Return M-die kit for A-die?

[tabascosauz]

I just picked up a 2x32GB Ripjaws 6000CL30 kit for the new CPU - it is just my luck that it is Hynix M: 042 code shows S820M.

For the same spec of 64GB 6000 30-40-40-96, there's also a Flare X SKU and another Ripjaws at the same price ($289). Drawback is that both of them are black; I picked up this Ripjaws kit specifically for white colour scheme.

There's also a 6400 32-39-39-102 kit of white Ripjaws, just $10 more.

Is it worth going back to the store to see if any of the others are A-die?

 

[Outback Bronze]

TBH, I don't think so, especially if you're after the white sticks.

A-die is really for diehard over clockers, and you would really need to be playing @ 1080p to see some results. You'll be lucky to get 1fps @ 4k going from 6k-8k MT/s.

Edit: We need @ir_cow he will know

 

[tabascosauz]

TBH, I don't think so, especially if you're after the white sticks.

A-die is really for diehard over clockers, and you would really need to be playing @ 1080p to see some results. You'll be lucky to get 1fps @ 4k going from 6k-8k MT/s.

Edit: We need @ir_cow he will know

I know........it's just that even at today's prices it's still my one 64GB upgrade, a good chunk o change and i'd like to not have to get another kit down the line. Seeing others in the intel 24/7 thread do 8000 on their regular A-die kits is causing me a lot of angst LOL, might not hit it today but who knows? Really don't feel great paying the same price for something with samsung-levels of potential

Should have closely examined them in-store. Not like the g.skill packaging is trying to hide anything



Ahh...never mind. Totally slipped my mind that all the 8000+ results are done on 16GB and 24GB density. Guess I'll be keeping my ripjaws.

 

[ir_cow]

I've been summoned lol.

M-Die (2GB) is inferior to A-Die (2GB) in just about ever way. I get lower tFAW values with M and maybe some other secondary by 1-2 ticks. But it does require more voltage for the same results.

If we are talking about just highest frequency in general. A-Die (2GB) wins, second to M-die (3GB) for the same voltages.

Now start to factor in the real world. First you need a Motherboard that can do 8000+ and a CPU as well. Switch to dual-rank and it's less about the motherboard because you'll max out at 6800-7200 on the Intel IMC.

So the question is where are you trying to go with this memory? I think M-die is perfectly fine for 6000-6400 range. Put a fan on it and let rip at 1.5v!

 

[tabascosauz]

I've been summoned lol.

M-Die (2GB) is inferior to A-Die (2GB) in just about ever way. I get lower tFAW values with M and maybe some other secondary by 1-2 ticks. But it does require more voltage for the same results.

If we are talking about just highest frequency in general. A-Die (2GB) wins, second to M-die (3GB) for the same voltages.

Now start to factor in the real world. First you need a Motherboard that can do 8000+ and a CPU as well. Switch to dual-rank and it's less about the motherboard because you'll max out at 6800-7200 on the Intel IMC.

So the question is where are you trying to go with this memory? I think M-die is perfectly fine for 6000-6400 range. Put a fan on it and let rip at 1.5v!

Good to know. Does this all still hold true for dual rank (2x32)?

I'm leaning towards giving the M-die a pass since I watched bz say that 2x32 will reduce A-die's headroom significantly. I think I will make a call to the store tomorrow morning and have them check to see if the 6400/32 white ripjaws is 820A - if so I might go back and exchange them since I haven't opened anything.

I was thinking of having a little fun on the latest agesa with 1:2 and seeing what it could do, if I had A-die. But again, not sure how 2x32GB will affect that. Of course, after that will probably just settle down at 6000 and tighten things down.

Board will be a B650E PG-ITX, so nothing too out of the ordinary for upper midrange ITX (10layer).

 

[dgianstefani]

I disagree that you need to play at 1080p to see results from better RAM.

1440p and even 4K monitors these days have 240 Hz + options. Besides, fast RAM improves the lows as well as the highs, so having 1%/0.1% lows closer to your average is a nice experience and will literally help you play better in competitive games.

(Driver limited to 237 FPS for latency reasons)

Considering you're talking about a new kit, finding A die may be worth it. If you were talking about upgrading an existing kit you'd already installed it would be a waste of time.

