SDRAM Device Minimum Auto-Refresh to Active/Auto-Refresh Command Period (t_RFC)	42, 40	69h, 00h	105.0 ns 35, CL = 5 28, CL = 4 21, CL = 3
Maximum device cycle time (t_CKmax)	43	80h	8.0 ns
SPD Revision	62	10h	Revision 1.0
Checksum for Bytes 0-62	63	BCh	188 (true)
Manufacturer’s JEDEC ID Code	64-71	FFh, 7Fh, 7Fh, 13h	Golden Empire
Module Part Number	73-90	—	GX21GB5300DC
Module Manufacturing Date	93-94	FFh, FFh	Not correct
Module Serial Number	95-98	20h, 02h, 07h, 25h	25070220h

According to the SPD data, the reviewed modules can operate with three different CAS# latencies, each one corresponds to its own cycle time (working frequency). Thus, the maximum CAS# latency (5) corresponds to DDR2-667 mode (3.0 ns, 333.3 MHz), the reduced CAS# latency (4) — to the DDR2-533 mode (3.75 ns, 266.7 MHz), and finally the least CAS# latency (3) — to the DDR2-400 mode (5.0 ns, 200.0 MHz). Note that the absolute timings values t_RCD and t_RP are somewhat reduced (12 ns) relative to the typical (15 ns) values for DDR2 memory. It consequently has an effect on the relative values in timing schemes, which can be written as follows:

Mode	Timings
DDR2-667	5-4-4-15
DDR2-533	4-3-3-12 (rounded off to the nearest integer)
DDR2-400	3-2-2-9 (rounded off to the nearest integer)

The timing scheme for DDR2-667 differs much from the 4-4-4(-12) scheme claimed on the module and on the description page, in a bad sense at that. The other schemes are obtained by rounding off t_RCD and t_RP to the nearest integer (in this case — they are rounded down). But the practice shows (we are going to see that again below) that almost all motherboards round the values up, when the memory system is configured automatically (to all appearances, on maximum stability grounds).

Among the other SPD peculiarities we can note a Part Number (which matches the one specified on the modules) and a serial number. You should also note that the module manufacturing date is not just absent but wrong.

Testbed Configurations and Software

Testbed #1

CPU: Intel Pentium 4 3.4 GHz (Prescott D0, 1 MB L2)
Chipset: Intel 925X
Motherboard: Gigabyte 8ANXP-D, BIOS F1 dated 06/07/2004
Memory: 2x512 MB GeIL DDR2-667
Video: Leadtek PX350 TDH, nVidia PCX5900
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: nVidia Forceware 62.01, Intel Chipset Utility 6.0.1.1002, DirectX 9.0c

Testbed #2

CPU: Intel Pentium 4 3.4 GHz (Prescott D0, 1 MB L2)
Chipset: Intel 925X
Motherboard: MSI 925X Neo, BIOS dated 06/18/2004
Memory: 2x512 MB GeIL DDR2-667
Video: Leadtek PX350 TDH, nVidia PCX5900
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: nVidia Forceware 62.01, Intel Chipset Utility 6.0.1.1002, DirectX 9.0c

Testbed #3

CPU: Intel Pentium 4 3.4 GHz (Prescott D0, 1 MB L2)
Chipset: Intel 915P
Motherboard: MSI 915P Neo2, BIOS V1.3B0 dated 09/08/2004
Memory: 2x512 MB GeIL DDR2-667
Video: Leadtek PX350 TDH, nVidia PCX5900
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: nVidia Forceware 62.01, Intel Chipset Utility 6.0.1.1002, DirectX 9.0c

Testbed #4

CPU: Intel Pentium 4 3.4 GHz (Prescott D0, 1 MB L2)
Chipset: Intel 915P
Motherboard: ECS PF4 Extreme, BIOS dated 06/01/2004
Memory: 2x512 MB GeIL DDR2-667
Video: Leadtek PX350 TDH, nVidia PCX5900
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: nVidia Forceware 62.01, Intel Chipset Utility 6.0.1.1002, DirectX 9.0c

Testbed #5

CPU: Intel Pentium 4 3.4 GHz (Prescott D0, 1 MB L2)
Chipset: Intel 915G
Motherboard: MSI 915G Combo, BIOS 080011 dated 07/14/2004
Memory: 2x512 MB GeIL DDR2-667
Video: Leadtek PX350 TDH, nVidia PCX5900
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: nVidia Forceware 62.01, Intel Chipset Utility 6.0.1.1002, DirectX 9.0c

Testbed #6

CPU: Intel Pentium 4 3.6 GHz (Prescott E0, 1 MB L2)
Chipset: Intel 925XE, 200 MHz FSB
Motherboard: ECS PF21 Extreme, BIOS dated 12/07/2004
Memory: 2x512 MB GeIL DDR2-667
Video: Leadtek PX350 TDH, nVidia PCX5900
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: nVidia Forceware 62.01, Intel Chipset Utility 6.0.1.1002, DirectX 9.0c

Testbed #7

CPU: Intel Pentium 4 3.6 GHz (Prescott E0, 1 MB L2) at 3.73 GHz (266 MHz x14)
Chipset: Intel 925XE, 266 MHz FSB
Motherboard: ECS PF21 Extreme, BIOS dated 12/07/2004
Memory: 2x512 MB GeIL DDR2-667
Video: Leadtek PX350 TDH, nVidia PCX5900
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: nVidia Forceware 62.01, Intel Chipset Utility 6.0.1.1002, DirectX 9.0c

Test Results

We tested GeIL DDR2-667 modules in the DDR2-533 mode, because (1) most motherboards in our tests (on i915P/G and i925X chipsets) do not support DDR2-667 mode at all; (2) it's useless in many respects, because the memory bandwidth is limited by the FSB throughput, which is 6.4 GB/sec at 200-MHz and 8.53 GB/sec at 266-MHz FSB. The peak DDR2-667 throughput in dual channel mode is 10.66 GB/sec.

