DDR4-3200 CL16 2x8Gb Kit: Corsair Vengeance LPX or G.Skill Ripjaws V or Patriot Viper Steel ?

[Coolio]

Hi folks,

I'm choosing the 3200 DDR4 RAM for my fanless Intel-based build (MoBo's RAM requirements) - office use and home theater with occasional Plex transcoding.

I won't overclock, so what I need is that the RAM works on real 3200 clocks, the heat is nicely dissipated, and the quality of modules is reputable.

The die they use may (?) differ, so if you're aware of these nuances, I will appreciate your comments on that.

Here are some more details:

• Corsair Vengeance LPX: CL16-18-18-35 (XMP 2.0 support, aluminium heat spreader, height 33.5mm/1.32")
• G.Skill Ripjaws V: CL16-18-18-38 (XMP 2.0 support, height 42mm/1.65")
• Patriot Viper Steel: CL16-18-18-36 (XMP 2.0 support, aluminium heat spreader, height 44.5mm/1.75")

Thank you for your comments!

 

[riev90]

Hi folks,

I'm choosing the 3200 DDR4 RAM for my fanless Intel-based build (MoBo's RAM requirements) - office use and home theater with occasional Plex transcoding.

I won't overclock, so what I need is that the RAM works on real 3200 clocks, the heat is nicely dissipated, and the quality of modules is reputable.

The die they use may (?) differ, so if you're aware of these nuances, I will appreciate your comments on that.

Here are some more details:

• Corsair Vengeance LPX: CL16-18-18-35 (XMP 2.0 support, aluminium heat spreader, height 33.5mm/1.32")
• G.Skill Ripjaws V: CL16-18-18-38 (XMP 2.0 support, height 42mm/1.65")
• Patriot Viper Steel: CL16-18-18-36 (XMP 2.0 support, aluminium heat spreader, height 44.5mm/1.75")

Thank you for your comments!

That corsair vengeance usually use hynix A/M-Die, which usually generate more heat due to it's dense chip.
I would grab the viper steel (usually use hynix C-Die) but you have to consider it's higher profile compared to vengeance.

 

[Coolio]

I would grab the viper steel (usually use hynix C-Die) but you have to consider it's higher profile compared to vengeance.

Good point regarding die-dependent level of heat! I've just found out that G.Skill Ripjaws V also has an aluminium heat spreader. Do you have any idea on which die it uses (i.e. how much heat it generates)?
Thank you!

[UPDATE] Just noticed that all those are DIMMs, while my mini-case build supports SO-DIMMs. Will research on those and be back with updated model list.

 

[riev90]

[UPDATE] Just noticed that all those are DIMMs, while my mini-case build supports SO-DIMMs. Will research on those and be back with updated model list.

if you have any online shop preference, I may help you search directly.
FYI, the 3200mhz kits usually use cas 22 and 1.2v for vdimm as the dram manufacturer want to ensure wire range of compatibility to the motherboards.

 

[Coolio]

if you have any online shop preference, I may help you search directly.

Thank you mate! I'm from the EU, so I usually search on Idealo or Computeruniverse. But you don't have to help, really, I don't want to spend your time!

FYI, the 3200mhz kits usually use cas 22 and 1.2v for vdimm as the dram manufacturer want to ensure wire range of compatibility to the motherboards.

I'm just curious which module shall I choose, if the choice is between e.g., 2400MHz CL15-15-15-35, 2666MHz CL18-18-18-43 and 3200MHz CL22-22-22-52 ?
Should frequency be my priority or timings, or does it rather depend on the nature of the tasks I will have?

 

[riev90]

Thank you mate! I'm from the EU, so I usually search on Idealo or Computeruniverse. But you don't have to help, really, I don't want to spend your time!

I'm just curious which module shall I choose, if the choice is between e.g., 2400MHz CL15-15-15-35, 2666MHz CL18-18-18-43 and 3200MHz CL22-22-22-52 ?
Should frequency be my priority or timings, or does it rather depend on the nature of the tasks I will have?

