• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

2400Mhz Double-Sided Fast & Unstable, Single-Sided Slow, Dull but Stable, Which one to Keep?

MyNewRig

New Member
Joined
Mar 7, 2014
Messages
13 (0.00/day)
Hello,

I have the G.Skill TridentX 2400Mhz 8GB Kit, on a Z87X-UD3H board and i5 4670K,

The kit is Double-Sided and has a great performance of 35/37/34GB/s Read/Write/Copy in AIDA64 Benchmark, the kit can also clock to 2600Mhz with voltage but is not very stable at that level.

When loading XMP1 of 2400Mhz the kit is unstable in two ways, the most annoying is that it makes OSX freeze every few hours, by running memtestOSX it produces errors and show that data in memory changes when running at 2400Mhz 10-12-12-31 CR2 1.65v, it also produce memory_management BSOD in windows but less frequently, like once every week or so.

If i run this kit on 2133Mhz 9-11-11-30 1.65 it is rock solid and all the symptoms above disappear.


I deemed the kit defective and ordered another identical kit of a newer production, when i received it turned out to be a Single-Sided kit, i said ok let me test it and see, the kit turned out to be rock solid at 2400Mhz by just loading XMP1, but there is a catch, the kit turned out to be 1) significantly slower than the double-sided kit, and 2) is very dull and can not move a bit over 2400Mhz and can not run on CL9, it is 3GB/s Slower than the double-sided kit which brings all benching scores down.

I compared Geekbench scores of the Double-sided kit running at 2133Mhz with the Single-Sided kit running at 2400Mhz and it turned out that the double-sided one is superior to the single-sided one even at 2133Mhz vs 2400Mhz for the SS.

That means that when running the double-sided on 2133Mhz it has a higher bandwidth by about 1.5GB/s than the single-sided running at 2400Mhz


another thing to add to the mix is that, when running at 2133Mhz my CPU requires only 1.85v to run stable at 4.2Ghz and of course has great low temps on air of below 70c under stress, when running the CPU on 2400Mhz RAM on 1.85v it gives BSOD after few hours of stress with Bug Check 0x124 and it requires around 2.4v to be stable which brings the temps up to 85c+ which is very high.


Now i have the two kits and i would like to keep one and return the other, the question is which one to keep? keep in mind that if return the Double-Sided kit i have it will be impossible to get my hands on another.

1) Do i keep the double-sided older kit, run it at 2133Mhz on 1.6v which will give me a similar performance of the single-sided kit running at 2400Mhz 1.65v and this will benefit by lowering the voltage on my IMC while achieving the same bandwidth, also this would allow me to maintain the low 1.85v for my CPU while remaining stable and thus benefit from lower temps? the down side of that is that i will keep a kit advertised at 2400Mhz which can not run at it rated speed and voltage, just because it has higher bandwidth

OR

2) Keep the single-sided and Dull kit that does not overclock a bit and is 3GB/s slower, because it provides stability and ease of setup because all that is required to run it is loading XMP1 and that is it, i lose significant performance and overclocking headroom for the sake of stability. and i lose the chance to own the double-sided because it seems to be very rarely manufactured.


OR

3) Get another non-G.skill kit that is known to be double-sided for example Corsair Vengeance Pro (i do not know if it is DS or not so i am asking)?


BTW, i also tried the Kingston HyperX Predator of 2400Mhz 11-13-13-30 and it has even worse performance than the Single-Sided TridentX, it could overclock easily to 2666Mhz at 1.65v but surprisingly the overall bandwidth gets worse, only write performance improves from 35Gb/s to 38Gb/s but, read and copy becomes much worse and dips below 30Gb/s


I am very confused because if i let the Double-sided TridentX kit go and have it refunded, i think will never be able to get my hands on another Double-sided TridentX because it seems that they are not being produced anymore.

What should i do? keep the Double-Sided kit and run it at 2133Mhz or return it for a refund and keep the newer production SS kit that is slower but stable?