Then again, personally I use M-Die and just run it at the bleeding edge, while still being synthetically stable. So it's up to you what approach you prefer. Regardless, you have to be on Intel for A die to shine, because it's frequency optimised much more than the other dies, and Intel Raptor Lake is the only platform that can reach those frequencies without ridiculous gearing.

You have ITX so automatic advantage of a two DIMM motherboard. Why not push it?

This is M die.

With DDR5 you get more performance from tuning timings than you do from simple MT increases. Obviously improving MT at the same timings makes for better performance, but I'd pick a latency optimised tune over an MT optimized tune anyday. If all you're going to do is set XMP/EXPO, most of the high speed kits have sucky timings anyway, same as the low speed kits, but at least those are cheap.

Just two timings that when tuned give you more performance than untuned MT are TRFC and TREFI.

 

[ir_cow]

@dgianstefani has more experience tuning memory on AM5. I just do the quick and dirty most of the time. Not a lot of time to experiment so I don't have many answers. I would like to know of course.

 

[tabascosauz]

@dgianstefani has more experience tuning memory on AM5. I just do the quick and dirty most of the time. Not a lot of time to experiment so I don't have many answers. I would like to know of course.

Just on the density question then, on either intel/amd platform, how much of an impact to speeds/volts/timings do you generally see going from single to dual rank?

I don't imagine i'll go as ham as in the above, I have a bit of time when I'll have the new setup on my openbenchtable just to dial in CO and memory, while on iGPU and a spare ssd. But won't be spending too much time on it.

What's also confusing to me is everyone saying that new = A-die........mine are stickered for Jan 2024...

 

[ir_cow]

Just on the density question then, on either intel/amd platform, how much of an impact to speeds/volts/timings do you generally see going from single to dual rank?

Timings don't change at all besides the ones related to temperature. tRFC and tREFI are examples. Two sided, means more heat output.

Voltage is the same per die type as far as I can tell for a given frequency.

Frequency is limited by die type and IMC. If I can do 8200 on single, I can only do 6800 for dual using the same CPU mem voltages (SA,TX,VDD2).

What's also confusing to me is everyone saying that new = A-die........mine are stickered for Jan 2024...

M-Die (2GB) is still in production I think. Mostly found in 6000 and lower because you don't need 1.45v for that.

 

[Taraquin]

Max speed for M-die is usually 7200-7600, for A-die it seems any kit can do 8000+. In theory dual rank can do that aswell, but it is way harder to run for the memorycontroller. If you are lucky 7600 in 2:1 may work if you have a 2dimm motherboard or a good 8-layer 4dimm. A-die can run RFC quite a bit tighter than M-die (80-100), but except for that they are quite similar timingswise. If you plan on holding on to that ram for years and maybe swap cpus without having to swap ram then I would go A-die. If you usually swap ram and cpu at the same time anyways there is less of a reason. You can get a measly 1-2% better performance when cpu bound if you tune RFC to max, but except for that little to gain.

 

[ThrashZone]

Hi,
Doubt you'd get different die fishing on the same series memory
Think you'd get net gain of wasted funds that way.

Trident-z or stay home with what you got hehe

 

[evernessince]

If you are just running EXPO it's unlikely A die would make a difference vs M die. As other's have said, A die is mostly if you are going in tweaking a lot. A die helped me get 100% stability with my tuned subtimings for example. There's a fringe scenario where if you have a terrible IMC on your CPU that can't even do 6000 with other EXPO settings it might help but I'd stress that as fringe and unlikely. People should be running MemTest86 with EXPO / XMP enabled on any new kit they get first thing to eliminate potential future surprises.

 

[tabascosauz]

Timings don't change at all besides the ones related to temperature. tRFC and tREFI are examples. Two sided, means more heat output.

Voltage is the same per die type as far as I can tell for a given frequency.

Frequency is limited by die type and IMC. If I can do 8200 on single, I can only do 6800 for dual using the same CPU mem voltages (SA,TX,VDD2).

