Reserved for information about DDR3.

The results so far.

i7 4790K @4.9. Asus Z87-A. GTX 780. Windows XP SP3.

Timings are in the CLx/xx-tRCDx/xx-tRPx/xx-tRASxx-tRFCxx/xxx-CRx format.

.

When overclocking I stick to (100:133) 1333, 1600, 1866 , 2133 , 2400 and 2666 MHz settings and 1.5V/1.65V and go for the best performing combination of speed (in MHz) and timings. My overclocked CPUs memory controller seem to have a hard time handling anything above 2400 MHz using the extremely safe voltages I use during this testing. I do set all secondary timings manually, DRAM Refresh Interval is set to its maximum value and I always try to set tRRD to 4 and tFAW to 16 if possible. I slack tRRD in steps of 1 but tFAW in steps of 4 (even if tRRD is stable at 4). I mostly leave the tertiary timings on auto unless I see something odd.

Will this method squeeze the absolute maximum amount of performance out of the memory kits? No of course not but it should be good enough to see how different memory settings, module layouts and memory ICs stack up against each other for actual use.

.

The testing methodology improved and produces scores with less variability. The first memory kits tested needed to be tested again, this is now done. Memory kit #7 that during the first test showed glitched performance using the SPD settings now performed just as expected.

.

SuperPI 32M (Standard deviation <0.5s)

6m 27.516s. OC @2400 10-11-10-22-222-1 tRRD4 tFAW16 @1.65V. 2x8GB dual rank Hynix 12800U 4Gbit MFR 1407. (#25) Page4
6m 28.500s. OC @2400 10-12-11-24-156-1 tRRD4 tFAW16 @1.65V. 2x4GB d.r G.Skill F3-17000CL11D Samsung 2Gbit E-die (#11) Page3
6m 29.672s. OC @2133 9-10-9-24-74-1 tRRD4 tFAW16 @1.65V. 2x2GB dual rank Elpida 10600U 1Gbit BDSE. (#4) Page2
6m 29.891s. OC @2400 11-13-11-25-230-1 tRRD4 tFAW16 @1.65V. 2x8GB dual rank Adata AX3U1600W8G9-DB 4Gbit. (#15) Page4
6m 30.531s. OC @2400 11-13-11-25-224-1 tRRD5 tFAW20 @1.65V. 2x8GB dual rank Samsung 12800U 4Gbit D-die. (#20) Page4
6m 30.531s. OC @2400 11-13-12-24-194-1 tRRD5 tFAW20 @1.65V. 2x8GB dual rank Hynix 12800U 4Gbit AFR. (#19) Page4
6m 30.578s. OC @2400 11-13-11-27-108-2 tRRD5 tFAW20 @1.65V. 2x4GB d.r Corsair Vengeance ver5.11 Hynix 2Gbit BFR. (#23) Page4
6m 30.593s. OC @2400 9-11-11-21-96-1 tRRD4 tFAW20 @1.65V. 2x2GB single rank Samsung 12800E 2Gbit D-die HCK0 (#9) Page2
6m 31.313s. OC @2400 10-11-11-22-96-1 tRRD4 tFAW16 @1.65V. 2x2GB s.r Samsung 10600U unmatched 2Gbit D-die HCH9. (#13) Page3
6m 32.093s. OC @2133 9-12-10-22-98-1 tRRD4 tFAW16 @1.65V. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1109. (#21) Page4
6m 33.219s. OC @2133 9-11-10-22-94-2 tRRD4 tFAW16 @1.65V. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1202. (#18) Page4
6m 34.500s OC @1866 8-10-9-20-60-1 tRRD4 tFAW16 @1.65V. 2x2GB dual rank Samsung 10600U 1Gbit E-die HCH9. (#6) Page2
6m 34.828s. OC @1866 8-10-9-20-61-1 tRRD4 tFAW16 @1.65V. 2x2GB dual rank Samsung 10600U 1Gbit E-die HCH9.(#1) Page1
6m 35.094s. OC @1866 9-9-9-22-58-1 tRRD4 tFAW16 @1.65V. 2x2GB d.r Corsair Dominator ver3.5 Micron 1Gbit D9KPT/KPV (#8) Page2
6m 35.125s. OC @1866 9-9-9-21-56-1 tRRD4 tFAW16 @1.65V. 2x2GB dual rank Micron 10600U 1Gbit D9KPT. (#16) page4
6m 35.188s. OC @1866 8-10-9-20-60-1 tRRD4 tFAW16 @1.65V. 2x2GB d.r Samsung 10600UE unmatched 1Gbit E-die HCH9. (#12) Page3
6m 35.297s. OC @1866 9-9-9-22-58-1 tRRD4 tFAW16 @1.65V. 2x2GB d.r Corsair Dominator ver3.1 Micron 1Gbit D9KPT/KPV (#17) Page4
6m 35.609s. OC @2133 9-11-10-27-84-1 tRRD4 tFAW16 @1.65V. 2x2GB s.r Crucial Ballistix 12800U Micron 2Gbit D9PFJ (#10) Page2
6m 36.188s. OC @2133 11-11-10-26-124-1 tRRD4 tFAW16 @1.5V. 2x2GB single rank Corsair Dominator ver8.14 Nanya 2Gbit (#2) Page1
6m 36.531s. XMP @2133 11-11-11-28-171-2 @1.6V. 2x4GB dual rank G.Skill F3-17000CL11D-8GBXL Samsung 2Gbit E-die. (#11) page3
6m 38.844s. OC @1866 9-9-8-20-100-1 tRRD4 tFAW16 @1.5V. 2x2GB single rank Kingston 10600U 2Gbit (PSC?). (#5) page2
6m 39.922s. OC @1600 8-8-7-20-84-1 tRRD4 tFAW16 @1.65V. 2x4GB d.r Corsair Vengeance ver2.12 Elpida 2Gbit BCBD/BCSE. (#22) Page4
