There are no rules like that about timings. You can set tRAS lower or higher without issues like 10-12-12-21 and some standard settings which you can find in XMP profiles are 10-12-12-31, 11-13-13-35, 9-9-9-24 etc. ... all depends what IC.
In DDR4 you can set even something like 15-15-15-15, 10-12-15-14 or 13-11-15-14. Simply don't make some random articles in the web limit you from trying various settings, even those that look weird.
There are no perfect memory OC guides just because there is too many IC around. Even though some act in similar way, you can have problems to run it in a different kit from the same series. Look at motherboard profiles made by manufacturers, barely any of them works but you see some pattern in which you can try tweaking your memory timings.
Started playing with the new setup last night. Running a 4960x (under water atm) on a Rampage IV Black Edition. Gave the 4x4 2400 tridentx a try....with everything pretty much on auto/default and xmp, it runs no problem at 2400. Using xmp profile 1, Aida read speed was just over 60K. Results were a bit faster with profile 2 (just under 61K).
How can I get the memory running at faster than 2400? Tried a couple of the Samsung profiles and was able to post at 2666 but it more or less instantly bsod'd once in windows. Set vccsa to 1.2 and vtt to 1.15 but it didn't help at all. I see others have made it to 2666-2676 for spi32. I know most of the similar builds top out around 2400 but would like to at least be able to test at higher speeds. Any ideas?
Will have a couple hours tonight...guess I'll see how the 4x 2200 pi's will do. Based on shamino's testing, figuring these will do better than tridentx. Figure I'll also try my other quad kit (binned crucial ballistix but since they're only rated at 1600, am not expecting much from those on this platform).
Just out of curiousity....would it be possible that the best sticks in quad channel wouldn't be as good as another set of sticks when running dual or single channel? For example, let's say the 2200 pi's score better than the 2400 tridentx in quad mode. Would it ever be possible that the tridentx in dual channel would be faster than the pi's in dual channel?
Forget about profiles on this board. I have no idea why ASUS is using Shamino's or other guys profiles on cherry picked memory kits when regular users can't set so tight timings.
I already said that you should use 125 bclk strap. I just try to make shorter posts lately so less stuff is being missed by readers
strap 100 = max x24 memory ratio = 2400 memory clock
strap 125/167 = higher ratios
Spent some time with the tridentx before switching to the 2300 pi's. What a difference! The pi's work with the psc presets without issue and read speeds are much higher than the tridentx. All the straps are working fine. Am sure there's more that can be fine tuned but here's the best result of the evening:
PS- Can't find the thread with everyone's aida results? iirc it has read speed and latency tables.
Have some time on my hands (ss and cascade are down until I can get them repaired) so figure it's a good time for more testing.
Wondering if, for ddr3, I can use the platform that can run ram the fastest (my z97/g3258 setup can get ram over 3000MHz) to test everything from 1600 upward. Trying to avoid having to test the sticks on each separate platform so thinking to take the following approach:
- Test each set of sticks at 1600 to identify the sticks that give the best Aida read results
- Repeat for 1800, 1900, 2000 (to cover options for 775 and other 'slower' setups)
- Repeat for 2133, 2200, 2400 and 2666
- Repeat at max possible speed for each set of sticks
Then...for any given platform, find max possible speed using sticks that will go faster than is possible given chipset/cpu and use the sticks that gave the best result at that particular speed. If they don't work, try the 2nd best, etc until identifying the best sticks for that chipset/cpu.
Whatcha think? Will that work or will I really need to test the sticks on every platform to figure out the best?
You can test it on one platform but some older platforms have limited memory controllers regarding max clock or performance so depends from platform you may see the same performance scalling or not.
Don't waste time on every setting.