Please advice,
Thanks :)
 
Joined
Oct 29, 2012
Messages
1,926 (0.46/day)
Location
UK
System Name TITAN Slayer / CPUCannon / MassFX
Processor i7 5960X @ 4.6Ghz / i7 3960x @5.0Ghz / FX6350 @ 4.?Ghz
Motherboard Rampage V Extreme / Rampage IV Extreme / MSI 970 Gaming
Cooling Phanteks PHTC14PE 2.5K 145mm TRs / Custom waterloop / Phanteks PHTC14PE + 3K 140mm Noctuas
Memory Crucial 2666 11-13-13-25 1.45V / G.skill RipjawsX 2400 10-12-12-34 1.7V / Crucial 2133 9-9-9-27 1.7V
Video Card(s) 3 Fury X in CF / R9 Fury 3840 cores 1145/570 1.3V / Nothing ATM
Storage 500GB Crucial SSD and 3TB WD Black / WD 1TB Black(OS) + WD 3TB Green / WD 1TB Blue
Display(s) LG 29UM67 80Hz/Asus mx299q 2560x1080 @ 84Hz / Asus VX239 1920x1080 @60hz
Case Dismatech easy v3.0 / Xigmatek Alfar (Open side panel)
Audio Device(s) M-audio M-track / realtek ALC 1150
Power Supply EVGA G2 1600W / CoolerMaster V1000 / Seasonic 620 M12-II
Mouse Mouse in review process/Razer Naga Epic 2011/Razer Naga 2014
Keyboard Keyboard in review process / Razer Blackwidow Ultimate 2014/Razer Blackwidow Ultimate 2011
Software Windows 7 Ultimate / Windows 7 ultimate / Windows 7 ultimate
Benchmark Scores cinebench 15.41 3960x @ 5.3ghz Wprime32m 3.352 3960x @ 5.25ghz Super PI 32m: 6m 42s 472ms @5.25ghz
See if the double sided kit will stabilize with 10-12-12-34 timings because the last timing is supposed to be the sun of the 3 before it. Also raising it should do nothing to performance.
Also try feeding the ram 1.675v and increasee the voltage of the memory controller because haswell has been know to not like cpu OC + high ram speed.
 

MyNewRig

New Member
Joined
Mar 7, 2014
Messages
13 (0.00/day)
I did not try to raise the 31 to 34 but i actually tried raising CL from 10 to 11 and System Agent Voltage (VTT) to +0.172 and also raised vcore from my tested to be stable 1.85v to 2.7v also i tried dropping CPU clock back to defaults .. all this done at once will not prevent MemtestOSX from throwing errors.

The way MemtestOSX works is by splitting available memory into two portions, writing the same exact data to both and then compare data, if the memory is stable then data integrity should be 100%, what happens is that it finds that data in one portion is changing from the one in the other and throw errors.

MemtestOSX is the best diagnostic tool i found because it exposes errors very easily in a 30 minutes test, otherwise Memtest86 passes with flying colors and the next best thing to know that they are not stable is to wait a week or two until windows throw the memory_management BSOD .. which is not a practical testing method.

Otherwise on 2400 11-12-12-31 tRFC 313 (raised from 193 XMP1 default) VTT +0.172, vcore 2.7v and even dropping CPU to default still MemtestOSX manages to find that data held in the memory is being changed ..

I did not try to raise vcore volt above 1.65v because i do not think i would like using the system at anything above that for 24/7 usage.
 

sneekypeet

Retired Super Moderator
Joined
Apr 12, 2006
Messages
29,407 (4.48/day)
System Name EVA-01
Processor Intel i7 13700K
Motherboard Asus ROG Maximus Z690 HERO EVA Edition
Cooling ASUS ROG Ryujin III 360 with Noctua Industrial Fans
Memory PAtriot Viper Elite RGB 96GB @ 6000MHz.
Video Card(s) Asus ROG Strix GeForce RTX 3090 24GB OC EVA Edition
Storage Addlink S95 M.2 PCIe GEN 4x4 2TB
Display(s) Asus ROG SWIFT OLED PG42UQ
Case Thermaltake Core P3 TG
Audio Device(s) Realtek on board > Sony Receiver > Cerwin Vegas
Power Supply be quiet DARK POWER PRO 12 1500W
Mouse ROG STRIX Impact Electro Punk
Keyboard ROG STRIX Scope TKL Electro Punk
Software Windows 11
my guess is that simply the main timings are not the issue. While I cannot give specific values to try, I would guess that changing the right secondary and tertiary timings would allow those kits to run well. It is simply that the XMP on the kits differs from what that motherboard needs for stability.