6m 40.625s. OC @1600 9-8-7-18-88-1 tRRD4 tFAW16 @1.65V. 2x4GB d.r G.Skill 12800CL9D-8GBSR2 Elpida 2Gbit BCBD/BCSE. (#14) Page3
6m 41.578s. OC @1866 10-9-9-23-168-1 tRRD4 tFAW16 @1.65V. 2x4GB single rank Corsair Value Ram 10600U 4Gbit. (#24) Page4
6m 44.422s. OC @1600 10-8-7-22-68-1 tRRD4 tFAW16 @1.65V. 2x2GB single rank Micron 10600U 2Gbit D9LGK. (#7) Page2
6m 44.843s. OC @1600 9-8-8-22-94-1 tRRD4 tFAW16 @1.65V. 2x2GB single rank Kingston 10600U Elpida 2Gbit BCSE. (#3) Page1
6m 45.140s. XMP @1600 8-8-8-24-88-2 @1.65V. 2x2GB dual rank Corsair Dominator ver3.5 Micron 1Gbit D9KPT/KPV. (#8) Page2
6m 45.843s. SPD @1600 9-9-9-28-128-1. 2x4GB dual rank G.Skill Sniper F3-12800CL9D-8GBSR2 Elpida 2Gbit BCBD/BCSE. (#14) page3
6m 46.344s. XMP @1600 9-9-9-24-74-2 @1.65V. 2x2GB dual rank Corsair Dominator ver3.1 Micron 1Gbit D9KPT/KPV (#17) Page4
6m 47.203s. XMP @1600 9-9-9-24-128-2 @1.65V. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1202. (#18) Page4
6m 47.297s. XMP @1600 9-9-9-24-128-2 @1.5V. 2x4GB dual rank Corsair Vengeance ver2.12 Elpida 2Gbit BCBD/BCSE. (#22) Page4
6m 47.500s. XMP @1600 9-9-9-24-128-2 @1.5V. 2x4GB dual rank Corsair Vengeance ver5.11 Hynix 2Gbit BFR. (#23) Page4
6m 47.687s. XMP @1600 9-9-9-24-128-2 @1.65V. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1109. (#21) Page4
6m 49.750s. SPD @1600 11-11-11-28-128-1. 2x4GB dual rank G.Skill F3-17000CL11D-8GBXL Samsung 2Gbit E-die. (#11) page2
6m 50.656s. SPD @1600 9-9-9-24-128-1. 2x2GB single rank Crucial Ballistix 12800U Micron 2Gbit D9PFJ. (#10) Page2
6m 51.032s. XMP @1600 8-8-8-24-128-2 @1.65V. 2x2GB single rank Corsair Dominator ver8.14 Nanya 2Gbit. (#2) Page1
6m 51.312s. XMP @1600 9-9-9-24-208-2 @1.5V. 2x8GB dual rank Adata AX3U1600W8G9-DB 4Gbit. (#15) Page4
6m 52.016s. SPD @1600 11-11-11-28-208-1. 2x8GB dual rank Samsung 12800U 4Gbit D-die. (#20) Page4
6m 52.078s. SPD @1600 11-11-11-28-208-1. 2x8GB dual rank Hynix 12800U 4Gbit AFR. (#19) Page4
6m 52.313s. SPD @1600 11-11-11-28-208-1. 2x8GB dual rank Hynix 12800U 4Gbit MFR 1407. (#25) Page4
6m 54.156s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Samsung 10600U 1Gbit E-die HCH9. (#1) Page1
6m 54.312s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Micron 10600U 1Gbit D9KPT. (#16) page4
6m 54.750s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Samsung 10600U 1Gbit E-die HCH9. (#6) Page2
6m 54.782s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Elpida 10600U 1Gbit BDSE. (#4) Page2
6m 54.782s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Corsair Dominator ver3.1 Micron 1Gbit D9KPT/KPV (#17) Page4
6m 54.828s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Samsung 10600UE unmatched 1Gbit E-die HCH9. (#12) Page3
6m 55.656s. SPD @1333 9-9-9-24-107-1. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1202. (#18) Page4
6m 55.672s. SPD @1333 9-9-9-24-107-1. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1109. (#21) Page4
6m 55.672s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Corsair Dominator ver3.5 Micron 1Gbit D9KPT/KPV. (#8) Page2
6m 55.844s. SPD @1333 9-9-9-24-107-1. 2x4GB dual rank Corsair Vengeance ver5.11 Hynix 2Gbit BFR. (#23) Page4
6m 55.906s. SPD @1333 9-9-9-24-107-1. 2x4GB dual rank Corsair Vengeance ver2.12 Elpida 2Gbit BCBD/BCSE. (#22) Page4
6m 56.172s. SPD @1600 11-11-11-28-128-1. 2x2GB single rank Samsung 12800E 2Gbit D-die HCK0. (#9) Page2
6m 59.515s. SPD @1333 9-9-9-24-174-1. 2x8GB dual rank Adata AX3U1600W8G9-DB 4Gbit. (#15) Page4
7m 01.328s. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Kingston 10600U 2Gbit (PSC?). (#5) page2
7m 01.875s. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Micron 10600U 2Gbit D9LGK. (#7) Page2
7m 01.891s. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Kingston 10600U Elpida 2Gbit BCSE. (#3) Page1
7m 02.031s. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Corsair Dominator ver8.14 Nanya 2Gbit. (#2) Page1
7m 02.203s. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Samsung 10600U unmatched 2Gbit D-die HCH9. (#13) Page3
7m 03.657s. SPD @1333 9-9-9-24-174-1. 2x4GB single rank Corsair Value Ram 10600U 4Gbit. (#24) Page4