You have 12th gen, right?
I'd rather max out the ram speed to 3200c22 as at 1.2v, the dram will not generate too much heat.

This one looks interesting:
https://www.computeruniverse.net/en/p/90823147

Team sodimm at 3200c22 usually uses micron b-die chip which is quite a good clocker.
If you plug the ram and turn the pc, the ram will boot directly to 3200c22 without any further tuning.

And I think for your use case, 16gb x2 (total 32gb) will be enough.

 

[Coolio]

Thank you so much!

If you plug the ram and turn the pc, the ram will boot directly to 3200c22 without any further tuning.

I'd rather max out the ram speed to 3200c22 as at 1.2v, the dram will not generate too much heat.

I know that a bit slower RAM with tight timings will outperform a bit faster RAM with modest timings. Do you recommend 3200CL22 because you think frequency should be prioritized vs. timings, or because these particular modules work at 3200 out of the box? I mean should my PC support XMP 2.0 profiles (checking that with Intel tech support), won't it be better to buy modules with tighter timings (e.g., 2666MHz CL18-18-18-43) and make them run at 3200 (which is the max freq. my MoBo supports) through XMP?

(I assume for PCs without XMP, your approach of going as high in "default" clocks as possible is the best and the only possible, right?)

Team sodimm at 3200c22 usually uses micron b-die chip which is quite a good clocker.

How would you prioritize dies in terms of their reliability? So B-die is a good thing, but which one should I avoid?

----
Finally, one very important thing - dual ranking. A bit of a theory I've learnt so far :

1. DIMMs must be distributed across both channels for dual-channel mode to activate. On desktop PC it will be like: 1 module goes to A1 slot, 1 module goes to B1 slot (slots A2 and B2 stay empty)
2. However, if only 2 slots (out of total 4 on MoBo) are occupied, the DIMMs must be dual-rank (2R), not single-rank (1R), as with 1R modules we'll have to occupy all 4 slots to achieve the similar performance
3. The problem is that (at least by the end of 2022) mostly big size DIMMs (e.g., 32Gb) are dual-rank, others are single-rank. Moreover, since 2R modules have lower frequencies vs. 1R modules, such 2-slot configuration works slower than 4-slot configuration with 1R modules

With all that in mind, I'm curious: since I don't need more than 16Gb in total and my MoBo has only "two 260-pin 1.2V DDR4 SDRAM SO-DIMM sockets", shall occupying both sockets with e.g., 2x8Gb modules turn the dual-channel mode ON? Am I right, that since I have only 2 sockets in total, I should go for single-rank (1R) modules to achieve the best performance (i.e., to do the opposite of what is recommended in point #2 above)?

Thank you for your kindest comments!

 

[riev90]

I know that a bit slower RAM with tight timings will outperform a bit faster RAM with modest timings.

That's correct.
3000mhz 16-18-18 will always be better than 3200mhz 22-22-22

Do you recommend 3200CL22 because you think frequency should be prioritized vs. timings, or because these particular modules work at 3200 out of the box?

1. Because your PC setup is Intel based and the memory tweaking doesn't really increase the performance
2. That's also correct. The 1240P is quite a performer for small-ish chip and I don't think it's worth it to tinker, for example tweaking the 3000 c16 sodimm (which is very hard to find these days) to 3200 16-18-18 only to gain 1~2% higher read/write/copy of data compared to 3200 22-22-22 (Plug and Play)

How would you prioritize dies in terms of their reliability? So B-die is a good thing, but which one should I avoid?

For sodimm? As long as you don't get dud chip (like hynix M/A-Die, Samsung C-Die, Nanya), it doesn't really matter.

With all that in mind, I'm curious: since I don't need more than 16Gb in total and my MoBo has only "two 260-pin 1.2V DDR4 SDRAM SO-DIMM sockets", shall occupying both sockets with e.g., 2x8Gb modules turn the dual-channel mode ON? Am I right, that since I have only 2 sockets in total, I should go for single-rank (1R) modules to achieve the best performance (i.e., to do the opposite of what is recommended in point #2 above)?

