Memory Module Analysis. Part 3: Kingston DDR-533 Modules

We proceed with the series of articles, dedicated to the low level analysis of the most important characteristics of memory modules using our RightMark Memory Analyzer test package. The object of today's review is a couple of Kingston DDR-533 modules, high performance HyperX series for overclockers.

Manufacturer Information

Module manufacturer: Kingston Technology
Manufacturer of module chips: Hynix Semiconductor
Web site of the module manufacturer: http://www.kingston.com/hyperx/products/khx.asp
Web site of the chip manufacturer:
http://www.hynix.com/eng/02_products/01_dram/index.jsp

Module Exterior

Photo of the memory module

Photo of the memory chip

Module and Chip Part Number

Module Part Number Expansion

Data sheet on Kingston HyperX DDR-533 modules does not contain the expansion of some Part Number components. The documentation provides only brief technical characteristics of the module.

Field	Value	Expansion
0	KHX4300/512	Module density: 512 MB Module organization: 64M x64 Voltage: Vdd: 2.7(+/-)0.1V, Vddq: 2.7(+/-)0.1V Timings (t_CL-t_RCD-t_RP-t_RAS): 3-4-4-8

Chip Part Number Expansion

Description of the part numbering system of Hynix memory chips:
http://www.hynix.com/eng/02_products/01_dram/down/DDR.pdf

Field	Value	Expansion
0-1	HY	Manufacturer code: HY = Hynix Memory
2-3	5D	Product family: 5D = DDR SDRAM
4	U	Process technology/voltage: U = VDD 2.5V, VDDQ 2.5V
5-6	56	Capacity and refresh: 56 = 256M, 8K refresh
7-8	8	Organization: 8 = x8
9	2	Number of banks: 2
10	2	Interface: 2 = SSTL_2
11	C	Chip generation C = 4th generation
12		Power Consumption: (blank) = normal
13	T	Package type T = TSOP
14		Die package type (blank) = a single die
15		Package material: (blank) = usual
16	—	(blank)
17-18	D5	Speed/Timings: D5 = DDR500, 3-4-4
19		Temperature conditions: (blank) = usual (0 — 70°C)

According to the expansion of the chip designation, these modules use DDR-500 chips, designed for 3-4-4 timings. However, the DDR-533 value in the Part Number expansion table of Hynix memory chips is not provided.

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 DDR:
JEDEC Standard No. 21-C, 4.1.2.4 — Appendix D, Rev. 1.0: SPD’s for DDR SDRAM

Parameter	Byte	Value	Expansion
Fundamental Memory Type	2	07h	DDR SDRAM
Number of Row Addresses on this assembly	3	0Dh	13 (RA0-RA12)
Number of Column Addresses on this assembly	4	0Ah	10 (CA0-CA9)
Number of DIMM Banks	5	02h	2 physical banks
Data Width of this assembly	6, 7	40h, 00h	64 bit
Voltage Interface Level of this assembly	8	04h	SSTL 2.5V
SDRAM Cycle time (t_CK) at maximum supported CAS# latency (CL X)	9	50h	5.00 ns (200.0 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	0Eh	BL = 2, 4, 8
Number of Banks on SDRAM Device	17	04h	4
CAS Latency (CL)	18	10h	CL = 3.0
Minimum clock cycle (t_CK) at reduced CAS# latency (CL X-0.5)	23	60h	6.00 ns (166.7 MHz)
Minimum clock cycle (t_CK) at reduced CAS# latency (CL X-1.0)	25	00h	Not defined
Minimum Row Precharge Time (t_RP)	27	48h	18.0 ns 3.60, CL = 3.0
Minimum Row Active to Row Active delay (t_RRD)	28	28h	10.0 ns 2.00, CL = 3.0
Minimum RAS to CAS delay (t_RCD)	29	48h	18.0 ns 3.60, CL = 3.0
Minimum Active to Precharge Time (t_RAS)	30	28h	40.0 ns 8.00, CL = 3.0
Module Bank Density	31	40h	256 MB
SDRAM Device Minimum Active to Active/Auto Refresh Time (t_RC)	41	3Ch	60.0 ns 12.00, CL = 3.0
SDRAM Device Minimum Auto-Refresh to Active/Auto-Refresh Command Period (t_RFC)	42	46h	70.0 ns 14.00, CL = 3.0
Maximum device cycle time (t_CKmax)	43	28h	10.0 ns
SPD Revision	62	10h	Revision 1.0
Checksum for Bytes 0-62	63	93h	147 (true)
Manufacturer’s JEDEC ID Code (only the first significant bytes are shown)	64-71	FFh, 98h	Kingston
Module Part Number	73-90	—	K
Module Manufacturing Date	93-94	04h, 19h	year 2004, week 25
Module Serial Number	95-98	4Ch, 2Ah, C5h, FFh	FFC52A4Ch