.

Frybench x86 (Render time) (Standard deviation <0.5s)

3m 29s. OC @2400 10-11-10-22-222-1 tRRD4 tFAW16 @1.65V. 2x8GB dual rank Hynix 12800U 4Gbit MFR 1407. (#25) Page4
3m 30s. OC @2400 10-12-11-24-156-1 tRRD4 tFAW16 @1.65V. 2x4GB d.r G.Skill F3-17000CL11D Samsung 2Gbit E-die (#11) Page3
3m 31s. OC @2133 9-10-9-24-74-1 tRRD4 tFAW16 @1.65V. 2x2GB dual rank Elpida 10600U 1Gbit BDSE. (#4) Page2
3m 31s. OC @2400 11-13-11-25-230-1 tRRD4 tFAW16 @1.65V. 2x8GB dual rank Adata AX3U1600W8G9-DB 4Gbit. (#15) Page4
3m 32s OC @2400 11-13-11-25-224-1 tRRD5 tFAW20 @1.65V. 2x8GB dual rank Samsung 12800U 4Gbit D-die. (#20) Page4
3m 32s. OC @2400 11-13-12-24-194-1 tRRD5 tFAW20 @1.65V. 2x8GB dual rank Hynix 12800U 4Gbit AFR. (#19) Page4
3m 32s. OC @2400 9-11-11-21-96-1 tRRD4 tFAW20 @1.65V. 2x2GB single rank Samsung 12800E 2Gbit D-die HCK0 (#9) Page2
3m 32s. OC @2400 10-11-11-22-96-1 tRRD4 tFAW16 @1.65V. 2x2GB s.r Samsung 10600U unmatched 2Gbit D-die HCH9. (#13) Page3
3m 33s. OC @2400 11-13-11-27-108-2 tRRD5 tFAW20 @1.65V. 2x4GB d.r Corsair Vengeance ver5.11 Hynix 2Gbit BFR. (#23) Page4
3m 33s. OC @2133 9-12-10-22-98-1 tRRD4 tFAW16 @1.65V. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1109. (#21) Page4
3m 34s. OC @2133 9-11-10-22-94-2 tRRD4 tFAW16 @1.65V. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1202. (#18) Page4
3m 34s. OC @1866 8-10-9-20-60-1 tRRD4 tFAW16 @1.65V. 2x2GB dual rank Samsung 10600U 1Gbit E-die HCH9. (#6) Page2
3m 34s. OC @1866 8-10-9-20-61-1 tRRD4 tFAW16 @1.65V. 2x2GB dual rank Samsung 10600U 1Gbit E-die HCH9. (#1) Page1
3m 34s. OC @1866 8-10-9-20-60-1 tRRD4 tFAW16 @1.65V. 2x2GB d.r Samsung 10600UE unmatched 1Gbit E-die HCH9. (#12) Page3
3m 34s. OC @1866 9-9-9-21-56-1 tRRD4 tFAW16 @1.65V. 2x2GB dual rank Micron 10600U 1Gbit D9KPT. (#16) page4
3m 34s. OC @1866 9-9-9-22-58-1 tRRD4 tFAW16 @1.65V. 2x2GB d.r Corsair Dominator ver3.5 Micron 1Gbit D9KPT/KPV (#8) Page2
3m 34s. OC @1866 9-9-9-22-58-1 tRRD4 tFAW16 @1.65V. 2x2GB d.r Corsair Dominator ver3.1 Micron 1Gbit D9KPT/KPV (#17) Page4
3m 36s. OC @2133 9-11-10-27-84-1 tRRD4 tFAW16 @1.65V. 2x2GB s.r Crucial Ballistix 12800U Micron 2Gbit D9PFJ(#10) Page2
3m 37s. OC @2133 11-11-10-26-124-1 tRRD4 tFAW16 @1.5V. 2x2GB single rank Corsair Dominator ver8.14 Nanya 2Gbit (#2) Page1
3m 37s. OC @1866 9-9-8-20-100-1 tRRD4 tFAW16 @1.5V. 2x2GB single rank Kingston 10600U 2Gbit (PSC?). (#5) page2
3m 38s. XMP @2133 11-11-11-28-171-2 @1.6V. 2x4GB dual rank G.Skill F3-17000CL11D-8GBXL Samsung 2Gbit E-die. (#11) page3
3m 39s. OC @1866 10-9-9-23-168-1 tRRD4 tFAW16 @1.65V. 2x4GB single rank Corsair Value Ram 10600U 4Gbit. (#24) Page4
3m 39s. OC @1600 8-8-7-20-84-1 tRRD4 tFAW16 @1.65V. 2x4GB d.r Corsair Vengeance ver2.12 Elpida 2Gbit BCBD/BCSE. (#22) Page4
3m 40s. OC @1600 9-8-8-22-94-1 tRRD4 tFAW16 @1.5V. 2x2GB single rank Kingston 10600U Elpida 2Gbit BCSE. (#3) Page1
3m 40s. XMP @1600 8-8-8-24-88-2 @1.65V. 2x2GB dual rank Corsair Dominator ver3.5 Micron 1Gbit D9KPT/KPV. (#8) Page2
3m 41s. OC @1600 9-8-7-18-88-1 tRRD4 tFAW16 @1.65V. 2x4GB d.r G.Skill 12800CL9D-8GBSR2 Elpida 2Gbit BCBD/BCSE. (#14) Page3
3m 41s. OC @1600 10-8-7-22-68-1 tRRD4 tFAW16 @1.65V. 2x2GB single rank Micron 10600U 2Gbit D9LGK. (#7) Page2
3m 42s. XMP @1600 9-9-9-24-128-2 @1.65V. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1202. (#18) Page4
3m 42s. XMP @1600 9-9-9-24-128-2 @1.65V. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1109. (#21) Page4
3m 44s. XMP @1600 9-9-9-24-128-2 @1.5V. 2x4GB dual rank Corsair Vengeance ver5.11 Hynix 2Gbit BFR. (#23) Page4
3m 45s. XMP @1600 9-9-9-24-74-2 @1.65V. 2x2GB dual rank Corsair Dominator ver3.1 Micron 1Gbit D9KPT/KPV (#17) Page4
3m 45s. SPD @1600 9-9-9-28-128-1. 2x4GB dual rank G.Skill Sniper F3-12800CL9D-8GBSR2 Elpida 2Gbit BCBD/BCSE. (#14) page3
3m 45s. XMP @1600 9-9-9-24-128-2 @1.5V. 2x4GB dual rank Corsair Vengeance ver2.12 Elpida 2Gbit BCBD/BCSE. (#22) Page4
3m 46s. SPD @1600 11-11-11-28-128-1. 2x4GB dual rank G.Skill F3-17000CL11D-8GBXL Samsung 2Gbit E-die. (#11) page2
3m 47s. SPD @1600 9-9-9-24-128-1. 2x2GB single rank Crucial Ballistix 12800U Micron 2Gbit D9PFJ. (#10) Page2
3m 47s. XMP @1600 8-8-8-24-128-2. @1.65V 2x2GB single rank Corsair Dominator ver8.14 Nanya 2Gbit. (#2) Page 1
3m 47s. XMP @1600 9-9-9-24-208-2 @1.5V. 2x8GB dual rank Adata AX3U1600W8G9-DB 4Gbit. (#15) Page4
3m 47s. SPD @1600 11-11-11-28-208-1. 2x8GB dual rank Samsung 12800U 4Gbit D-die. (#20) Page4
3m 47s. SPD @1600 11-11-11-28-208-1. 2x8GB dual rank Hynix 12800U 4Gbit MFR 1407. (#25) Page4
3m 47s. SPD @1600 11-11-11-28-208-1. 2x8GB dual rank Hynix 12800U 4Gbit AFR. (#19) Page4
3m 52s. SPD @1600 11-11-11-28-128-1. 2x2GB single rank Samsung 12800E 2Gbit D-die HCK0. (#9) Page2
3m 52s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Samsung 10600U 1Gbit E-die HCH9. (#6) Page2
3m 52s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Samsung 10600U 1Gbit E-die HCH9. (#1) Page1
3m 52s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Samsung 10600UE unmatched 1Gbit E-die HCH9. (#12) Page3
3m 52s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Elpida 10600U 1Gbit BDSE. (#4) Page2
3m 52s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Micron 10600U 1Gbit D9KPT. (#16) page4
3m 52s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Corsair Dominator ver3.1 Micron 1Gbit D9KPT/KPV (#17) Page4
3m 52s. SPD @1333 9-9-9-24-107-1. 2x4GB dual rank Corsair Vengeance ver5.11 Hynix 2Gbit BFR. (#23) Page4
3m 52s. SPD @1333 9-9-9-24-107-1. 2x4GB dual rank Corsair Vengeance ver2.12 Elpida 2Gbit BCBD/BCSE. (#22) Page4
3m 53s. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Corsair Dominator ver3.5 Micron 1Gbit D9KPT/KPV. (#8) Page2
3m 53s. SPD @1333 9-9-9-24-107-1. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1109. (#21) Page4
3m 54s. SPD @1333 9-9-9-24-107-1. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1202. (#18) Page4
3m 54s. SPD @1333 9-9-9-24-174-1. 2x8GB dual rank Adata AX3U1600W8G9-DB 4Gbit. (#15) Page4
3m 57s. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Kingston 10600U 2Gbit (PSC?). (#5) page2
3m 57s. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Samsung 10600U unmatched 2Gbit D-die HCH9. (#13) Page3
3m 57s. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Corsair Dominator ver8.14 Nanya 2Gbit. (#2) Page1
3m 57s. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Kingston 10600U Elpida 2Gbit BCSE. (#3) Page1
3m 57s. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Micron 10600U 2Gbit D9LGK. (#7) Page2
3m 58s. SPD @1333 9-9-9-24-174-1. 2x4GB single rank Corsair Value Ram 10600U 4Gbit. (#24) Page4

.