1. It really depends on the on-die chip on your purchased ram. I have found some interesting sodimms I've mentioned above:
> 1 sodimm (16gb) is made of Hynix AFR, dual rank, but Thaiphoon read it as Hynix CJR. XMP is 2666 c19, and can't go higher than 3200mhz with timings 18-17-9-22 @1.35v
> 2 sodimms (16gb) is made of Hynix MFR, dual rank, but Thaiphoon read it as Hynix CJR. XMP is 2666 c19, and max it can go is 3333mhz with timings 16-17-9-22 @1.35v or set 3133 20-19-8-22 @1.2v only
> 2 sodimms (8gb) is made of Micron B/D-Die, dual rank even though it's 8gb kit, XMP is 3200 c22, and it can boot to 4000mhz with timings 22-24-8-22 @1.35. Now both running stable at 3666mhz 22-24-8-22 @1.2v only (Asrock X300, Ryzen 4350g)

2. For 4 dimms board, to run dual channel, each kit shall be installed on A2 and B2. For 2 dimms, just populate the dimms. I would order the identical kit just to ensure getting the same on-die chip.

 

[Coolio]

riev90
Thank you for the clarification! So the logic in search for a good RAM in my case will be: 1) long-list only those modules that work at 3200 out of the box (without XMP o/clocking), 2) shortlist to those 3200s which have tighter possible timings.

Because your PC setup is Intel based and the memory tweaking doesn't really increase the performance

Does this logic apply to all Intel-based builds (i.e., not to specific CPU series, e.g., mobile ones)? Just to know for the future.

As for the Dual Channel Mode, from what you said I understand that I just need to fill both slots with identical modules (which goes without saying) - and Dual Channel Mode will be ON. Good!

I have found some interesting sodimms I've mentioned above:
> 1 sodimm (16gb) is made of Hynix AFR, dual rank, but Thaiphoon read it as Hynix CJR. XMP is 2666 c19, and can't go higher than 3200mhz with timings 18-17-9-22 @1.35v
> 2 sodimms (16gb) is made of Hynix MFR, dual rank, but Thaiphoon read it as Hynix CJR. XMP is 2666 c19, and max it can go is 3333mhz with timings 16-17-9-22 @1.35v or set 3133 20-19-8-22 @1.2v only
> 2 sodimms (8gb) is made of Micron B/D-Die, dual rank even though it's 8gb kit, XMP is 3200 c22, and it can boot to 4000mhz with timings 22-24-8-22 @1.35.

I'm surprised that modules of the same series, but from different supply type (kit/non-kit) differ in terms of the die used, and (even more so!) the same die-based modules can be o/clocked with the different result (possibly MFR from your example is manufactured from the binned dies, though - that will explain why they are faster).
Nevertheless, with that considered, seems like in my case there's just no sense to pay attention whether the module is 1R or 2R (as I have only 2 slots anyways), but what really matters is that very out-of-the-box frequency/timings mentioned earlier, since o/clocking is not recommended in my case. Correct?

 

[riev90]

Does this logic apply to all Intel-based builds (i.e., not to specific CPU series, e.g., mobile ones)? Just to know for the future.

It applies to your current build because it probably use the H type of board which has dram overclocking limit, or should I say locked oc feature.
If you have Intel 12th gen with b/z690 board or 13th gen with b/z790, and also long dimm board then I will suggest much better dram kits.

seems like in my case there's just no sense to pay attention whether the module is 1R or 2R (as I have only 2 slots anyways)

I would still prefer dual rank kit even though it's harder to stabilize those on the higher speed.
But in your use case, just plug in the ram, then even if it's dual rank, it will boot and run at 3200 22-22-22 @1.2v.

what really matters is that very out-of-the-box frequency/timings mentioned earlier, since o/clocking is not recommended in my case. Correct?

Indeed.
Also no need to feel burden, I will try to help if I can.