Rules are simple:
- double sided modules > single sided
- lower main timings > higher timings
- lower capacity > higher capacity ( generally 2GB > 4GB > 8GB if you look at max OC, exception can be 1GB modules which are usually worse than 2GB )
- 775/1156/1366 = better are Elpida BBSE or Hyper/Micron/PSC at tight timings
- anything newer like 1155/1150/2011 = better are Elpida BBSE/PSC/Samsung or double sided higher binned Hynix
The main idea is that I'm trying to figure out which sticks to keep and which to sell....I definitely have a lot more ddr3 than anyone really needs but I want to make sure to keep the best sticks for any ddr3 platform I'll ever run. So...trying to figure out the best, easiest and quickest way to identify the sell sticks vs. the keepers.
As an example of what I was thinking....let's think about 775 with the REX. The max ram speed I've ever managed on that platform is in the 2000 to 2100MHz range (with 2 sticks) but some 775's I've only been able to get to 1700-1800 so (for 775 use), want to figure out my best pair(s) of sticks at 1700-1800 and 2000-2100. My thought was to use the z97 setup and test all the sticks at those speeds (maybe shoot for 1750 and 2050) and find the ones that give the best aida read speed. The sticks with the highest read speeds at those 2 settings (1750 and 2050) would be the keepers.
Once I have those identified, I'd think about 2011 (am pretty much ditching 1155 at this point) which I've had running in the 2133-2500 range and (again, using z97 which seems to run anything and everything) test all sticks at a couple speeds (maybe 2200 and 2450) to figure out the keepers for 2011. Finally...test for 1150 at maybe 2500-2750 or so to identify those keeper sticks.
After testing all sticks on z97 and figuring out the 775, 2011 and 1150 keepers, sell the rest of the sticks.
Does that make sense? Maybe with your guidance, could reduce the number of sticks to be tested? If I post pics of all the sticks, think you might be able to identify ones that should be ruled out and not even tested?
Memory on 1155/1150 and 2011 is overclocking almost the same with some exceptions. So you can check all kits on 1155 or 2011 as 1150 CPUs have better IMC. Still 1155/2011 will run at least up to 2600-2800.
2011/1150 don't like hypers, qimonda or older micron based IC.
You can also make full list of RAM and I bet that some are keepers without tests.
From DDR3 I'm only keeping:
2 kits - 2x4GB TridentX 2400 CL9 ( Samsung ) - right now in my gaming PC
1 kit - 2x4GB TridentX 2600 CL10 ( Samsung )
1 kit - 2x2GB RipjawsX 2133 CL9 ( PSC )
2 kits - 2x2GB Dominator GT 2000 C8 ( PSC )
1 kit - 3x2GB ADATA Gaming 2000 C9 ( PSC )
1 kit - 2x4GB HyperX 2666 C11 ( Hynix MFR - can make 3700-3800 on air ) - right now in my NAS
1 kit - 2x4GB ADATA 2933 C12 ( Hynix ) - similar to HyperX but currently trying to sell as market price is higher
OK...here's the ddr3 I found around...still missing a few but these are most of them. Would definitely welcome your (and anyone else's) opinions on
- which are definite keepers
- which are definite 'to be sold'
- which should be tested further before deciding
Am going to use the sticks for 775, 2011, fm2 and 1150. May also still play with x58, 1155 and some of the amd ddr3 chipsets/sockets.
For any sticks that are 'to be sold'...any ideas on value/pricing would be very welcomed.
1600-GSkill Ripjaws http://www.newegg.com/Product/Product.aspx?Item=N82E16820231344
Elpida BBSE
f3-12800cl7t-3gbrh, 3x1GB, 7-7-7-24
Only tested to confirm everything runs properly at default. Prior owner ran them from 1000c6 (1.73v) up to 1200c9 (1.68v)
1600-Crucial Ballistix http://www.newegg.com/Product/Product.aspx?Item=N82E16820148146
part# bl12864ba1608.8sfb, batch# cl1118b.m3, 2x1GB, (assuming 8-8-8-24 as for 05-1600-crucial2)
Have had these running 2000 at stock timings and volts. 2-4 sticks of these crucials are supposed to be binned and cherry-picked.