Cadaveca is a good guy to ask, but then again he may be busy.
 

cadaveca

My name is Dave
Joined
Apr 10, 2006
Messages
17,232 (2.63/day)
BTW, i also tried the Kingston HyperX Predator of 2400Mhz 11-13-13-30 and it has even worse performance than the Single-Sided TridentX, it could overclock easily to 2666Mhz at 1.65v but surprisingly the overall bandwidth gets worse, only write performance improves from 35Gb/s to 38Gb/s but, read and copy becomes much worse and dips below 30Gb/s

With this information, I blame your CPU. Bandwidth tends to drop off when IMC (integrated memory controller) is not stable.

Have you tried XMP2 on the dual-sided sticks?


I see that you list 2.4V...2.7V...on CPU voltage? That's CPU input, and running that high is not recommended. Stock should be around 1.75V or so. Unless, of course, you made a typo.


If you increase memory speed, and you "have" to increase CPU voltage for stability...that is not the right approach, and hints at other issues. The Analog I/O and Digital I/O as well as VCCSA need an increase to bring stability to memory OCs.
 

MxPhenom 216

ASIC Engineer
Joined
Aug 31, 2010
Messages
12,944 (2.61/day)
Location
Loveland, CO
System Name Ryzen Reflection
Processor AMD Ryzen 9 5900x
Motherboard Gigabyte X570S Aorus Master
Cooling 2x EK PE360 | TechN AM4 AMD Block Black | EK Quantum Vector Trinity GPU Nickel + Plexi
Memory Teamgroup T-Force Xtreem 2x16GB B-Die 3600 @ 14-14-14-28-42-288-2T 1.45v
Video Card(s) Zotac AMP HoloBlack RTX 3080Ti 12G | 950mV 1950Mhz
Storage WD SN850 500GB (OS) | Samsung 980 Pro 1TB (Games_1) | Samsung 970 Evo 1TB (Games_2)
Display(s) Asus XG27AQM 240Hz G-Sync Fast-IPS | Gigabyte M27Q-P 165Hz 1440P IPS | Asus 24" IPS (portrait mode)
Case Lian Li PC-011D XL | Custom cables by Cablemodz
Audio Device(s) FiiO K7 | Sennheiser HD650 + Beyerdynamic FOX Mic
Power Supply Seasonic Prime Ultra Platinum 850
Mouse Razer Viper v2 Pro
Keyboard Razer Huntsman Tournament Edition
Software Windows 11 Pro 64-Bit
With this information, I blame your CPU. Bandwidth tends to drop off when IMC (integrated memory controller) is not stable.

Have you tried XMP2 on the dual-sided sticks?


I see that you list 2.4V...2.7V...on CPU voltage? That's CPU input, and running that high is not recommended. Stock should be around 1.75V or so. Unless, of course, you made a typo.


If you increase memory speed, and you "have" to increase CPU voltage for stability...that is not the right approach, and hints at other issues. The Analog I/O and Digital I/O as well as VCCSA need an increase to bring stability to memory OCs.


I was also going to say, with XMP, his board might have one of the secondary/third timings wrong(too tight/loose), compared to what the SPD in the BIOS shows like how mine was Dave when you helped me get my XMP to work at 2133 on my board.
 

cadaveca

My name is Dave
Joined
Apr 10, 2006
Messages
17,232 (2.63/day)
I was also going to say, with XMP, his board might have one of the secondary/third timings wrong(too tight/loose), compared to what the SPD in the BIOS shows like how mine was Dave when you helped me get my XMP to work at 2133 on my board.
Gigabyte isn't ASUS.
 

MyNewRig

New Member
Joined
Mar 7, 2014
Messages
13 (0.00/day)
ok so you guys seem to suspect that settings are to blame and not ICs that are unstable @ 2400Mhz?

The thing is that XMP on the DS kit is identical to that of the SS kit except for the tRFC which is 193 on the DS and 313 on the SS which is to be expected from high density ICs

the motherboard ignores some of the secondary timings in both kits.

I tried XMP2 which has tighter timing but no juice still throws errors

yes I tried vcore of 2.7v just for testing this was not a typo I normally run 4.2Ghz on 1.85v which is very stable except when you run at 2400Mhz and stress the memory for hours, you end up with bug check 124 BSOD this does not happen when running at 2133Mhz and same low vcore of 1.85v

interesting to know that bandwidth drop on higher memory clocks indicates unstable IMC, unfortunately I have returned the HyperX preditor kit because I got sick from so much testing with different kits.