Doom III (Standard deviation <1FPS)

673.1 FPS. OC @2400 10-11-10-22-222-1 tRRD4 tFAW16 @1.65V. 2x8GB dual rank Hynix 12800U 4Gbit MFR 1407. (#25) Page4
672.9 FPS. OC @2400 9-11-11-21-96-1 tRRD4 tFAW20 @1.65V. 2x2GB single rank Samsung 12800E 2Gbit D-die HCK0 (#9) Page2
671.9 FPS. OC @2400 10-12-11-24-156-1 tRRD4 tFAW16 @1.65V, 2x4GB d.r G.Skill F3-17000CL11D Samsung 2Gbit E-die (#11) Page3
671.5 FPS. OC @2400 10-11-11-22-96-1 tRRD4 tFAW16 @1.65V. 2x2GB s.r Samsung 10600U unmatched 2Gbit D-die HCH9. (#13) Page3
670.8 FPS. OC @2400 11-13-11-25-224-1 tRRD5 tFAW20 @1.65V. 2x8GB dual rank Samsung 12800U 4Gbit D-die. (#20) Page4
670.4 FPS. OC @2400 11-13-11-25-230-1 tRRD4 tFAW16 @1.65V. 2x8GB dual rank Adata AX3U1600W8G9-DB 4Gbit. (#15) Page4
670.2 FPS. OC @2400 11-13-12-24-194-1 tRRD5 tFAW20 @1.65V. 2x8GB dual rank Hynix 12800U 4Gbit AFR. (#19) Page4
669.4 FPS. OC @2400 11-13-11-27-108-2 tRRD5 tFAW20 @1.65V. 2x4GB d.r Corsair Vengeance ver5.11 Hynix 2Gbit BFR. (#23) Page4
669.2 FPS. OC @2133 9-10-9-24-74-1 tRRD4 tFAW16 @1.65V. 2x2GB dual rank Elpida 10600U 1Gbit BDSE. (#4) Page2
667.5 FPS. OC @2133 9-12-10-22-98-1 tRRD4 tFAW16 @1.65V. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1109. (#21) Page4
666.9 FPS. OC @2133 9-11-10-27-84-1 tRRD4 tFAW16 @1.65V. 2x2GB s.r Crucial Ballistix 12800U Micron 2Gbit D9PFJ. (#10) Page2
666.3 FPS. OC @2133 11-11-10-26-124-1 tRRD4 tFAW16 @1.5V. 2x2GB single rank Corsair Dominator ver8.14 Nanya 2Gbit (#2) Page1
665.6 FPS. OC @2133 9-11-10-22-94-2 tRRD4 tFAW16 @1.65V. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1202. (#18) Page4
664.2 FPS. OC @1866 8-10-9-20-60-1 tRRD4 tFAW16 @1.65V. 2x2GB dual rank Samsung 10600U 1Gbit E-die HCH9. (#6) Page2
664.0 FPS. OC @1866 8-10-9-20-61-1 tRRD4 tFAW16 @1.65V. 2x2GB dual rank Samsung 10600U 1Gbit E-die HCH9. (#1) Page1
663.8 FPS. OC @1866 9-9-9-21-56-1 tRRD4 tFAW16 @1.65V. 2x2GB dual rank Micron 10600U 1Gbit D9KPT. (#16) page4
663.0 FPS. OC @1866 8-10-9-20-60-1 tRRD4 tFAW16 @1.65V. 2x2GB d.r Samsung 10600UE unmatched 1Gbit E-die HCH9. (#12) Page3
662.6 FPS. OC @1866 9-9-9-22-58-1 tRRD4 tFAW16 @1.65V. 2x2GB d.r Corsair Dominator ver3.1 Micron 1Gbit D9KPT/KPV (#17) Page4
662.6 FPS. OC @1866 9-9-9-22-58-1 tRRD4 tFAW16 @1.65V. 2x2GB d.r Corsair Dominator ver3.5 Micron 1Gbit D9KPT/KPV (#8) Page2
661.5 FPS. OC @1866 9-9-8-21-102-1 tRRD4 tFAW16 @1.5V. 2x2GB single rank Kingston 10600U 2Gbit (PSC?). (#5) page2
661.5 FPS. OC @1866 10-9-9-23-168-1 tRRD4 tFAW16 @1.65V. 2x4GB single rank Corsair Value Ram 10600U 4Gbit. (#24) Page4
661.3 FPS. XMP @2133 11-11-11-28-171-2 @1.6V. 2x4GB dual rank G.Skill F3-17000CL11D-8GBXL Samsung 2Gbit E-die. (#11) page3
658.3 FPS. OC @1600 8-8-7-20-84-1 tRRD4 tFAW16 @1.65V. 2x4GB d.r Corsair Vengeance ver2.12 Elpida 2Gbit BCBD/BCSE. (#22) Page4
656.9 FPS. OC @1600 9-8-7-18-88-1 tRRD4 tFAW16 @1.65V. 2x4GB d.r G.Skill 12800CL9D-8GBSR2 Elpida 2Gbit BCBD/BCSE. (#14) Page3
656.7 FPS. XMP @1600 8-8-8-24-88-2 @1.65V. 2x2GB dual rank Corsair Dominator ver3.5 Micron 1Gbit D9KPT/KPV. (#8) Page2
655.3 FPS. OC @1600 9-9-8-22-94-1 tRRD4 tFAW16 @1.5V. 2x2GB single rank Kingston 10600U Elpida 2Gbit BCSE. (#3) Page1
654.7 FPS. OC @1600 10-8-7-22-68-1 tRRD4 tFAW16 @1.65V. 2x2GB single rank Micron 10600U 2Gbit D9LGK. (#7) Page2
652.7 FPS. XMP @1600 9-9-9-24-74-2 @1.65V. 2x2GB dual rank Corsair Dominator ver3.1 Micron 1Gbit D9KPT/KPV (#17) Page4
652.7 FPS. XMP @1600 9-9-9-24-128-2 @1.65V. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1109. (#21) Page4
652.5FPS. XMP @1600 9-9-9-24-128-2 @1.5V. 2x4GB dual rank Corsair Vengeance ver5.11 Hynix 2Gbit BFR. (#23) Page4
651.9 FPS. XMP @1600 8-8-8-24-128-2 @1.65V. 2x2GB single rank Corsair Dominator ver8.14 Nanya 2Gbit. (#2) Page1
651.5 FPS. SPD @1600 9-9-9-28-128-1. 2x4GB dual rank G.Skill Sniper F3-12800CL9D-8GBSR2 Elpida 2Gbit BCBD/BCSE. (#14) page3
651.3 FPS. XMP @1600 9-9-9-24-128-2 @1.5V. 2x4GB dual rank Corsair Vengeance ver2.12 Elpida 2Gbit BCBD/BCSE. (#22) Page4
651.1 FPS. XMP @1600 9-9-9-24-128-2 @1.65V. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1202. (#18) Page4
650.7 FPS. SPD @1600 9-9-9-24-128-1. 2x2GB single rank Crucial Ballistix 12800U Micron 2Gbit D9PFJ. (#10) Page2
649.7 FPS. XMP @1600 9-9-9-24-208-2 @1.5V. 2x8GB dual rank Adata AX3U1600W8G9-DB 4Gbit. (#15) Page4
647.6 FPS. SPD @1600 11-11-11-28-128-1. 2x4GB dual rank G.Skill F3-17000CL11D-8GBXL Samsung 2Gbit E-die. (#11) page2
645.2 FPS. SPD @1600 11-11-11-28-208-1. 2x8GB dual rank Samsung 12800U 4Gbit D-die. (#20) Page4
645.2 FPS. SPD @1600 11-11-11-28-208-1. 2x8GB dual rank Hynix 12800U 4Gbit MFR 1407. (#25) Page4
644.5 FPS. SPD @1600 11-11-11-28-208-1. 2x8GB dual rank Hynix 12800U 4Gbit AFR. (#19) Page4
644.5 FPS. SPD @1600 11-11-11-28-128-1. 2x2GB single rank Samsung 12800E 2Gbit D-die HCK0. (#9) Page2
642.9 FPS. SPD @1333 9-9-9-24-107-1. 2x4GB dual rank Corsair Vengeance ver2.12 Elpida 2Gbit BCBD/BCSE. (#22) Page4
642.3 FPS. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Samsung 10600U 1Gbit E-die HCH9. (#6) Page2
642.3 FPS. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Corsair Dominator ver3.1 Micron 1Gbit D9KPT/KPV (#17) Page4
642.2 FPS. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Elpida 10600U 1Gbit BDSE. (#4) Page2
642.0 FPS. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Samsung 10600U 1Gbit E-die HCH9. (#1) Page1
641.8 FPS. SPD @1333 9-9-9-24-107-1. 2x4GB dual rank Corsair Vengeance ver5.11 Hynix 2Gbit BFR. (#23) Page4
641.6 FPS. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Samsung 10600UE unmatched 1Gbit E-die HCH9. (#12) Page3
641.2 FPS. SPD @1333 9-9-9-24-107-1. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1109. (#21) Page4
640.6 FPS. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Micron 10600U 1Gbit D9KPT. (#16) page4
640.4 FPS. SPD @1333 9-9-9-24-74-1. 2x2GB dual rank Corsair Dominator ver3.5 Micron 1Gbit D9KPT/KPV. (#8) Page2
639.9 FPS. SPD @1333 9-9-9-24-107-1. 2x4GB dual rank Corsair XMS3 ver5.11 Hynix 2Gbit BFR 1202. (#18) Page4
639.5 FPS. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Samsung 10600U unmatched 2Gbit D-die HCH9. (#13) Page3
638.5 FPS. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Micron 10600U 2Gbit D9LGK.
638.5 FPS. SPD @1333 9-9-9-24-174-1. 2x8GB dual rank Adata AX3U1600W8G9-DB 4Gbit. (#15) Page4
638.3 FPS. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Corsair Dominator ver8.14 Nanya 2Gbit. (#2) Page1
638.0 FPS. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Kingston 10600U 2Gbit (PSC?). (#5) page2
636.8 FPS. SPD @1333 9-9-9-24-107-1. 2x2GB single rank Kingston 10600U Elpida 2Gbit BCSE. (#3) Page1
635.7 FPS. SPD @1333 9-9-9-24-174-1. 2x4GB single rank Corsair Value Ram 10600U 4Gbit. (#24) Page4