 

[Coolio]

Also no need to feel burden, I will try to help if I can.

OMG, you're amazing, mate - thank you so much!

It applies to your current build because it probably use the H type of board which has dram overclocking limit, or should I say locked oc feature.

My board is NUC12WSBi3 (tech spec) which is far from 690/790 boards, but maybe this tech spec will bring some more light on the situation (I'm, honestly, not that smart in reading them )

I would still prefer dual rank kit even though it's harder to stabilize those on the higher speed.

Got it, but won't 2R potentially bring me some headache? I.e., what may this stabilization "problem" mean to me in practice?

Thank you once again - I highly appreciate your help!

 

[riev90]

My board is NUC12WSBi3 (tech spec) which is far from 690/790 boards, but maybe this tech spec will bring some more light on the situation (I'm, honestly, not that smart in reading them )

Yup, that's NUC board with limited overclocking options (and probably limited bios option), too, so that any 8gb / 16gb / 32gb 3200c22 kit will just be plugged, turn on the PC, check the bios whether both ram are perfectly installed, save and then boot to OS. It's basically a plug and forget

Got it, but won't 2R potentially bring me some headache? I.e., what may this stabilization "problem" mean to me in practice?

IMHO it won't give you headache, I assure you haha.
Like I've said, I have 2 of those 3200c22 with micron chip, and the first boot into bios (I use ryzen 4350g + Asrock X300 deskmini) shown the ram running at 3200 22-22-22-22 @1.2v without any change to the settings.

As per stability issues? So far haven't found anything while trying / testing the xmp 3200c22:
No WHEA (windows machine error)
No restart or BSOD
No hiccups on windows or when playing
Everything basically run fine

Besides, the 1.2v really have lowering the heat profile.

 

[Coolio]

riev90
Thank you! PnP is probably the best in my case, since it's just an office machine (albeit pretty powerful one).
You're right regarding this board's limitations, e.g. Intel support just confirmed this morning there's no XMP capability.

I am waiting for some details from a couple of DDR4 manufacturers and will be back soon with a shortlist of modules to choose from.

 

[dasa]

1. Because your PC setup is Intel based and the memory tweaking doesn't really increase the performance

This is a myth Intel sees gains from latency\bandwidth just like AMD does.
In the past RAM speed has been enough to make i3 NUC faster than i5 NUC in tasks that are sensitive to it.

Nice to see that Intel has added XMP support to NUC again after removing it for a few generations.

 

[riev90]

This is a myth Intel sees gains from latency\bandwidth just like AMD does.

Yup, besides newest Intel's ddr5 platform, Intel gen 3 ~ gen 11 + ddr4 board only saw negligible performance increase from ram tuning.

Note: except for extreme tuning of more than 4000mhz c12 (which requires 1.8v++ on vdimm)

 

[Coolio]

Nice to see that Intel has added XMP support to NUC again after removing it for a few generations.

Not to all NUCs. My model is not on the list.

 

[dasa]

Yup, besides newest Intel's ddr5 platform, Intel gen 3 ~ gen 11 + ddr4 board only saw negligible performance increase from ram tuning.

With 3770K, 6700K, 7700K, I found that overclocking RAM with ~1.5Vdimm made a larger difference to performance in most games than overclocking the CPU and putting both together made for a very noticeable improvement. But I guess you could say that each on there own was negligible at ~7% from a fast 3200c14 XMP to 4000c17 but it becomes ~14% when combined with the CPU OC and it is a lot more if you consider stock the max supported speed of 2133c15.
https://www.overclock.net/threads/3...133c15-3000c12-4000c17.1611359/#post_25789786
Extra VDIMM didn't help much as these systems were hitting a wall with the memory controller.

I tweaked the RAM in our NUC5i3 and I belive it added a nice bit of zip to them and I remember seeing benchmarks of the NUC5i3 with fast RAM vs the NUC5i5 with slow RAM and the RAM speed closed the gap significantly with what was just a few MHz difference between the CPUI.