I tried raising VCCSA from Auto to +0.173 but did not play with analog and digital I/O which is the only thing I did not try
 

cadaveca

My name is Dave
Joined
Apr 10, 2006
Messages
17,232 (2.63/day)
I tried raising VCCSA from Auto to +0.173 but did not play with analog and digital I/O which is the only thing I did not try


I/O voltages are crucial when pushing "weak" IMCs. I've had chips that boost 2933 MHz, the highest multiplier, with no problem. My current CPU requires a boost to I/O before it can boot @ 2933, even.

You can try adjusting timings in BIOS to match the SPD of the sticks, if you can get the proper data off the sticks, but based on what you say, with x124 error, it's your CPU. Yes, in the past, x124 points to a need for vCORE, but not with current chips..it can be both IMC and math cores.


That said, 10-12-12-31 @ 2400 Mhz is nothing special...even 9-11-11-31 isn't that great, honestly. Many sticks out there both Samsung- and Hynix-based can do those timings and clocks.


you can try 11-13-13-35 on the dual-sided sticks, see if that works...if it does, than I'd look a bit more at the memory, but even then, I'd still look at CPU.
 

MxPhenom 216

ASIC Engineer
Joined
Aug 31, 2010
Messages
12,944 (2.61/day)
Location
Loveland, CO
System Name Ryzen Reflection
Processor AMD Ryzen 9 5900x
Motherboard Gigabyte X570S Aorus Master
Cooling 2x EK PE360 | TechN AM4 AMD Block Black | EK Quantum Vector Trinity GPU Nickel + Plexi
Memory Teamgroup T-Force Xtreem 2x16GB B-Die 3600 @ 14-14-14-28-42-288-2T 1.45v
Video Card(s) Zotac AMP HoloBlack RTX 3080Ti 12G | 950mV 1950Mhz
Storage WD SN850 500GB (OS) | Samsung 980 Pro 1TB (Games_1) | Samsung 970 Evo 1TB (Games_2)
Display(s) Asus XG27AQM 240Hz G-Sync Fast-IPS | Gigabyte M27Q-P 165Hz 1440P IPS | Asus 24" IPS (portrait mode)
Case Lian Li PC-011D XL | Custom cables by Cablemodz
Audio Device(s) FiiO K7 | Sennheiser HD650 + Beyerdynamic FOX Mic
Power Supply Seasonic Prime Ultra Platinum 850
Mouse Razer Viper v2 Pro
Keyboard Razer Huntsman Tournament Edition
Software Windows 11 Pro 64-Bit
ok so you guys seem to suspect that settings are to blame and not ICs that are unstable @ 2400Mhz?

The thing is that XMP on the DS kit is identical to that of the SS kit except for the tRFC which is 193 on the DS and 313 on the SS which is to be expected from high density ICs

the motherboard ignores some of the secondary timings in both kits.

I tried XMP2 which has tighter timing but no juice still throws errors

yes I tried vcore of 2.7v just for testing this was not a typo I normally run 4.2Ghz on 1.85v which is very stable except when you run at 2400Mhz and stress the memory for hours, you end up with bug check 124 BSOD this does not happen when running at 2133Mhz and same low vcore of 1.85v

interesting to know that bandwidth drop on higher memory clocks indicates unstable IMC, unfortunately I have returned the HyperX preditor kit because I got sick from so much testing with different kits.

I tried raising VCCSA from Auto to +0.173 but did not play with analog and digital I/O which is the only thing I did not try

Wait, are you talking vcore here, or Input Voltage. 2 very different settings. Like Dave said, its not recommended to run those kinds of voltages on either settings. Im not even sure motherboards allow 2.7v on the vcore.
 

MyNewRig

New Member
Joined
Mar 7, 2014
Messages
13 (0.00/day)
ok you are making some great mind-opening points, lets break this down a little:

1- the x124 happens on both kits when running @ 2400Mhz and CPU @ 4.2Ghz with vcore 1.85v and running HCI design memtest 4 instances utilizing 95% of available memory, running 10 passes for about 5 hours or so, it always ends up with x124 BSOD

this is fixed by either a) running RAM @ 2133 or b) raising vcore to 2.4v or c) dropping CPU back to default

The x124 is not the main issue at hand that I am currently trying to solve, it is just meant to give the full picture


2- if the SS kit can run XMP1 of 2400 10-12-12-31 without throwing errors although it is 3GB/s slower in bandwidth, why can't the DS kit run on the same exact profile with the same exact mobo and CPU without throwing errors? without the need for highe I/O voltage?

