Overclock.net › Forums › Components › Memory › Memory theory - help needed
New Posts  All Forums:Forum Nav:

# Memory theory - help needed

I've spent quite a bit of time over the past couple of weeks trying to learn about overclocking my system since I finally have time to mess with it. I've never overclocked anything before this which is why I want to understand as much of the underlying theory, principles and basics as possible.

So far I've tried to figure out my max FSB for the CPU at stock and max voltages and I will get back to work on the CPU after I figure out the memory which is where I am now. I'm trying to understand more specifically how timings vs. speed actually play out. For example, I can tighten my timings down to their lowest values except for TWTR (2, won't boot at 1) and get a max DDR2 clock of 306Mhz (612Mhz effective) at 1.9v or 343Mhz clock (686Mhz effective) at 2.2v. At the other end I can set the usual timing specs to 5-5-5-15-1T and the rest to auto and get a 458Mhz clock (916Mhz effective) at 1.9v or 506Mhz clock (1012Mhz effective) at 2.2v.

From reading at thetechrepository site they explained that you multiply the CAS value by the cycle time of your frequency to figure out the actual latency. To me this says that the latency is more representative/accurate of the time to do a single instruction. For example using my memory,
DDR2 800Mhz (400Mhz core) - Theoretical max bandwidth is 800Mhz x 8bytes/instruction = 6.4GB/s
DDR2 800Mhz (400Mhz core) @ CAS 5, latency = 2.5ns x 5 = 12.5ns/clock cycle
DDR2 800Mhz (400Mhz core) @ CAS 5 bandwidth = (1second / 12.5ns) x 16bytes/clock cycle = 1280MB/s = 1.28GB/s

There are two things that I want to point out to see if I am correct on a side note, the first being that DDR memory has a bandwidth of 16bytes per clock cycle. The second being instead of trying to figure out latency in nanoseconds and then figuring out what frequency that corresponds to it is easier to divide the core DDR speed by the CAS value to find what frequency is obtained when factoring in CAS and/or divide the max rated bandwidth by the CAS value.

I've read that basically if you can maintain similar latencies that faster rated ram will provide more bandwidth and I do not understand how. To me DDR 800 @ CAS 4 should be the same as DDR 1600 @ CAS 8 since their latencies are the same yet they both transmit 8 bytes twice per clock cycle. I think I understand that the 1600 works faster, twice as fast but isn't that negated since the CAS delay is twice as long? It all seems to make sense yet from all of the stuff I've read I feel like I am putting the pieces together wrong or I am missing something all together.
Edited by meltmanbob - 9/15/10 at 6:59am
 My System (13 items)
CPUMotherboardGraphicsRAM
7750 B.E. (overclocking) (lapped) Asrock 780GXE/128m (n.b.+s.b. lapped) onboard 3200 2gb DDR2 800 OCZ Gold
Hard DriveOptical DriveOSMonitor
ADATA S599 128gb, Maxtor SataII 320gb LiteOn IDE DVD burner Windows 7 home 64bit Chimei 19" CMV939D
KeyboardPowerCaseMouse
Logitech something CoolerMax V-600 Rosewill Challenger Wal-mart brand wireless
generic
 My System (13 items)
CPUMotherboardGraphicsRAM
7750 B.E. (overclocking) (lapped) Asrock 780GXE/128m (n.b.+s.b. lapped) onboard 3200 2gb DDR2 800 OCZ Gold
Hard DriveOptical DriveOSMonitor
ADATA S599 128gb, Maxtor SataII 320gb LiteOn IDE DVD burner Windows 7 home 64bit Chimei 19" CMV939D
KeyboardPowerCaseMouse
Logitech something CoolerMax V-600 Rosewill Challenger Wal-mart brand wireless
generic
I am not an engineer and I do not have a complete grasp of the the nuances of how all the timings interact, but the jist of it is:

You can't directly convert timings to bandwidth. Core timings indicate the delay in getting the first data, they do not indicate a limit in how much can be transfered.