Performance tests

The first series of tests used the "standard" timing values, set by BIOS "by SPD". As we have already noted, most BIOS routines for configuring the memory system tend to round fractional figures up — which in this case resulted in the 4-4-4-12 scheme, regular for DDR2-533.

Note that the modules under review refused to work with ECS PF21 Extreme motherboard (Testbeds 6 and 7). Of course, the motherboard was configured for the 200 MHz and 266 MHz modes beforehand with other memory modules.

Parameter	Testbed 1	Testbed 2	Testbed 3	Testbed 4	Testbed 5	Testbed 6	Testbed 7^*
Timings	4-4-4-12	4-4-4-12	4-4-4-12	4-4-4-12	4-4-4-12	—	—
Average memory read bandwidth, MB/sec	4892	4943	4770	4769	4719	—	—
Average memory write bandwidth, MB/sec	2113	2017	2017	2031	2002	—	—
Max. memory read bandwidth, MB/sec	6435	6497	6436	6436	6371	—	—
Max. memory write bandwidth, MB/sec	4287	4327	4323	4309	4266	—	—
Minimum Pseudo-Random Access Latency, ns	48.7	48.2	51.1	51.0	51.7	—	—
Maximum Pseudo-Random Access Latency, ns	56.0	55.4	58.3	58.3	59.0	—	—
Minimum Random Access Latency^**, ns	118.5	117.2	122.1	122.7	123.9	—	—
Maximum Random Access Latency^**, ns	137.7	136.0	140.9	141.3	142.7	—	—

^*FSB frequency — 266.7 MHz
^**16MB block size

The best memory system performance results (maximum memory bandwidth, minimum latency) were demonstrated by motherboards based on the i925X chipset — MSI 925X Neo (Testbed 2) and Gigabyte 8ANXP-D (Testbed 1). Motherboards on the i915P chipset — MSI 915P Neo2 (Testbed 3) and ECS PF4 Extreme (Testbed 4) are in between, and the last place is expectedly taken by the combo DDR/DDR2 solution MSI 915G Combo (Testbed 5) on the i915G chipset.

Stability tests

Timing values (except for t_CL) were adjusted "on the fly" due to the built-in RMMA feature that allows to change dynamically memory settings supported by the chipset. Memory operation stability was evaluated in our specially developed utility, which will soon be released as a stand-alone application and will come shipped with RMMA.

The minimal timing values, which we managed to get from these modules on all motherboards, 3-3-2, can be considered quite good. Namely, we managed to reduce t_CL by one, t_RCD — to haul up to the nominal value claimed in the SPD (which is just a tad lower than the absolute record of 3-2-2, achieved by Corsair XMS2 PRO modules), and to reduce t_RP by 2 cycles relative to the standard and by one relative to the nominal value. The last parameter (t_RAS), in the chipset settings, is ignored by these memory modules as well as by the majority of other DDR2 modules.

Parameter	Testbed 1	Testbed 2	Testbed 3	Testbed 4	Testbed 5	Testbed 6	Testbed 7^*
Timings	3-3-2	3-3-2	3-3-2	3-3-2	3-3-2	—	—
Average memory read bandwidth, MB/sec	5143	5191	5002	4990	4931	—	—
Average memory write bandwidth, MB/sec	2302	2320	2228	2248	2252	—	—
Max. memory read bandwidth, MB/sec	6509	6567	6538	6519	6460	—	—
Max. memory write bandwidth, MB/sec	4287	4331	4323	4309	4266	—	—
Minimum Pseudo-Random Access Latency, ns	46.6	46.2	48.5	48.8	49.3	—	—
Maximum Pseudo-Random Access Latency, ns	53.6	53.2	56.0	56.1	56.9	—	—
Minimum Random Access Latency^**, ns	103.4	102.8	107.9	108.3	109.2	—	—
Maximum Random Access Latency^**, ns	124.3	123.4	128.0	128.4	129.7	—	—

^*FSB frequency — 266.7 MHz
^**16MB block size

Overclocking timings allows to improve the results obtained on motherboards based on the i915P chipset (Testbeds 3 and 4), that is to bring them closer to the results of the leaders on the i925X chipset (Testbeds 1 and 2), as well as to reduce noticeably the random access latencies — by 13-15 ns.

Bottom line

The reviewed GeIL DDR2-667 memory modules from the XMS2 PRO series offer excellent performance (their memory bandwidth is almost equal to the maximum values and their latencies are very low) and demonstrate a significant timing overclocking potential — the minimal values (3-3-2) are very close to the absolute record, recently achieved by Corsair XMS2 PRO modules. Nevertheless, we should pay your attention to the "non 100%" compatibility of these modules with modern motherboards on 915 and 925 series chipsets — the modules refused to operate on one of these motherboards on the i925XE chipset, which allows to reveal the DDR2 potential to a greater extent.

Dmitri Besedin (dmitri_b@ixbt.com)
April 21, 2005.

Write a comment below. No registration needed!