3- VTT, Digital & Analog I/O are presented in offsets and not raw numbers and you have no way of telling what the final value is if you increment them, when you say raise them, what number are we talking about, like +0.xxx how can I calculate that?

4- is the extra bandwidth in the DS kit causing the IMC to act up and effect data integrity stored into the RAM even though an identical settings and kit runs flawlessly at 2400Mhz on the IMC just because its write and copy speed is 3GB/s faster?
 

cadaveca

My name is Dave
Joined
Apr 10, 2006
Messages
17,232 (2.63/day)
1- the x124 happens on both kits when running @ 2400Mhz and CPU @ 4.2Ghz with vcore 1.85v and running HCI design memtest 4 instances utilizing 95% of available memory, running 10 passes for about 5 hours or so, it always ends up with x124 BSOD

this is fixed by either a) running RAM @ 2133 or b) raising vcore to 2.4v or c) dropping CPU back to default

The x124 is not the main issue at hand that I am currently trying to solve, it is just meant to give the full picture ?

No, actually, it IS the main problem here. Both kits give x124 = IMC not capable @ those speeds without tweaking. the i5-4670k only supports memory speeds of up to 1600 MHz... so at 2400 MHz, you're pushing a 50% OC on the IMC. 50% OC, without proper voltage adjustments sounds right to you?


2- if the SS kit can run XMP1 of 2400 10-12-12-31 without throwing errors although it is 3GB/s slower in bandwidth, why can't the DS kit run on the same exact profile with the same exact mobo and CPU without throwing errors? without the need for highe I/O voltage??

The single-sided kit places less load on the controller, evidenced by the lower bandwidth. That's why it appears to work better.

3- VTT, Digital & Analog I/O are presented in offsets and not raw numbers and you have no way of telling what the final value is if you increment them, when you say raise them, what number are we talking about, like +0.xxx how can I calculate that??

Boot into BIOS, load optimized defaults, and then set offsets for each domain at the minimal value. Save, reboot, and you'll have the rough stock values so you can calculate your offset needed.

4- is the extra bandwidth in the DS kit causing the IMC to act up and effect data integrity stored into the RAM even though an identical settings and kit runs flawlessly at 2400Mhz on the IMC just because its write and copy speed is 3GB/s faster?

It's not just that it's 3Gb/s faster...it's WHY it is faster that makes it unstable.
 

MyNewRig

New Member
Joined
Mar 7, 2014
Messages
13 (0.00/day)
Wait, are you talking vcore here, or Input Voltage. 2 very different settings. Like Dave said, its not recommended to run those kinds of voltages on either settings. Im not even sure motherboards allow 2.7v on the vcore.

sorry I am talking about the value after the decimal point, I am sorry the correct numbers I meant are 1.24 and 1.27 not 2.7 of course lo,

1.185v is what I normally use, yes all these values were a typo

apologies, it is 4AM in the morning here in Europe, I am typing on a tablet and so exhausted from testing memory all day long

please adjust all the numbers in my previous posts, they are what comes after the decimal point 1.xx
 

cadaveca

My name is Dave
Joined
Apr 10, 2006
Messages
17,232 (2.63/day)
Try 1.25V on CPU to eliminate it from the equation, & work on the memory. 1.25V on vCPU, 1.25V on Cache, 1.2 on VCCSA, DIO, AIO should be enough to max out your chip.


NO worries about the typos...it happens.


Of course, you want to keep those values all as low as is possible, and really, you should not need to exceed what I posted above. I simply just would not at all, if on air-cooling.
 

MyNewRig

New Member
Joined
Mar 7, 2014
Messages
13 (0.00/day)
Wow cadaveca, you have just caused a series of paradigm shifts to occur in my brain that I jumped off bed and came to my desktop computer so that I can type faster, let me address some of your points in order in an attempt to understand what I am doing, so kindly bear with me :)

No, actually, it IS the main problem here. Both kits give x124 = IMC not capable @ those speeds without tweaking. the i5-4670k only supports memory speeds of up to 1600 MHz... so at 2400 MHz, you're pushing a 50% OC on the IMC. 50% OC, without proper voltage adjustments sounds right to you?.