DDR-800 CAS 4 and DDR-1600 CAS 8 will have very similar latencies, but once the data is accessed, the later will transfer it almost twice as fast.
 Primary (15 items) Secondary (13 items) Vishera Testbed (11 items)
CPUMotherboardGraphicsRAM
5820K @ 4.3GHz, 1.225v Gigabyte X99 SOC Champion (F4m) 2x Sapphire R9 290X Tri-X OC New Edition (10036... 4x4GiB Crucial @ 2667, 12-12-12-28-T1, 1.35v
Hard DriveHard DriveHard DriveCooling
Plextor M6e 128GB (fw 1.05) M.2 (PCI-E 2.0 2x) 2x Crucial M4 256GB 4x WD Scorpio Black 500GB Cooler Master Nepton 280L
OSMonitorKeyboardPower
Windows 7 Professional x64 SP1 BenQ BL3200PT Filco Majestouch Tenkeyless (MX Brown) Corsair RM1000x
CaseMouseAudio
Fractal Design Define R4 CM Storm Spawn Realtek ALC1150 + M-Audio AV40
CPUMotherboardGraphicsRAM
X5670 Gigabyte X58A-UD5 r2.0 w/FF3mod10 BIOS Reference R9 290X w/Stilt's MLU 1000e / 1375m E... 2x Samsung MV-3V4G3D/US @ 1600MT/s 7-8-8-19-T1,...
RAMHard DriveHard DriveHard Drive
1x Crucial BLT4G3D1608ET3LX0 @ 1600MT/s 7-8-8-1... OCZ (Toshiba) Trion 150 120GB Hitachi Deskstar 7k1000.C 1TB 2x Seagate 7200.10 RAID 0
CoolingOSPowerCase
Prolimatech Genesis + 2x140mm Cougar 1200rpm Windows Server 2008 R2 Antec TP-750 Antec P182
Audio
ASUS Xonar DS
CPUMotherboardGraphicsRAM
FX-9590 @ 5GHz, 1.55v ASUS Sabertooth 990FX R2.0 Reference NVIDIA GTX 780 2x8GiB G.Skill DDR3-1866 10-11-10-30-T1, 1.5v
Hard DriveHard DriveCoolingOS
Crucial M500 480GB 2x Samsung Spinpoint F1 1TB XSPC RX360 + X20 750 + Raystorm Windows 7 Pro SP1 x64
MonitorPowerCase
Dell S2740L Seasonic SS-860XP2 Coolermaster HAF-932
 Primary (15 items) Secondary (13 items) Vishera Testbed (11 items)
CPUMotherboardGraphicsRAM
5820K @ 4.3GHz, 1.225v Gigabyte X99 SOC Champion (F4m) 2x Sapphire R9 290X Tri-X OC New Edition (10036... 4x4GiB Crucial @ 2667, 12-12-12-28-T1, 1.35v
Hard DriveHard DriveHard DriveCooling
Plextor M6e 128GB (fw 1.05) M.2 (PCI-E 2.0 2x) 2x Crucial M4 256GB 4x WD Scorpio Black 500GB Cooler Master Nepton 280L
OSMonitorKeyboardPower
Windows 7 Professional x64 SP1 BenQ BL3200PT Filco Majestouch Tenkeyless (MX Brown) Corsair RM1000x
CaseMouseAudio
Fractal Design Define R4 CM Storm Spawn Realtek ALC1150 + M-Audio AV40
CPUMotherboardGraphicsRAM
X5670 Gigabyte X58A-UD5 r2.0 w/FF3mod10 BIOS Reference R9 290X w/Stilt's MLU 1000e / 1375m E... 2x Samsung MV-3V4G3D/US @ 1600MT/s 7-8-8-19-T1,...
RAMHard DriveHard DriveHard Drive
1x Crucial BLT4G3D1608ET3LX0 @ 1600MT/s 7-8-8-1... OCZ (Toshiba) Trion 150 120GB Hitachi Deskstar 7k1000.C 1TB 2x Seagate 7200.10 RAID 0
CoolingOSPowerCase
Prolimatech Genesis + 2x140mm Cougar 1200rpm Windows Server 2008 R2 Antec TP-750 Antec P182
Audio
ASUS Xonar DS
CPUMotherboardGraphicsRAM
FX-9590 @ 5GHz, 1.55v ASUS Sabertooth 990FX R2.0 Reference NVIDIA GTX 780 2x8GiB G.Skill DDR3-1866 10-11-10-30-T1, 1.5v
Hard DriveHard DriveCoolingOS
Crucial M500 480GB 2x Samsung Spinpoint F1 1TB XSPC RX360 + X20 750 + Raystorm Windows 7 Pro SP1 x64
MonitorPowerCase
Dell S2740L Seasonic SS-860XP2 Coolermaster HAF-932
That's kind of what I was thinking yet even though I can't quite put together how all of the timings relate to each other in terms of order of operations, it seems that all of the timings specify a specific timeline in the memory access scheme. So basically for that to work there has to be a window that isn't described by the memory timings where the speed is only limited by the frequency it's being run at and not a timing parameter.