1600-Crucial Ballistix http://www.newegg.com/Product/Product.aspx?Item=N82E16820148146
part# bl12864ba1608.8sfb, batch# cl1118v.c8, 2x1GB, 8-8-8-24
Have had these running 2000 at stock timings and volts. 2-4 sticks of these crucials are supposed to be binned and cherry-picked.
1600-Crucial Ballistix http://www.newegg.com/Product/Product.aspx?Item=N82E16820148146
part# bl12864ba1608.8sfb, batch# cl1119f.xz, 2x1GB, 8-8-8-24
Have had these running 2000 at stock timings and volts. 2-4 sticks of these crucials are supposed to be binned and cherry-picked.
1600-Crucial Ballistix
D9GTR
batch# cl1114j.8z, 2x1GB, 8-8-8
Haven't tried these that I can recall. Prior owner ran them at 7-7-7@1800 or 7-7-7@2000. Can run up to 7-7-7@2100 w/ 2.05v. Ran them at 8-8-8@2000 in a Z77 board with 1.65v.
2000-Geil Evo2 http://www.geil.com.tw/German/reviews/show/id/212
PSC
get34gb2000c6dc, 2x2GB, 6-9-6-24
These sticks won't run on a lot of setups. On z77, have had these running 6-9-6-24 1T @1910, 1.65v, 5-8-6-24 1T with 1.75v. Prior owner ran them 2600, 8-12-8-28 1T @ 1.87v and 2667, same timings @ 1.92v on z87.
2133-GSkill Perfect Storm http://www.newegg.com/Product/Product.aspx?Item=N82E16820231269
Elpida BBBG/BBSE
f3-17066cl9t-6gpbps, 3x2GB, 9-9-9-24
Prior owner tested the kit in various combinations for 1105 7-9-7-24 1,62v, 1143 7-9-7-24 1,66v, 1200 7-10-7-24 1,75v and also for 1000 6-8-6-24. The kit boots and does 1m 1300 8-11-8-22 at 1,89v on M5F, but fails 32m.
2400-GSkill TridentX http://www.newegg.com/Product/Product.aspx?Item=N82E16820231638
Samsung
f3-2400c9d-8gtxd, 4x4GB, 9-11-11-31
On z97, have had these running as high as 2933, 11-13-13-38 @ 1.87v but aida read speed was much faster at 2400, 9-11-11-31 with stock voltage. Ran 2786, 10-12-12-30 2T @ 1.87v on fm2.
2400-GSkill PI's http://www.newegg.com/Product/Product.aspx?Item=N82E16820231338
PSC
f3-19200cl9d-4gbpis, 2x2GB, 9-11-9-28
Need to play with these more...they were giving me some trouble running stock speed and timings at stock voltage on z97. Prior owner ran them 2667, 8-12-8-28 1N @ 1.86v on z97.
2666-GSkill TridentX http://www.newegg.com/Product/Product.aspx?Item=N82E16820231595
Samsung?
f3-2666c11d-8gtxd, 2x4GB, 11-13-13-35
Played with these a bit on z97. Aida read speeds were higher with the 2400 tridents but got them to run 2933 without issue. Wasn't very impressed but maybe worth more testing?
2933-GSkill TridentX http://www.newegg.com/Product/Product.aspx?Item=N82E16820231683
SAMSUNG?
f3-2933c12d-8gtxdg, 2x4GB, 12-14-14-35
Played with these a bit on z97. Aida read speeds were higher with the 2400 tridents but got them to run over 3000 without issue. Wasn't very impressed but maybe worth more testing?
BBSE are good in general but hard to say in this case. I think this kit could be good only for older platforms. Test how high can it go. Like 2000 8-8-8/9-9-9 1.65V is a good start.