Yes at first this sounded right because:

a) even though Intel officially only supports 1600Mhz, in its XMP support PDF list, they actually list speeds of up to 2933Mhz which they themselves test and approve for use via loading XMP profiles, so yes Intel does promote these speeds in a way, but they just do not want to be RMAing CPUs because people can not run them at 1866Mhz+, and

b) XMP profile data seem to contain IMC voltage settings as well, as seen in AIDA64 SPD data, in AIDA64 SPD data I can see the following as part of XMP1 -> Voltage 1.65 V (Memory Controller: 1.25 V) <- , "whatever that is I am not sure what is meant by Memory Controller in XMP data?", so in my understanding it should be. load XMP1 and go, leaving everything else on Auto. (maybe this is one of the values that the mobo ignores?)



The single-sided kit places less load on the controller, evidenced by the lower bandwidth. That's why it appears to work better.

What is reason behind the "less load" thing, when both run at exactly the same frequency, timings and voltage? does bandwidth transfer rate, relate directly to stress on IMC? G.Skill tech support seemed to advice otherwise as he related IMC stress to memory frequency and not to actual bandwidth the ICs are capable of.



Boot into BIOS, load optimized defaults, and then set offsets for each domain at the minimal value. Save, reboot, and you'll have the rough stock values so you can calculate your offset needed.

Here is where I have a problem, in Gigabyte BIOS, and even in windows tools like TweakLuncher2 and EasyTune, I have no way of seeing the raw values of these things, I only see offsets all the way, which Is stupid I know, but I am not sure how to actually know what the end voltage will be after adjusting the offsets.


It's not just that it's 3Gb/s faster...it's WHY it is faster that makes it unstable.

Please elaborate? I have been told by G.Skill tech that the reason it is faster is that it uses 256Mb ICs which runs at tRFC of 193 as apposed to the 512Mb ICs in the SS kit, and would therefore be faster, how is that contributing to instability? because this seems like one of the critical points to understand in this situation.


Try 1.25V on CPU to eliminate it from the equation, & work on the memory. 1.25V on vCPU, 1.25V on Cache, 1.2 on VCCSA, DIO, AIO should be enough to max out your chip.


NO worries about the typos...it happens.


Of course, you want to keep those values all as low as is possible, and really, you should not need to exceed what I posted above. I simply just would not at all, if on air-cooling.


This is a problem as indicated above,

a) I have the Hyper 212 Evo cooler so 4.2 and 1.250v makes temps go in the high 80s which I think is extreme when running under stress, this is why I keep it at 1.185v because I seem to have a good low voltage chip that is 100% stable at 4.2Ghz 1.185v up to 2133Mhz RAM speed, but of course I can set it up to 1.250v just for testing.

b) I am not sure what cache voltage is? but I will search for it in voltages

c) VCCSA, DIO, AIO like I said are displayed in offsets all the way, through BIOS and windows tools, how can I check the final raw value? this is one of the reason I felt very reluctant, changing these values from Auto, because I would not know what the heck I am changing them to without a tool that displays raw values, and I am not sure what this tool would be in Gigabyte mobo?
 
Last edited:

cadaveca

My name is Dave
Joined
Apr 10, 2006
Messages
17,232 (2.63/day)
Although XMP is "rated" by Intel as possible...doesn't mean every CPU will do it. Most will not, actually, without a large voltage boost.


OK, so you got 2 sets of sticks, 1x 256 mbit, 1x512 mbit. 256 mbit uses 16 memory chips to give 4 GB, but 512 mbit, only uses 8, as you've noticed(single-sided). That is literally less banks/ranks for the IMC to deal with, and therefore, less load. By load..there is less bank interleaving, etc, which makes the controller actually work less. It's a minor difference, really, but it does play a factor. Of course, as you've seen , the real difference isn't enough to really go form "unstable" to "stable", but it does play a role. I've tested quite a few chips with nearly every DDR3 IC on the market, and getting the best out of each IC does require different voltages at times, too. For example, I found Hynix-based stick to require a bit more voltage than Samsung...but Samsung does offer better bandwidth. So what the G.Skill tech told you is accurate for sure... it just isn't as simple as just that though.

I'll suggest at this point that you read my OC guide, to give a general baseline of info that we can build off of. Once you see my opinions and such in that guide, perhaps we will be on the same page. I see you are using software to look at votlages...but it actually requires a multimeter. There are probe points on the top-right corner of your board for expressly this purpose. Aio and Dio voltages as well as VCCSA are individual to each chip, so I cannot tell you anything other than suspected ranges as given by Intel's documentation, which you will find repeated in my OC guide.