Except for some details provided below, SPD data looks typical. There is only one CAS# latency value supported — 3.0, it corresponds to the standard (for DDR-400, as the latest certified JEDEC standard) cycle time - 5.0ns. It should be noted that the manufacturer uses the standard value, it differs its approach from some other manufacturers (to use the true but non-standard cycle time — for example, 4.3ns). It's obviously done to ensure 100% compatibility of this module with different motherboards and BIOS versions, some of them refuse to automatically setup timings of non-standard memory modules. In this case timings can be defined as 3.0-3.6-3.6-8, 3.0-4-4-8 actually (considering that t_RCD, t_RP, and t_RAS cannot be non-integral values), which matches the specifications provided in the datasheet. Though this module claims to support only a single CL timing (3.0), the cycle time field at reduced CAS# latency (CL X-0.5 = 2.5) is also filled, it's set to 6.0ns, which corresponds to operating at 166.7 MHz, i.e. DDR-333 mode. We also want to note the following peculiarities in SPD data: it uses the only letter K to identify the module Part Number (which is probably characteristic of all Kingston modules) as well as correct data on the manufacturing and module serial number.

Testbed Configurations and Software

Testbed 1

CPU: 3.4 GHz Intel Pentium 4 (Prescott core, Socket 478)
Chipset: Intel 865PE
Motherboard: Albatron PX865PE Pro, BIOS dated 03/01/2004
Memory: 2x256 MB Kingmax DDR-466
Video: ATI Radeon 9800Pro
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: Intel Chipset Utility 6.0.1.1002

Testbed 2

CPU: 3.4 GHz Intel Pentium 4 (Prescott core, Socket 478)
Chipset: Intel 865G
Motherboard: ASUS P4P800-VM, BIOS 1012.002 dated 03/22/2004
Memory: 2x256 MB Kingmax DDR-466
Video: ATI Radeon 9800Pro
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: Intel Chipset Utility 6.0.1.1002

Testbed 3

CPU: 2.8 GHz Intel Pentium 4 (Prescott core, LGA775)
Chipset: Intel 865PE
Motherboard: ASUS P5P800, BIOS 1001.007 dated 07/08/2004
Memory: 2x256 MB Kingmax DDR-466
Video: ATI Radeon 9800Pro
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: Intel Chipset Utility 6.0.1.1002

Testbed 4

CPU: 2.8 GHz Intel Pentium 4 (Prescott core, LGA775)
Chipset: Intel 915G
Motherboard: ASUS P5GD1-VM, BIOS 1003.001 dated 07/14/2004
Memory: 2x256 MB Kingmax DDR-466
Video: ATI Radeon 9800Pro
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: Intel Chipset Utility 6.0.1.1002

Testbed 5

CPU: 2.8 GHz Intel Pentium 4 (Prescott core, LGA775)
Chipset: Intel 915G
Motherboard: Albatron PX915G Pro, BIOS dated 05/26/2004
Memory: 2x256 MB Kingmax DDR-466
Video: ATI Radeon 9800Pro
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: Intel Chipset Utility 6.0.1.1002

Testbed 6

CPU: 2.8 GHz Intel Pentium 4 (Prescott core, LGA775)
Chipset: Intel 915P
Motherboard: Albatron PX915P Pro, BIOS dated 05/26/2004
Memory: 2x256 MB Kingmax DDR-466
Video: ATI Radeon 9800Pro
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb
Drivers: Intel Chipset Utility 6.0.1.1002