My current test system for DDR3 is based around an Asus Z87-A motherboard with an overclocked i7 4790K CPU running Windows XP SP3. My CPUs memory controller is pretty weak so the results I get will not be anything that is very hard to achieve. The video card used will be a GTX 780 I need to test. I could have used a GTX 980ti but one of the things I will test is Doom 3 and I feel that if I'm using the optimal video card for that my focus would shift from memory testing to trying to get the best Doom 3 score for the Doom3 thread.

.

The benchmarks used for comparison will be.

SuperPI 32M. For Windows XP I feel that this is the most important benchmark as it has stayed the same since forever which makes it easy to compare results.

Frybench. This is a rendering benchmark that is very hard on the system. If the memory is stable in Frybench it's probably stable in...

Doom III. Do we get any (for us) relevant benefit of the increased memory performance?

.

I will also bench.

Aida64 Cache & Memory Benchmark. Gives an indication of memory latency and throughput.

Aida64 PhotoWorxx. Scales well with memory latency and throughput.

3dmark2001. I thought it would scale with memory speed but after testing 4 memory kits it seems the 4790K has too much cache for memory to matter at all. I will only post outliers if any as the result otherwise ends up being ~127000 points with no obvious correlation to memory speed or latency.

http://www.xtremesystems.org/forums/showthrea … -DDR3-IC-thread

Lets start the testing with this package with 31 "untested" DDR3 modules I bought some months ago.

I saw something moderately interesting in the sellers pictures showing the lot and put in the minimum bid without really expecting to win. Surprisingly I ended up winning the lot and the price ended up being about 80 euro cents per module shipped. I have not opened the package until now, there wasn't any hurry as I would never try to return cheap "untested" stuff.

.

Inside the package we find an assortment of DDR3 modules. It's mostly older 2GB modules but for 0.8 euro each it's hard to complain and for Windows XP 2GB sticks are perfect anyhow.

.

After removing orphaned lone modules we are left with [Edit] 13 [/Edit] "untested" DDR3 kits.

[Edit] I married a Samsung 2GB dual ranked ECC module with 1Gbit E-die with a similar non ECC module. I also traded a single ranked Samsung module with 2Gbit C-die from the scrap lot with a similar module but with D-die from my "orphaned memory" box and by doing so created a second single ranked D-die kit. [/Edit]

.

Lets see what these memory kits can do!

The Serpent Rider wrote on 2023-06-26, 09:31:

http://www.xtremesystems.org/forums/showthrea … -DDR3-IC-thread

xtremesystems.org is my go to forum for information about hardware from 2003 - 2013 or so, it's one of few of the more technical older hardware forums that is still mostly accessible.

i4memory.com was a great resource but sadly long gone. I'm a bit worried that Charles Wirth won't be keeping xtremesystems.org online forever, Vogons feels like a safer bet.

Lets start with these 2x2GB Samsung 10600U modules.

There were actually two of these kits in the untested lot but I have so far only tested one of them.

To keep track of this memory kit in case I end up testing more of the exact same kind lets call them #1.

The modules are dual ranked (double sided) with 16 1Gbit "E-die" K4B1G0846E-HCH9 memory chips per module. The ICs were probably fabricated 2009 but the modules were made week 14 2010.