By the way, Intel DOES support OC within warranty. You can purchase an "extended" Tuning Plan warranty from Intel directly for about $25 that will cover OC problems, and offer easy RMA. with built-in temperature and current protections built into these chips, it should be fairly hard for normal use to hurt these chips... although I have killed quite a few Haswell chips myself, and most often it's IMC that dies... after de-lidding. I have spoken to a few users now that have found the same to occur too, but nothing to outright say "IF YOU DE-LID, YOUR CHIP WILL DIE!!".


Intel Tuning Plan can be found here:

http://click.intel.com/tuningplan/


My overclocking guide can be found here:

http://www.techpowerup.com/reviews/Intel/Haswell_OC_Guide/
 
Last edited:

MyNewRig

New Member
Joined
Mar 7, 2014
Messages
13 (0.00/day)
OK a few more quick things before I pull the plug on this test round just to make sure that everything is taken into account:


1- I just want to make sure that you are aware that, this DS kit produces errors in MemtestOSX (as in data held in memory being altered or changed when read back) even when CPU is running at stock clock and voltage, everything on Auto without any overclocking and just loading XMP1, but it does not produce x124 BSOD in windows, the point is that, by reverting CPU back to stock with the SS kit I can eliminate BOTH problems, but with the DS kit I still get data changed in memory.

2- Which is completely something on the side not related to stability itself but is very important for my testing, will removing and reinserting memory modules many times for the sake of this testing make my memory banks on the motherboard degrade physically? am I damaging my mobo by replacing kits few times a day to complete these tests?

3- Because I have been testing for over a week and very exhausted from this whole situation, can I temporarily skip your guide by setting some rough voltages and testing the DS kit with MemtestOSX? the idea is that I would set known to be stable loose voltage levels and test again if data can maintain integrity in memory, if it passes I then read your guide and go out and buy a multimeter to tweak things properly, if not I would just return this kit.

4- Finally are these 512mbit single-sided 2400 CL10 new production TridentX really trash like people say in the forums? I mean if I could not get the DS kit stable, what are my better alternatives? should I keep the new SS one? or what kit do you recommend I get as a replacement? keeping in mind that 2400Mhz is very important for me, because I run a 3D simulator that is CPU/RAM bound and I need the max performance I could get in this app.

Sorry for the trouble and thanks again :)
 

MyNewRig

New Member
Joined
Mar 7, 2014
Messages
13 (0.00/day)
Hello again cadaveca, I hope you are around,

I went through most of your guide, and posts here, and the first thing i have tried after reinstalling the OLD double-sided modules and returning the new single-sided ones back to the store is raise the timing to 11-13-13-35 and here goes the errors in MemtestOSX, so it seems like they are the modules themselves that were causing the problem and not the IMC (the IMC would i guess still produce x124 if heavily stressed for long hours) but for now the errors that were produced by a simple memtest have gone away by increasing the timings ..

I would not raise the voltage to more than 1.65v for 24/7 use even though i suspect this will stabilize the kit ..

I will next try leaving things at 10-12-12 like they are in XMP1 and only increase the 31 to 34 and see if this fixes the problem ..

Could you please advice on what is a good 8GB double-sided kit is available in the market these days? i have tried the HyperX Predator and know for sure that they are single-sided 1 Rank kit ... is the Vengeance Pro any good? or what else is out there that is double-sided in 4GB sticks?
 

cadaveca

My name is Dave
Joined
Apr 10, 2006
Messages
17,232 (2.63/day)
I'm always around, just not always motivated. :roll:

You can try G.Skill TridentX ( i have a current review on the front page of 8 GB modules), or other G.Skill modules.

Crucial has some nice sticks for lower speeds, 1866 and lower, with 1.35V.

Corsair Vengeance Pro and Dominator Platinum are good too, for higher-end stuff, as is Avexir's new Blitz 1.1 ( I ordered a custom set of these myself to use for board testing, with a customized SPD).

Really, brand doesn't matter too much, as long as you go with modules that are certified to work with your motherboard. This is the most important part, as there are timings that some board do not offer adjustment for that differs from band-to-brand.

Personally, I like all of these brands for different purposes, and what matters to me is warranty and such, and Corsair, Crucial and G.Skill are all very good in this regard. I have not used Avexir's RMA service as of yet.


Within those brands, each has specific modules that will feature specific ICs, but at this point DDR3 is on the decline, and many speed bins from all brand use 1R DIMMs. This is why I tend to not report IC info in my memory reviews, since the ICs can and will be changed according to market availability. Corsair is pretty good for having revision numbers right out in the open on each set, but then you are left to the whim of retailers in relating such info.