Testbed #7

CPU: AMD Athlon 64 3500+ (NewCastle core, 2.2 GHz, Socket 939)
Chipset: VIA K8T800Pro
Motherboard: ASUS A8V Deluxe, BIOS 1005.027 dated 06/17/2004
Memory: 2x256 MB Kingmax DDR-466
Video: ATI Radeon 9800Pro
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb

Testbed #8

CPU: AMD Athlon 64 3500+ (NewCastle core, 2.2 GHz, Socket 939)
Chipset: NVIDIA nForce3 250
Motherboard: Gigabyte K8NS Ultra-939, BIOS F2 dated 09/24/2004
Memory: 2x256 MB Kingmax DDR-466
Video: ATI Radeon 9800Pro
HDD: WD Raptor WD360, SATA, 10000 rpm, 36Gb

Test Results

According to our method, the memory modules were tested in two modes. The first series of tests (performance tests) were carried out in standard mode (DDR-400) with standard timings, set in BIOS by default according to SPD data (Memory Timings: "by SPD") The second series (stability tests) is carried out in the extreme mode, still at the standard frequency, but with minimum possible timings for this module on a given motherboard.

Performance tests

Parameter^*	Testbed 1 (PAT)	Testbed 1	Testbed 2	Testbed 3 (PAT)	Testbed 4
Timings	2.5-3- 3-7	3.0-4- 4-8	3.0-4- 4-8	3.0-4- 4-8	3.0-4- 4-8
Average memory read bandwidth, MB/sec	5343	4732	4357	4918	4433
Average memory write bandwidth, MB/sec	1822	1635	1617	1735	1855
Max. memory read bandwidth, MB/sec	6305	6188	6172	6163	6181
Max. memory write bandwidth, MB/sec	4269	4248	4220	4231	4256
Minimum Pseudo-Random Access Latency^**, ns	45.9	54.7	60.0	50.1	55.0
Maximum Pseudo-Random Access Latency^**, ns	54.1	65.2	70.4	61.4	64.0
Minimum Random Access Latency^**, ns	98.5	118.7	129.3	111.2	138.0
Maximum Random Access Latency^**, ns	119.3	137.7	149.2	135.3	161.5

Parameter^*	Testbed 5	Testbed 6	Testbed 7	Testbed 8
Timings	3.0-4- 4-8	3.0-4- 4-8	2.5-4- 4-6	3.0-4- 4-8
Average memory read bandwidth, MB/sec	4454	4456	4104	3910
Average memory write bandwidth, MB/sec	1847	1848	2644	2478
Max. memory read bandwidth, MB/sec	6210	6217	6230	6204
Max. memory write bandwidth, MB/sec	4256	4267	6201	6234
Minimum Pseudo-Random Access Latency^**, ns	54.6	54.7	33.7	36.2
Maximum Pseudo-Random Access Latency^**, ns	63.8	63.7	37.7	40.2
Minimum Random Access Latency^**, ns	137.9	137.6	86.9	74.4
Maximum Random Access Latency^**, ns	161.6	161.0	92.1	78.5

^*the best results are marked in bold (all other things being equal)
^**16MB block size

The majority of motherboards set timings correctly in the first series of tests (3.0-4-4-8), i.e. the ones claimed by the manufacturer and written in the SPD chip (that's one of the advantages of the above mentioned standard approach to writing SPD data for non-standard modules). The only exceptions are Albatron PX865PE Pro motherboards (Testbed 1) with enabled PAT — lower timings (in our case — 2.5-3-3-7) are obviously one of the components (not the major one) of Performance Mode, implemented in this motherboard. For some vague reasons, ASUS A8V Deluxe (Testbed 7) does approximately the same thing, using the 2.5-4-4-6 timings by default.

The best result in most parameters among Pentium 4 based platforms (Testbed 1-6) is demonstrated by Albatron PX865PE Pro (Testbed 1) with enabled PAT. This is achieved not so much due to PAT as to a faster timing scheme — because the same motherboard with disabled PAT and standard timings takes the third place, while the second one is taken by ASUS P5P800 (Testbed 3), which also runs in PAT mode. ASUS P4P800-VM (Testbed 2) is a tad slower, it's based on i865G versus i865P, while 915-series motherboards (Testbeds 4-6) are noticeably inferior, especially in memory latencies, being the slowest models.