The highest JEDEC specification the modules officially can handle is 1333 (666ddr) MHz with CL 9, tRCD 9, tRP 9 and tRAS 24 timings at 1.5V. There is not much information about Samsung's 1Gbit "E-die" on the Internet so I had no idea what to expect from these modules.

.

Before trying to optimize the memory settings it's a good idea to see that the memory actually works at stock speed.

DDR3 1333 CL9-9-9-24-1. Aida64 Cache & Memory Benchmark plus PhotoWorxx.

DDR3 1333 CL9-9-9-24-1. Frybench.

DDR3 1333 CL9-9-9-24-1. SuperPI 32M.

.

And finally the benchmark that actually matters...

DDR3 1333 CL9-9-9-24-1. Doom III 1024*768 "Ultra".

.

With the stock results as a baseline lets find out how much performance we can gain.

It seems I now need to do some actual real work so the next post will have to wait until later.

.

(I updated the results for this memory kit, the reason can be found a few posts below)

Lets continue

Now early on there will be some double posting of images showing memory kits as I want a picture of the actual memory kit tested in every post with test results. Later I will probably just do a single post with every memory kit except for those with an XMP-profile (or surprising SPD results) and just add the stock "auto" result to the result lists.

I do find the performance with "auto" settings important as this is how most people run their memory. Two memory kits with similar looking SPD profiles do not necessarily perform the same.

.

Lets see what this (#1) Samsung dual ranked memory kit with "E-die" K4B1G0846E-HCH9 memory chips can do!

.

The "best" memory settings I could get this kit to run at was: 1866MHz CL8-10-9-20-61-1 TRRD4 TFAW16 @1.65V

These settings are a huge improvement over the SPD-profile.

.

DDR3 1866 CL8-10-9-20-61-1 TRRD4 TFAW16. Aida64 Cache & Memory Benchmark plus PhotoWorxx.

.

DDR3 1866 CL8-10-9-20-61-1 TRRD4 TFAW16. Frybench x86.

.

DDR3 1866 CL8-10-9-20-61-1 TRRD4 TFAW16. SuperPI 32M

.

DDR3 1866 CL8-10-9-20-61-1 TRRD4 TFAW16. Doom III 1024*768 "Ultra"

.

All in all I'm very happy with the performance of this memory kit.

A few % extra performance in Doom III might not sound like a lot but considering that it's free extra performance just from changing some memory settings I think it's awesome. The cost/performance value is good with the modules priced at 1.6 euro for the pair. 😁

(I updated the results for this kit, the reason can be found a couple of posts down)

The second memory kit tested is a more fancy Corsair Dominator kit.

Fancy heat spreaders and XMP profiles do sadly not guarantee stellar performance.

.

.

These (#2) Corsair Dominator "DHX Pro" CMP4GX3M2A1600C8 ver8.14 are single rank modules with 2Gbit Nanya ICs.

The highest JEDEC specification the modules officially can handle is 1333 (666ddr) MHz with CL 9, tRCD 9, tRP 9 and tRAS 24 timings at 1.5V. The modules have an XMP profile @1600 MHz with 8-8-8-24-41-2 timings @1.65V. The "2" at the end is the "command rate" value and the 41 is the tRC value. When I write timings in this format ("x-x-x-xx-xx-x) the xx in bold is instead the REF cycle time (tREF) as this is how CPU-Z presents the timings. I do not have much experience with 2Gbit Nanya ICs but they are supposed to be pretty bad.

*On Asus motherboards the user do not have direct control over the tRC value, it is derived from other timings.

.

Lets see how these modules behave using the "auto" (SPD) settings.

DDR3 1333 CL9-9-9-24-107-1. Aida64 Cache & Memory Benchmark plus PhotoWorxx.

DDR3 1333 CL9-9-9-24-107-1. Frybench x86.

DDR3 1333 CL9-9-9-24-107-1. SuperPI 32M.

DDR3 1333 CL9-9-9-24-107-1. Doom III 1024*768 "Ultra".

.

We can see that this Nanya "single rank" memory do not perform as well as Samsung "dual rank" memory in SuperPI and Frybench when running at the 1333MHz auto SPD settings. Doom III do not seem to suffer from the same performance hit. In theory "dual rank" memory should outperform "single rank" memory when everything else is the same.

Edit

Results updated as the testing methodology has improved for less run to run variance.

/Edit

lets see how these Corsair Dominator performs using the XMP profile.

.

.

This version (8.14) of the Corsair Dominator "DHX Pro" is single ranked, is specified to run at 8-8-8-24 timings at 1.65V and as mentioned earlier uses 2Gbit Nanya memory chips. The fact that the modules are single ranked in combination with the fact that the XMP profile relaxes the "command rate" to 2 doesn't bode well for the performance.

When I loaded the XMP profile some spread spectrum and clock filter settings got reset making the FSB drop a tiny bit from just over 100 MHz to just under 100MHz. I didn't notice this at once but unsurprisingly it turns out that it affects the performance a tiny tiny bit. As 99.98MHz is the reference clock speed the system runs at when loading BIOS defaults I decided to not restore the changed settings but instead create a new BIOS-profile with BIOS defaults + the minimum changes needed to run my 4.9 GHZ CPU overclock. I will load this profile when changing memory kits from now on. I will retest the Samsung (#1) kit already tested using the correct settings but the results posted are valid. If anything the results for the Samsung memory will improve a tiny bit as I forgot to remove the network cable when first testing it. Making sure the system isn't connected to the Net is good practice for consistent results. [edit] The results for the Samsung kit (#1) is now updated. [/edit]

.

So how does the XMP profile perform?

XMP 1600 CL8-8-8-24-128-2. Aida64 Cache & Memory Benchmark plus PhotoWorxx.

XMP 1600 CL8-8-8-24-128-2. Frybench x86

XMP 1600 CL8-8-8-24-128-2. SuperPI 32M

XMP 1600 CL8-8-8-24-128-2. Doom III 1024*768 "Ultra"

Not really great is it?

The next question is obviously if manual overclocking can save the Corsair Dominator Nanya (#2) dumpster fire.

.

.

The best I could do with these (#2) Dominator 8.14 single ranked modules with Nanya 2Gbit memory chips was 2133 MHz with 11-11-10-26-124-1 timings. This is actually somewhat better than I had expected.

.

Lets see if the decent overclocking result actually gives us a decent performance increase.

DDR3 2133 CL11-11-10-26-124-1. Aida64 Cache & Memory Benchmark plus PhotoWorxx.

DDR3 2133 CL11-11-10-26-124-1. Frybench x86.

DDR3 2133 CL11-11-10-26-124-1. SuperPI 32M.

DDR3 2133 CL11-11-10-26-124-1. Doom III 1024*768 "Ultra"

.

The results are great! Overclocking turned these modules from pure e-waste to being faster than many 2133 DDR3 kits running their XMP profiles.

The next kit up for testing is this (#3) single ranked Kingston kit with Elpida memory ICs.

Internet suggests that this model of Elpida memory chips often are seen on 1866MHz kits with 9-9-9 timings so it can't be that bad... can it?