Not to all NUCs. My model is not on the list.

Bugger.

 

[riev90]

With 3770K, 6700K, 7700K, I found that overclocking RAM with ~1.5Vdimm made a larger difference to performance in most games than overclocking the CPU and putting both together made for a very noticeable improvement. But I guess you could say that each on there own was negligible at ~7% from a fast 3200c14 XMP to 4000c17 but it becomes ~14% when combined with the CPU OC and it is a lot more if you consider stock the max supported speed of 2133c15.
https://www.overclock.net/threads/3...133c15-3000c12-4000c17.1611359/#post_25789786
Extra VDIMM didn't help much as these systems were hitting a wall with the memory controller.

I tweaked the RAM in our NUC5i3 and I belive it added a nice bit of zip to them and I remember seeing benchmarks of the NUC5i3 with fast RAM vs the NUC5i5 with slow RAM and the RAM speed closed the gap significantly with what was just a few MHz difference between the CPUI.

What I was trying to say in my initial post that you replied:
1. The overall gain from 2666mhz c19 to 3200mhz c22 of OP's platform will not be noticeable / negligible because Intel already optimize it's arch to not really rely on the dram performance.
2. Do note that OP's NUC platform will be limited to max 3200mhz so that's why I recommend to buy the 3200 c22 kits and don't bother with tuning it further, just plug and forget.
3. Do note also in my reply to you that the extreme tuning like 4000c12 will give performance bump but honestly, how many users (per installation of Intel from gen 2 to gen 11) you know are prepared to sacrifice their time to tuning their ram and test it using memtest or TM5 and be patient to wait for more than 1.5 hours just to find out that in the middle of test, there would be error?

That's why Intel release XMP product to introduce "Auto Overclocking" on dram kits to make it easier for user base to auto tuning their ram and leave it at that.

4. I've used Intel 12100F + Asus B660I Strix with ddr5 kits:
> Kingston 32gb kit 6000 c40 Hynix M-Die
> Patriot 8gb x2 4800 c40 Hynix M-Die
I've tried both kits to run on xmp (4800c40) and overclock both to 6600c30 or 6800c32 and only AIDA and other dram sensitive apps are seeing performance jump. Not talking about gaming as at the time I use them, I only have gtx 750ti (fuck crypto!!!) to play with.

 

[dasa]

1. The overall gain from 2666mhz c19 to 3200mhz c22 of OP's platform will not be noticeable / negligible because Intel already optimize it's arch to not really rely on the dram performance.

It will be negligible due to the timings going up at a similar rate to the frequency making the end latency much the same but with a little more bandwidth which would benefit the IGP but not make much difference to the CPU.
So not really anything to do with the Intel arch in my mind.

2. Do note that OP's NUC platform will be limited to max 3200mhz so that's why I recommend to buy the 3200 c22 kits and don't bother with tuning it further, just plug and forget.

Fair enough I missread that it supported XMP but it seems that most NUC12 still do not. Sorry about that.

 

[riev90]

It will be negligible due to the timings going up at a similar rate to the frequency making the end latency much the same but with a little more bandwidth which would benefit the IGP but not make much difference to the CPU.
So not really anything to do with the Intel arch in my mind.

OP's intention is to increase his overall performance not just the IGPU, and the NUC is locked hard so kit of 3200c22 is sufficient for his needs.

But I can agree with you about increased performance of APU by further tuning of dram.
I can attest to this as I'm using Asrock Deskmini and I tuned my ram from 3200c22 to 3666 22-24-8-22 and gained staggering ~90% performance bump in one test (Unigine Valley, 1080p, low).

 

[dasa]

Do you recommend 3200CL22 because you think frequency should be prioritized vs. timings, or because these particular modules work at 3200 out of the box?

3200CL22 13.75ns
2666CL18 13.5ns
2400CL15 12.5ns
So there is a very small latency loss for a fair increase in bandwidth.