Microns, can be good almost only for 775. Can play with them on haswell but for that they will require high voltages and better cooling. Personally I would sell them but I have no 775 DDR3 platform and I'm not sure if you find anything much better for these older boards
1600-Crucial Ballistix
D9GTR
batch# cl1114j.8z, 2x1GB, 8-8-8
Haven't tried these that I can recall. Prior owner ran them at 7-7-7@1800 or 7-7-7@2000. Can run up to 7-7-7@2100 w/ 2.05v. Ran them at 8-8-8@2000 in a Z77 board with 1.65v. View attachment 161502
The same story as with Cellshocks. These sticks are probably really cheap so I'm not sure if are worth to be sold. On the other hand I doubt we will see any new DDR3 platform so they can be useless for you in some time.
You can keep them to play around on some platforms as I see they're overclocking pretty good and can make tight timings @2000+. Probably can push them to make something like CL6/7 2200+ but on 2.2V+ and better cooling.
Microns or something similar like Qimonda. I wouldn't count on much more than ~1800-2000 at tight timings.
1800-Corsair Dominator
cm3x1024-1800c7d, ver 3.1, 2x1GB, 7-7-7-20
Haven't played with these in years. Don't recall how they ran. View attachment 161503
Probably better kit than Microns for 775 platform. I would test if you can run them higher or at least the same as above 2x1GB/3x1GB Microns. If yes then sell Microns and heep this kit. Should be also good for Z77 and maybe even Z87/97 ( depends from board ).
Looks like hypers. The same as above Dominator kit. Test it on new and old platforms. If you make it run on anything newer @2400 CL8-9-8 or 8-8-8 or tighter then it's a keeper.
2000-Geil Evo2 http://www.geil.com.tw/German/reviews/show/id/212
PSC
get34gb2000c6dc, 2x2GB, 6-9-6-24
These sticks won't run on a lot of setups. On z77, have had these running 6-9-6-24 1T @1910, 1.65v, 5-8-6-24 1T with 1.75v. Prior owner ran them 2600, 8-12-8-28 1T @ 1.87v and 2667, same timings @ 1.92v on z87. View attachment 161507
2133-GSkill Perfect Storm http://www.newegg.com/Product/Product.aspx?Item=N82E16820231269
Elpida BBBG/BBSE
f3-17066cl9t-6gpbps, 3x2GB, 9-9-9-24
Prior owner tested the kit in various combinations for 1105 7-9-7-24 1,62v, 1143 7-9-7-24 1,66v, 1200 7-10-7-24 1,75v and also for 1000 6-8-6-24. The kit boots and does 1m 1300 8-11-8-22 at 1,89v on M5F, but fails 32m. View attachment 161508
All above are worth to keep. Just test them for max performance and pick the best kits. It will be hard to buy anything similar so better have 1-2 spare kits for each platform. Some of them looks really good, others are above average too.
Both above TridentX are Hynix based CFR/MFR. Good almost only for high clocks but not high performance ( at least not on air and clocks below 3000 ). If you are not planning to push them for max clocks then I would sell them.
Summary:
- Check Micron based kits - if you find Elpida as good or better then sell all Microns. If price will be really low then can keep them.
- If you won't use Hynix based kits to make high clocks then sell them as for benching they're useless comparing to all these PSC/Samsungs.
- Other kits based on 1GB modules - I would sell them but the same as with Microns, check how are they acting comparing to other kits on 775. Problem with 1GB modules is that on Win7/8+ you already need ~2.5GB+ RAM and on XP you run almost only SPI32M and most other tests are not reacting good on memory performance so can use anything slower.
- Test PSC/Hypers and pick best of them for all platforms. This and Samsung are best for benching. Hypers depends from kit and used motherboard as not all are performing/overclocking good on Z77 or anything newer. My Hypers stuck at not much above 2200 8-8-8/9-9-9 depends from board so I keep them only for AMD.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