M-Die (2GB) is still in production I think. Mostly found in 6000 and lower because you don't need 1.45v for that.

roger that, 8200>6800 is roughly in line with the kind of dual rank penalty I was reading about. If that is expected across the board, then it looks like I would be going nowhere fast on A-die too. Remembering the whole debacle about high VSOC

If you plan on holding on to that ram for years and maybe swap cpus without having to swap ram then I would go A-die. If you usually swap ram and cpu at the same time anyways there is less of a reason.

Yup, this is exactly the crux of it. I guess being 2x32 evens the playing field a bit and A-die wouldn't be quite as strong as I thought it would be.

Do you reckon 6400 32-39-39 would get a better bin, for better performance tightening at the same 6000? Or are these all pretty much the same lower end Hynix bins

Hi,
Doubt you'd get different die fishing on the same series memory
Think you'd get net gain of wasted funds that way.

Trident-z or stay home with what you got hehe

I was eyeing the 6400CL32 kit more than the FlareX/Ripjaws 6000CL30 because only the former is also white like the current kit. Technically the two are kinda equivalent on latency but not quite, the 6400 kit is 32-39-39 while the three 6000 kits are all 30-40-40

Trident Z - hell no! Not doing G.skill RGB ever again. What a freaking nightmare ripjaws/flare for a very good reason lol.

Also, am not decided on keeping FC140, getting white FC140, or moving to PS120. The 120mm towers want lower height DIMMs so the ripjaws height is much appreciated

 

[ThrashZone]

Hi,
OpenRGB off

 

[danc]

Gskill is bad because no thermal contact on the Pmic.
Switch to cheaper Tcreate Experts with 10L PCB and thermal contacts.
Only bummer is Teamgroup software washed A die recognition, so you cannot use Hwinfo and the likes to readout if they are A die.
But 64GB 6400MT is safely A die.

 

[tabascosauz]

Gskill is bad because no thermal contact on the Pmic.
Switch to cheaper Tcreate Experts with 10L PCB and thermal contacts.
Only bummer is Teamgroup software washed A die recognition, so you cannot use Hwinfo and the likes to readout if they are A die.
But 64GB 6400MT is safely A die.

Is that an actual issue at AM5 1:1 speeds though? The Reddit accounts where this is a xmp stability problem seem to be in some absolutely appalling airflow conditions (50c at idle). Even if SFF conditions led to those temps I wouldn't let them run like that and just put a fan on them.

T-creates are 322 on Amazon for 6400 34-44-44, and extended fulfillment time from outside Canada. Not the 289 price listed on pcpartpicker.

 

[ir_cow]

roger that, 8200>6800 is roughly in line with the kind of dual rank penalty I was reading about. If that is expected across the board, then it looks like I would be going nowhere fast on A-die too. Remembering the whole debacle about high VSOC

Well that's the current on Intel. AMD is all about staying within that 1:1 ratio, so 6400 is probably where you will land. But you could get even better results by keeping the FLCK in step.

@dgianstefani has his tuned up

 

[dgianstefani]

Faster bins generally tune down better. So getting a fast A die then running it tuned at <6400 MT will still have better results than a 6400 MT kit tuned. Especially if you want to guarantee you're getting A die.

Don't bother with expo specific kits entirely unnecessary and generally more expensive. Cheapest kit with fastest timings/MT stock, with a thermal pad on PMIC.

Bear in mind MT guarantees from AMD drop off sharply as you add memory ranks.

At the upper limit of ram capacity they only guarantee 3600 MT.

At which point you're crippling the system and everything that runs using the infinity fabric to get capacity.

 

[tabascosauz]

Well that's the current on Intel. AMD is all about staying within that 1:1 ratio, so 6400 is probably where you will land. But you could get even better results by keeping the FLCK in step.

@dgianstefani has he's tuned up

Yup yup. I know. Just getting a sense for more realistic expectations on what 2x32 will do, rather than what 2x16 can do

The store just confirmed that the 6400/32 kit is 821A, so I'm gonna head over and swap these two kits.

Faster bins generally tune down better. So getting a fast A die then running it tuned at <6400 MT will still have better results than a 6400 MT kit tuned. Especially if you want to guarantee you're getting A die.

Don't bother with expo specific kits entirely unnecessary and generally more expensive. Cheapest kit with fastest timings/MT stock, with a thermal pad on PMIC.