If you dont want to lose any latency you could find a kit of Kingston FURY Impact 3200CL20 12.5ns
Same latency as 2400CL15 due to the frequency increase offsetting the timing increase but still has the bandwidth benefits from the higher frequency.

The actual performance difference may be measurable but certainly not noticeable.
As mentioned you could also get a little more performance from a dual rank kit which would be in the Kingston .pdf spec as 1r or 2r.
Unfortunatly unless you can find some old stock about the only way to get 2r these days is with 32GB sticks.

 

[Coolio]

riev90 dasa

Thank you guys for your explanations, it was very helpful! I've finally got replies from all 4 manufacturers' tech support, and honestly it was not very helpful, since only Patriot agreed to share which die they use for their SO-DIMMs, and only Patriot and Kingston clearly defined which rank their modules are.
I prioritized RAM models in the descending "buy this one" order like follows:

• Kingston FURY Impact [3200, CL20-20-22-22], p/n KF432S20IBK2/16 [Rank - 1R; XMP - Yes]
  • tech sheet with exra info
• Patriot Viper Steel [3200, CL22-22-22-52], p/n PVS48G320C8S [Die - Hynix A/C/D; Rank - 1R; XMP - Yes]
  • tech sheet with extra info
• Crucial [3200, CL22-22-22-22], p/n CT2K8G4SFRA32A [Rank - not guaranteed on arrival]
• Corsair Vengeance Performance [3200, CL22-22-22-53], p/n CMSX16GX4M2A3200C22 [Rank - not guaranteed on arrival]

I placed Kingston FURY Impact on the 1st place, as this is the only CL20 module in the list. Patriot Viper Steel is the 2nd, because it supports XMP - same as Kingston. While my CPU doesn't have XMP profiles, I assumed that XMP-enabled RAM modules may be binned ones (to be able to handle o/clocking),so they should be more reliable in terms of quality. Maybe I am mistaken in such assumption.

What will you recommend from this list? Just in case - here are the memory module requirements for my build.

Thank you guys!

 

[riev90]

Thank you guys for your explanations, it was very helpful! I've finally got replies from all 4 manufacturers' tech support, and honestly it was not very helpful, since only Patriot agreed to share which die they use for their SO-DIMMs, and only Patriot and Kingston clearly defined which rank their modules are.
I prioritized RAM models in the descending "buy this one" order like follows:

• Kingston FURY Impact [3200, CL20-20-22-22], p/n KF432S20IBK2/16 [Rank - 1R; XMP - Yes]
  • tech sheet with exra info

Based on this part of datasheet:

It supports default (plug & play) at 3200 20-22-22, then this Kingston kit is the clear choice.

 

[daglesj]

I got the Kingston Fury 3200 CL20 kit for my Framework as it was the lowest latency 1.2V I could find. As mentioned only the 64GB kit (2x32GB) is dual rank.

I just got the 2x8GB kit and it works fine.

 

[Coolio]

I got the Kingston Fury 3200 CL20 kit for my Framework

It supports default (plug & play) at 3200 20-22-22, then this Kingston kit is the clear choice.

Thank you! What will be your choice #2 riev90 ? Just in case... Looks like the 3 left are almost identical, but maybe there are some differences a newbie like me doesn't pay attention to?

 

[dasa]

The kingston kit is the fastest but depending on price it's probably not the best value.

 

[Coolio]

dasa They all cost +/- $2 for the kit, so price is definitely not a marker here.

 

[dasa]

Kingston is a easy choice then as it has a slight speeed advantage with no down side.

 

[Coolio]

dasa Yes, that's clear and noted. I'm trying to find the best backup option out of those 3 left, just in case.

 

[dasa]

Everything else looks to be much the same.

 

[riev90]

Thank you! What will be your choice #2 riev90 ? Just in case... Looks like the 3 left are almost identical, but maybe there are some differences a newbie like me doesn't pay attention to?

agreed with dasa , the other 3 are pretty much same in terms of default xmp.
go with the kingston.

 

[Coolio]

OK, got it mates, thank you for your support!