From what I've read tRAS, tRP, tRCD, and CAS more or less describe the entire process which would mean that each step is bound by timings. I guess the next step is to try and figure out exactly what the order of operations are with regard to the individual timings. I think I understand what you are suggesting though and the best way I can think of it is like driving a fast car vs a slow car through a city. There will be some common "latencies" such as both waiting the same amount of time at the same red light yet the faster car could reasonably reach the next light faster and already have waited the pre-determined time at that red light by the time the slow car even gets there.

On a slight side note towards actual application, at what point would you consider a good trade off in terms of Mhz to base timings? For example I can hit 458Mhz with 5-5-5-15-1T, 440Mhz @ 4-4-4-12-1T and 306Mhz @ 3-3-3-8-1T and I'm trying to figure out when do I compromise one for the other.
 My System (13 items)
CPUMotherboardGraphicsRAM
7750 B.E. (overclocking) (lapped) Asrock 780GXE/128m (n.b.+s.b. lapped) onboard 3200 2gb DDR2 800 OCZ Gold
Hard DriveOptical DriveOSMonitor
ADATA S599 128gb, Maxtor SataII 320gb LiteOn IDE DVD burner Windows 7 home 64bit Chimei 19" CMV939D
KeyboardPowerCaseMouse
Logitech something CoolerMax V-600 Rosewill Challenger Wal-mart brand wireless
generic
 My System (13 items)
CPUMotherboardGraphicsRAM
7750 B.E. (overclocking) (lapped) Asrock 780GXE/128m (n.b.+s.b. lapped) onboard 3200 2gb DDR2 800 OCZ Gold
Hard DriveOptical DriveOSMonitor
ADATA S599 128gb, Maxtor SataII 320gb LiteOn IDE DVD burner Windows 7 home 64bit Chimei 19" CMV939D
KeyboardPowerCaseMouse
Logitech something CoolerMax V-600 Rosewill Challenger Wal-mart brand wireless
generic
I'm pretty sure out of those 440Mhz at 4-4-4-12 would be best
Thanks for the suggestion! I think I've got a decent set of programs to start using to try and see how the different settings perform. For the memory I have Memtest, MaxxMem, Everest Home, SiSoft Sandra and Intel Burn Test, Orthos, OCCT, Prime95 and HyperPi. I skipped on PCMark or 3DMark since they cost money.

On a slightly different note I have been trying to isolate each major part for overclocking to find their individual limits and I ran into some odd things yesterday. When isolating the north bridge I could get it to 2.4Ghz on stock volts, 2.6Ghz @ 1.425v and could not hit 2.8Ghz even with 1.5v. I moved on to some other parts of the system and when I came back to try and combine the CPU overclock and the north bridge overclock I could not get the north bridge to 2.6Ghz even at 1.5v. All of that overclocking was simply done with the multiplier and a base clock of 200Mhz just to get a rough idea of where their limits were. I was however able to get 2.604Ghz on the north bridge but I had to raise the base clock and keep the multiplier at 12, any thoughts as to why it would not work at the 13x multiplier anymore? I couldn't reproduce my earlier results even by isolating just the north bridge.

Also which forum would be best to post and get advice towards an entire system overclock since I'm working on the CPU, RAM, north bridge, HT, and onboard graphics? Thanks again!
 My System (13 items)
CPUMotherboardGraphicsRAM
7750 B.E. (overclocking) (lapped) Asrock 780GXE/128m (n.b.+s.b. lapped) onboard 3200 2gb DDR2 800 OCZ Gold
Hard DriveOptical DriveOSMonitor
ADATA S599 128gb, Maxtor SataII 320gb LiteOn IDE DVD burner Windows 7 home 64bit Chimei 19" CMV939D
KeyboardPowerCaseMouse
Logitech something CoolerMax V-600 Rosewill Challenger Wal-mart brand wireless
generic
 My System (13 items)
CPUMotherboardGraphicsRAM
7750 B.E. (overclocking) (lapped) Asrock 780GXE/128m (n.b.+s.b. lapped) onboard 3200 2gb DDR2 800 OCZ Gold
Hard DriveOptical DriveOSMonitor
ADATA S599 128gb, Maxtor SataII 320gb LiteOn IDE DVD burner Windows 7 home 64bit Chimei 19" CMV939D
KeyboardPowerCaseMouse
Logitech something CoolerMax V-600 Rosewill Challenger Wal-mart brand wireless