Bear in mind MT guarantees from AMD drop off sharply as you add memory ranks.

At the upper limit of ram capacity they only guarantee 3600 MT.

At which point you're crippling the system and everything that runs using the infinity fabric to get capacity.

Seems like 6000 at the least should still not be a problem at all for single CCD dual rank (? That's all I really care about at the moment, the 1:2 stuff would just be fun tinkering.

64GB is the whole point of this adventure so it's not worth stepping down to 2x16 just to let A-die stretch its legs.

 

[dgianstefani]

Yup yup. I know. Just getting a sense for more realistic expectations on what 2x32 will do, rather than what 2x16 can do

The store just confirmed that the 6400/32 kit is 821A, so I'm gonna head over and swap these two kits.



Seems like 6000 at the least should still not be a problem at all for single CCD dual rank (? That's all I really care about at the moment, the 1:2 stuff would just be fun tinkering.

64GB is the whole point of this adventure so it's not worth stepping down to 2x16 just to let A-die stretch its legs.

It's quad rank. Two dual rank sticks.

Good that they confirmed it. Hopefully it's true.

 

[Durhamranger]

Faster bins generally tune down better. So getting a fast A die then running it tuned at <6400 MT will still have better results than a 6400 MT kit tuned. Especially if you want to guarantee you're getting A die.

Don't bother with expo specific kits entirely unnecessary and generally more expensive. Cheapest kit with fastest timings/MT stock, with a thermal pad on PMIC.

Bear in mind MT guarantees from AMD drop off sharply as you add memory ranks.

At the upper limit of ram capacity they only guarantee 3600 MT.

At which point you're crippling the system and everything that runs using the infinity fabric to get capacity.

Never had a problem with 4 x 16gb dimms myself....

 

[ir_cow]

Never had a problem with 4 x 16gb dimms myself....

4x single rank is different than 4x dual. Last attempt with 4x single, I managed 6200 stable.
Highest I could go with 4x dual is 4800. Both before AGESA 1.0.7

Officially AMD only says 3600 is guarantee to work for either.

in either case your results and mine are 100% not plug n play.

 

[tabascosauz]

Never had a problem with 4 x 16gb dimms myself....

4x single rank is different than 4x dual. Last attempt with 4x single, I managed 6200 stable.
Highest I could go with 4x dual is 4800. Both before AGESA 1.0.7

Officially AMD only says 3600 is guarantee to work for either.

in either case your results and mine are 100% not plug n play.

$660 flagship board...I sure hope it has no issues running 6400

I was originally planning on getting the A620I Lightning to save a buck. But the order ran into some issues and the rear I/O is also REALLY lacking, so the B650E will have to do. Not incredibly enthused about $350, but it's about the same as my AM4 Strix ITX was so (the one about to receive the handmedown 5800X3D)...

It's quad rank. Two dual rank sticks.

Good that they confirmed it. Hopefully it's true.

I mean, yes, but for a couple years now that's only used to refer to 4 ranks per channel, because everyone and their mother running DR B-die would have been "quad rank" which wasn't the case

Anyways thanks everybody for the help. Got the A-die.

 

[ir_cow]

The A620 suck ass... Not impressed. The Gaming X (which I never completed for review) could only do 5600 in 1:1 for single rank.

It's fine for budget computer. Budget CPU, mem , and GPU. But at that point might as well get a gaming console.

 

[tabascosauz]

The A620 suck ass... Not impressed. The Gaming X (which I never completed for review) could only do 5600 in 1:1 for single rank.

It's fine for budget computer. Budget CPU, mem , and GPU. But at that point might as well get a gaming console.

Most of the ITX A620s looked fine to me. 8-layer fare, comparable to some of the midrange full size stuff. ASRock is doing that PCB sharing thing so the A620I and B650I Lightnings basically are the same. Being newer boards both have a pretty recent QVL with 7200 advertised. Unlike Gigabyte's A620, VRM looked fine for single CCD, 8 x 60A SPS or something. Could have saved enough money to buy a whole separate 2x16 Hynix kit too.

It was the supposed A620 limitations that I didn't want to play the lottery on. No sense in losing performance for no reason other than because AMD said so.