I've only kept one set of 4 GB sticks myself, a Corsair Dominator Platinum kit which is based on Samsung ICs, and I have those Avexir modules coming, which are likely Hynix-based, but may very well be single-sided, in which case I'm gonna be pretty disappointed. That said, I'd look for Samsung-based sticks, as they offer the best performance and tightest latency support per clock, but it will be hard to find really good kits...I wouldn't let go of mine for anything less than $1000. I'd rather keep them forever than sell them cheaply, and I have had quite a few requests to sell them for top dollar already.
 

MyNewRig

New Member
Joined
Mar 7, 2014
Messages
13 (0.00/day)
TridentX is what my story is all about :D

The 4GB modules that are being manufactured these days are single sided, hence my issue

I see the ones in your review are double-sided (yummy) ... but these are expensive for my needs, i need only 8GB dual-channel kit, i am not willing to pay for 16GB kit because i do not need it and would be paying more than i would like ..

UPDATE: i was also able to stabilize the kit by running them on 1.675v and loading XMP1 of 2400 10-12-12-31 ... so my issue is indeed caused by the modules themselves not stable at advertised specs ... and not the IMC (IMC also not stable for prolonged stress at 2400) so i think my issue is complex, 1) unstable kit at rated specs, 2) unstable IMC at 2400 .. lol

So now back to square one, the 8GB TridentX 2400 C10 kit replacements available to me in the current market are single-sided .. i can also have my retailer replace it with the Vengeance Pro 2400 11-13-13-31 (i am not sure if these are double sided or not)

So now back to square one:

1- Should i keep my current kit and run it on 1.675v? too high for 24/7 use?
2- Keep my current kit and run at 2200 or 2133 CL9 1.65v?
3- Keep the new single-sided, slow, dull and stable kit?
4- Get the Vengeance Pro 2400 11-13-13-31?
5- Kill myself? LOL :D

BTW: how can i identify the ICs? or what should my purchasing decision be based on?

Please be motivated enough to answer this one haha

EDIT1: What do you mean by DDR3 on the decline? they will stop manufacturing them soon for DDR4?
 
Last edited:

cadaveca

My name is Dave
Joined
Apr 10, 2006
Messages
17,232 (2.63/day)
1- Should i keep my current kit and run it on 1.675v? too high for 24/7 use?

should be fine for the memory. Haswell seem to like lower DRAM voltages though I have yet to see any issues from running ~1.75 V

2- Keep my current kit and run at 2200 or 2133 CL9 1.65v?

well.. 2133 MHz seem the best optimized divider for most boards. That @ 1.6 V or less, to me, is ideal for long-term use.

3- Keep the new single-sided, slow, dull and stable kit?

Meh. I'd keep it, shove 1.85V into it, and have some fun.

4- Get the Vengeance Pro 2400 11-13-13-31?

I wouldn't. Not that there is anything wrong with them... I simply would choose a different model for this application.

5- Kill myself? LOL :D

LuLz. First-world problems.

BTW: how can i identify the ICs? or what should my purchasing decision be based on?

By physical identification and voltage response is best, otherwise...

EDIT1: What do you mean by DDR3 on the decline? they will stop manufacturing them soon for DDR4?

Sure. I mean, it would be really nice if everyone could have that nice Samsung stuff, but it is EOL as far as I know. Maybe someone has a stash of it somewhere, or something, but if you are into benchmarking, you're binning cheaper 2 GB sticks right now anyway. Lots of those out there.
 

MyNewRig

New Member
Joined
Mar 7, 2014
Messages
13 (0.00/day)
Ok great, thanks a bunch for answering this one, this thread has been the most useful, thanks to you and others for the valuable input,

Like you said, i think 2133 CL9 1.6v on the double-sided kit is the best of both worlds, since the DS performance at 2133 is almost similar to SS performance at 2400 yet i win many things by keeping the old DS one and running at 2133

doing this i stress the IMC less by both running at lower DRAM voltage and a not so high frequency, in windows it is stable without any voltage increase to the IMC like VTT, DIO, AIO, and in OSX it is also stable ..

I also have the opportunity to replace it later if i find that i want to run at 2400 without so much hassle ..

So keeping the old DS kit and running at 2133 and returning the new SS back to store .. case closed :)
 
Top