.

.

This (#3) Kingston 10600U (1333MHz) kit is a single ranked 2x2GB memory kit with Elpida 2Gbit BCSE chips.

The highest JEDEC specification the modules officially can handle is 1333 (666ddr) MHz with CL 9, tRCD 9, tRP 9 and tRAS 24 timings at 1.5V. While the memory ICs type have the potential to be decent Elpida memory ICs are known to vary a lot in their capabilities even within the exact same memory IC model so there are no guarantees.

.

First the 1333MHz SPD auto settings.

DDR3 1333 CL9-9-9-24-107-1. Aida64 Cache & Memory Benchmark plus PhotoWorxx.

DDR3 1333 CL9-9-9-24-107-1. Frybench x86.

DDR3 1333 CL9-9-9-24-107-1. SuperPI 32M.

DDR3 1333 CL9-9-9-24-107-1. Doom III 1024*768 "Ultra".

.

The performance isn't great but that's how it is with single sided 10600U sticks.

Edit
I retested this kit. The results are just as boring as before.
/Edit

Skyscraper wrote on 2023-06-28, 10:42:

The next kit up for testing is this (#3) single ranked Kingston kit with Elpida memory ICs. […]
Show full quote

The next kit up for testing is this (#3) single ranked Kingston kit with Elpida memory ICs.

Internet suggests that this model of Elpida memory chips often are seen on 1866MHz kits with 9-9-9 timings so it can't be that bad... can it?

Kinston 2x2GB single rank 10600U Elpida 2Gbit BCSE kit #3.jpg

.

This (#3) Kingston 10600U (1333MHz) kit is a single ranked 2x2GB memory kit with Elpida 2Gbit BCSE chips.

The highest JEDEC specification the modules officially can handle is 1333 (666ddr) MHz with CL 9, tRCD 9, tRP 9 and tRAS 24 timings at 1.5V. While the memory ICs type have the potential to be decent Elpida memory ICs are known to vary a lot in their capabilities even within the exact same memory IC model so there are no guarantees.

.

First the 1333MHz SPD auto settings.

DDR3 1333 CL9-9-9-24-107-1. Aida64 Cache & Memory Benchmark plus PhotoWorxx.

i7 4790K @4.9 DDR3 @1333 CL9-9-9-24-1 1.5V Aida64 Cache & Memory Benchmark plus Photoworxx Elpida 2x2GB BCSE single rank from scrap lot.JPG

DDR3 1333 CL9-9-9-24-107-1. Frybench x86.

i7 4790K @4.9 DDR3 @1333 CL9-9-9-24-1 1.5V Frybench x86 Elpida 2x2GB BCSE single rank from scrap lot.JPG

DDR3 1333 CL9-9-9-24-107-1. SuperPI 32M.

i7 4790K @4.9 DDR3 @1333 CL9-9-9-24-1 1.5V SuperPI 32M Elpida 2x2GB BCSE single rank from scrap lot.JPG

DDR3 1333 CL9-9-9-24-107-1. Doom III 1024*768 "Ultra".

i7 4790K @4.9 DDR3 @1333 CL9-9-9-24-1 1.5V GTX 780 Doom3 1024Ultra Elpida 2x2GB BCSE single rank from scrap lot.JPG

.

It's not surprising that this memory kit using its 1333MHz SPD profile slots in right behind the other tested single ranked kit at the bottom of the lists. With this kit even the Doom III result is remarkably bad.

Kingston is third-party and is too relaxed timings too, and used open-market chips and many times blank part number or crossed out chips. OEM that sells brands like Micron/Crucial, Samsung, Hynix have much better timings and highest quality and is genuine also chips were made by themselves. The one that sells that are not OEM are called third party had to buy chips of any variable quality levels or low quality all they can get and assemble them. Third party: e.g.: Mushkin, Kingston, Patriot, Corsair, G. Skil. so on, you know what I mean.

The quality third-party I can confirm are Kingston and Mushkin. Had least failure rate and high success rate in use. The ones I had problems the most were other third-party eg: compatibility, incorrectly detected also failure rate. One time I had pair of corsair modules not detected correctly despite it is stated DDR3-1600 on the packaging, actually ran at 1333 on many boards, including HP computer. I ended up reprogramming the SPD from a compatible module and sold them.

Cheers,

pentiumspeed wrote on 2023-06-28, 17:34:

Kingston is third-party and is too relaxed timings too, and used open-market chips and many times blank part number or crossed out chips. OEM that sells brands like Micron/Crucial, Samsung, Hynix have much better timings and highest quality and is genuine also chips were made by themselves. The one that sells that are not OEM are called third party had to buy chips of any variable quality levels or low quality all they can get and assemble them. Third party: e.g.: Mushkin, Kingston, Patriot, Corsair, G. Skil. so on, you know what I mean.

The quality third-party I can confirm are Kingston and Mushkin. Had least failure rate and high success rate in use. The ones I had problems the most were other third-party eg: compatibility, incorrectly detected also failure rate. One time I had pair of corsair modules not detected correctly despite it is stated DDR3-1600 on the packaging, actually ran at 1333 on many boards, including HP computer. I ended up reprogramming the SPD from a compatible module and sold them.

Cheers,

I will have to add a bit of a counter argument here even if I think we mostly agree.

The memory chip makers (Samsung, Hynix, Micron and the others) makes different models of memory chips (and revisions of those models) with different capability's. Not every memory chip of the same model and revision come out of the fabrication process exactly the same, just like CPUs they vary in what speeds they can handle.

The memory chip makers can choose if they test all the chips and sort them after capability (a.k.a binning) or just sell them unsorted with specifications even the worst can handle. I think the trend is that the memory chip manufacturers do less and less testing themselves and let the customers (Kingston, Corsair, G.Skill and others) handle the "binning process".

So what do the memory makers use for their "in house" memory? I think in most cases they use "unbinned" memory or when it's binned they use the lower/mid (but not bottom) bins and sell the best bins for a lot of money to Kingston, Corsair and the others. This is one of the reasons you always see Samsung, Hynix and Micron using the slowest JEDEC timings (1333 9-9-9 or 1600 11-11-11 and so on) for their memory. The exception being if the memory chip maker has it's own in house "performance memory" brand like Microns Crucial, then they of course get good memory bins for their higher end memory modules.

.

So why then is vanilla Samsung or Hynix "in house" memory often decent when it comes to overclocking while Kingston value ram using the same type of memory chips often is total garbage? The reason is (probably) that Kingston buys a huge lot of unsorted memory ICs then tests and sorts them. The good memory chips ends up in their premium memory and the not so good ones ends up in kits like the one I just tested.

To sum it up. The memory makers own "in house" "vanilla" non fancy memory modules have the potential to be decent as they are (probably) mostly made from unbinned (unsorted) memory ICs. The third party low end "value" offerings often suck because they are (probably) mostly made from memory chips they themselves have tested and found not good enough for their more expensive premium fancy memory.