It's not correct to compare the above results with those obtained on the Athlon 64 platform, so we'll have to confine ourselves to comparing the results obtained on two motherboards — ASUS A8V Deluxe (Testbed 7) and Gigabyte K8NS Ultra-939 (Testbed 8). Though the memory controller is integrated into a processor, there are still some differences between the boards, strange and noticeable at that. A8V Deluxe (Testbed 7) takes the lead in most parameters, while Gigabyte K8NS Ultra-939 (Testbed 8) is the first in terms of random access latencies. When we reviewed stranger behaviour of motherboards on VIA K8T800Pro and NVIDIA nForce3 250 (right up to complete inoperability of memory modules on one of the motherboards, but different operation on another), we assumed that the differences must have to do with different configuration methods for the integrated memory controller in Athlon 64, performed at BIOS POST.

Stability tests

The second series of tests was carried out with minimum possible timing values not resulting in memory glitches.

Parameter^*	Testbed 1 (PAT)	Testbed 1	Testbed 2	Testbed 3 (PAT)	Testbed 4
Timings	2.5-3- 3-5	2.5-3- 3-5	2.5-3- 3-5	2.5-3- 3-5	2.5-3- 3-4
Average memory read bandwidth, MB/sec	5343	4764	4396	4978	4495
Average memory write bandwidth, MB/sec	1827	1808	1774	1862	1943
Max. memory read bandwidth, MB/sec	6306	6269	6222	6274	6276
Max. memory write bandwidth, MB/sec	4269	4257	4228	4248	4256
Minimum Pseudo-Random Access Latency^**, ns	46.0	54.7	59.9	49.8	54.5
Maximum Pseudo-Random Access Latency^**, ns	54.0	62.4	67.9	57.8	63.5
Minimum Random Access Latency^**, ns	98.6	113.6	124.2	106.9	129.1
Maximum Random Access Latency^**, ns	119.3	133.6	143.7	129.9	153.9

Parameter^*	Testbed 5	Testbed 6	Testbed 7	Testbed 8
Timings	2.5-3- 3-4	2.5-3- 3-4	2.5-3- 3-5	2.5-3- 3-5
Average memory read bandwidth, MB/sec	4496	4489	4100	4064
Average memory write bandwidth, MB/sec	1957	2048	2671	2483
Max. memory read bandwidth, MB/sec	6255	6244	6277	6242
Max. memory write bandwidth, MB/sec	4268	4268	6119	6238
Minimum Pseudo-Random Access Latency^**, ns	54.4	54.5	33.5	33.8
Maximum Pseudo-Random Access Latency^**, ns	63.5	63.6	37.6	37.8
Minimum Random Access Latency^**, ns	128.8	128.8	78.6	66.8
Maximum Random Access Latency^**, ns	153.5	153.6	83.5	70.8

^*the best results are marked in bold
^**16MB block size

You may easily notice that you can set any t_RAS in configuration registers of the chipset (up to 4 inclusive) without damaging memory operating stability. It means that this parameter is ignored by these modules (as well as by the majority of other modules), they use an internal value instead (nobody knows what exactly). The other timing values are common to all tested systems — 2.5-3-3, you cannot possibly call them record breaking.

Switching the memory to extreme mode did not change the alignment of forces among platforms with Pentium 4 as well as (Testbeds 1-6) Athlon 64 processors (Testbeds 7-8).

Bottom line

The Kingston HyperX DDR-533 modules tested demonstrate excellent compatibility with various motherboards on Intel 865P/G and 915P/G, VIA K8T800Pro and NVIDIA nForce 3 chipsets, average overclocking capacity in terms of timings (they allow 2.5-3-3 timings in DDR-400 mode), and good operating stability in these conditions. We can note the following motherboards that lead in memory performance: Albatron PX865PE Pro (Testbed 1) and ASUS P5P800 (Testbed 3), partially due to the PAT option. The worst results with these modules are demonstrated by motherboards based on i915P/G chipsets — ASUS P5GD1-VM (Testbed 4), Albatron PX915G Pro (Testbed 5), and Albatron PX915P Pro (Testbed 6).

Dmitri Besedin (dmitri_b@ixbt.com)

June 24, 2005.

Write a comment below. No registration needed!