This is of course simplifications and generalizations. Some memory IC models/revisions are known to be almost generally good and others more or less generally bad and for these the "binning" obviously matters less. On the other hand I think it's more likely you will find the generally bad chips on Kingston value ram compared to the generally good ones.

When it comes to failure rates I think that has more to do with quality control than not having access to good memory chips. Another thing is when a chip manufacturer have had a quality issue that didn't get caught fast enough like Elpida did in 2008 with some of their 1Gbit DDR3 ICs.

We will see just how bad these third party low end value offerings can be in my next post.

Lets see just how bad a low end 1333MHz kit (#3) from Kingston made with memory chips that are probably "binning rejects" can be.

With "binning rejects" I mean that these could very well be Elpida BCSE memory chips Kingston bought in bulk and tested but that didn't meet the standards for their faster more premium memory. Supposedly G.Skill used Elpida BCSE on modules rated up to 2000MHz with 9-9-9 timings so this chip revision do at least have the potential to be decent.

Before I tried to overclock this kit I fully expected that they would overclock better than the two kits previously tested. Not because I thought that these chips were a good bin of Elpida BCSE but if a good bin can handle 2000MHz with 9-9-9 timings then surly even a low end bin must be able to do 1866 MHz with the same timings?

.

.

To my surprise the best settings I could get this memory kit (#3) to run at was 1600MHz with 9-8-8-22-94-1 timings at 1.65V. This might have been alright for some low end chip fabricated in the first half of 2008 but it's no wonder these Elpida BCSE ICs made in 2011 ended up in this bottom of the barrel Kingston kit.

.

Just how bad is it?

DDR3 1600 CL9-8-8-22-94-1. Aida64 Cache & Memory Benchmark plus PhotoWorxx.

DDR3 1600 CL9-8-8-22-94-1. Frybench x86.

DDR3 1600 CL9-8-8-22-94-1. SuperPI 32M.

DDR3 1600 CL9-8-8-22-94-1. Doom III 1024*768 "Ultra".

.

Well pretty horrible but pretty much what could be expected.

Edit
Retested: Still sucks.
/Edit

The memory chip makers can choose if they test all the chips and sort them after capability (a.k.a binning) or just sell them unsorted with specifications even the worst can handle. I think the trend is that the memory chip manufacturers do less and less testing themselves and let the customers (Kingston, Corsair, G.Skill and others) handle the "binning process".

I think there is a 2-step binning. Worst chips, are sold for less reputable brands with less money to spare. Chips which have less headroom or can't work stable in more harsh conditions. So at the very least, buying from manufacturers can guarantee that chips met certain standard, which is always above listed specs.

On another end of the spectrum, we have factory top tier bins, like legendary Samsung TCCD, which was practically sold only as overclocked memory.

So why then is vanilla Samsung or Hynix "in house" memory often decent when it comes to overclocking while Kingston value ram using the same type of memory chips often is total garbage? The reason is (probably) that Kingston buys a huge lot of unsorted memory ICs then tests and sorts them. The good memory chips ends up in their premium memory and the not so good ones ends up in kits like the one I just tested.

I doubt that Kingston bothers (before RAM biz split anyway) to bin value RAM on large scale and just resell what manufacturers deemed as meeting requirements. But Corsair, Gskill and other more enthusiast focused brands probably have bins inside bins inside bins.

P.S.
Maybe add 3DMark2001? That benchmark is sensitive to RAM and timings.

The Serpent Rider wrote on 2023-06-29, 11:25:

I think there is a 2-step binning. Worst chips, are sold for less reputable brands with less money to spare. Chips which have le […]
Show full quote

The memory chip makers can choose if they test all the chips and sort them after capability (a.k.a binning) or just sell them unsorted with specifications even the worst can handle. I think the trend is that the memory chip manufacturers do less and less testing themselves and let the customers (Kingston, Corsair, G.Skill and others) handle the "binning process".

I think there is a 2-step binning. Worst chips, are sold for less reputable brands with less money to spare. Chips which have less headroom or can't work stable in more harsh conditions. So at the very least, buying from manufacturers can guarantee that chips met certain standard, which is always above listed specs.

On another end of the spectrum, we have factory top tier bins, like legendary Samsung TCCD, which was practically sold only as overclocked memory.

So why then is vanilla Samsung or Hynix "in house" memory often decent when it comes to overclocking while Kingston value ram using the same type of memory chips often is total garbage? The reason is (probably) that Kingston buys a huge lot of unsorted memory ICs then tests and sorts them. The good memory chips ends up in their premium memory and the not so good ones ends up in kits like the one I just tested.

I doubt that Kingston bothers (before RAM biz split anyway) to bin value RAM on large scale and just resell what manufacturers deemed as meeting requirements. But Corsair, Gskill and other more enthusiast focused brands probably have bins inside bins inside bins.

P.S.
Maybe add 3DMark2001? That benchmark is sensitive to RAM and timings.

I don't think Kingston spent a minute specifically on binning value ram either but the binning rejects definitely ended up in those product lines. 😁

I think the binning process was/is different depending on how much the capabilities of the memory chips vary within chip revisions and even batches. It probably also depends on general memory prices and how much market demand there is for more expensive binned memory.

There is not much use binning if there is zero market demand for DD3 memory faster than 1866 MHz CL11 and all memory chips you make can do that speed.

.

I had planned to include 3dmark2001 but sadly there is zero scaling with memory in 3dmark2001 using this Haswell system.

If the scaling had been bad I would still have included the results but there is really no scaling and the up to 1% variance in 3dmark2001 points between different memory kits and different runs with the same memory kit show no correlation with memory throughput and latency.

The 3 tests I use for comparison scales decently with memory speed and also do usually not show any notable variance between runs on this system when using the exact same settings and exact same benching method.

Why do I not show 3dmark2001 results even if I have them?

This is why.

First 3 results using 3 memory kits running their 1333MHz auto SPD timings. One kit is dual ranked, two are single ranked.

127000 points! Awesome!

.

Now lets look at a result at overclocked 1866 MHz settings.

Again ~127000 points!

.

Finally lets look at a kit running at overclocked 2133 MHz settings.

Once again 127000 points! 😁

.

The decrease in measured latency going from the non overclocked to overclocked settings above is about 20% and the increase in memory throughput 50%. It's hard to conclude anything other than that 3dmark2001 doesn't scale with memory speed or latency on this overclocked Haswell i7 4790K system.

3dmark2001 is like The Serpent Rider mentions known to scale with faster memory so why doesn't it? Your guess is as good as mine, perhaps the 8MB cache clocked at 4500MHz nullifies the improvement faster memory usually brings in this old 3d benchmark?

With in mind, there are grades of silicon electronic parts being graded at manufacture. Top grade goes to military and for example memory chips is graded too. A grade is used internally to manufacture their own memory modules, and also A grade goes to long lead orders by thousands. Anything of B grade or lesser, is sold on open market, (that's what Kingston gets). With exception of Samsung and Crucial Ballistrix for gaming and overclocking (Micron manufactures and their normal modules them with A grade stuff. All the other gaming memory modules and regular third party memory modules are grade B and with cheapest ones grade C.

Cheers,