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

Hynix A or M die for "futureproofed" AM5 build?

Status
Not open for further replies.
There is no future proofing in PC hardware. Buy what will serve your needs today, be prepared to replace it when it no longer serves those needs.
No one seems to note the quotes... ;)
 
Searching the interwebs for DDR5 memory advice I've seen most people recommend modules using Hynix chips (and surprisingly recommend staying away from Micron, apparently :eek:). However, there seems to be a difference of opinion on whether Hynix A or M dice are the better choice. Apparently, A dice can clock higher while M dice have tighter timings.

Since current AMD CPUs cannot use higher clocked RAM as well as Intel ones, many people recommend going with M dice for AM5. But what about future AM5 CPUs (for instance Zen6/"Medusa")? Would it not make sense to prefer A dice, since Zen5 already has somewhat better memory support (as far as the current IO die can handle it) and Zen6 is sure to greatly improve on that, considering it'll get a brand new IO die?

I'm looking for memory that'll serve me well now, and, hopefully, even better once Zen6 comes around. Bonus question: Is A or M the newest iteration?
- There are 16gb M-dies, avoid these, they max out around 7000-7400. Rfc at 160ns-ish, RP similar to RCD.
- 16gb A-die have best timings and do 8000+, they run very low RP and RFC, similar to 16gb M-dies on other timings. Rfc at 120ns-ish, RP 2-6 below RCD.
- 24gb M-dies have similar timings to 16gb M-dies, but can run 10 000+. All world record with speed is set if 24gb M-dies if I remember correctly.
 
There is no future proofing in PC hardware. Buy what will serve your needs today, be prepared to replace it when it no longer serves those needs.
The AM5 platform was future proofed when we know AMD will release another generation of processors on it.

That is the only future proofing we get though.


 
I think that topic is only for opinions.

I'm happy that i bought 2x 32GIB Hynix A DDR5 modules. I'm still on 1.25V for the DRAM - like the two stock profiles. I bought 5600 MT/S stock profile DRAM which I clock at 6000 MT/s with a bit tighter timings. Later many DRAM modules use again 1.35V or maybe even 1.45V for DDR5.

I would go for bigger modules. 2x48GiB was not available at that time.

Code:
*-bank:1
             description: DIMM Synchronous Unbuffered (Unregistered) 4824 MHz (0.2 ns)
             product: KF556C36-32
             vendor: Kingston
             physical id: 1
             serial: ?
             slot: DIMM 1
             size: 32GiB
             width: 64 bits
             clock: 529MHz (1.9ns)

        *-bank:3
             description: DIMM Synchronous Unbuffered (Unregistered) 4824 MHz (0.2 ns)
             product: KF556C36-32
             vendor: Kingston
             physical id: 3
             serial: ?
             slot: DIMM 1
             size: 32GiB
             width: 64 bits
             clock: 529MHz (1.9ns)

I think bigger DRAM modules will keep more their monetary values than smaller ones, like 16GiB.
 
What's the reason?
Their chips just have piss poor perfomance vs the competition this gen. They just don't clock as high and have looser latencies on ddr5. Their product is solid if you don't care about OCing though.
 
CUDIMM is the future, the question is when will AMD support this and if Zen6 supports 10,000+ speeds. Otherwise getting as close to 6400 CL28 is the goal for now.
What about 6400 MT and higher latency? Would that be vastly preferable to 5600 MT and slightly lower latency? For something like a 25% markup? I wonder if they're the same modules or if the 6400 MT ones are actually much higher binned. Right now I'm trying to decide between this kit and this kit. And yes, they're vastly more expensive than regular DDR5 kits. :)
 
One of the best kits I have tested so far on X670E Gene with 9950X is the G.SKILL 2x16GB A-Die Z5 Royal Neo DDR5-6000CL26. The kit works fine in EXPO, or can be overclocked depending on your CPU IMC to 6400, 6600 and 8000, to 8400 with tight timings on my system.


Screenshot 2025-02-18 054140.png
 
What about 6400 MT and higher latency? Would that be vastly preferable to 5600 MT and slightly lower latency? For something like a 25% markup? I wonder if they're the same modules or if the 6400 MT ones are actually much higher binned. Right now I'm trying to decide between this kit and this kit. And yes, they're vastly more expensive than regular DDR5 kits. :)
When did ECC come into the picture?
 
When did ECC come into the picture?
For me? Right from the start. I just didn't want this thread to get tainted by comments on how speed and ECC cannot coexist. I never buy anything else.
 
For me? Right from the start. I just didn't want this thread to get tainted by comments on how speed and ECC cannot coexist. I never buy anything else.
Yet this is the first mention you've made of it in this thread, and it changes the whole equation, because I'm utterly baffled why you're interested in high clocks and low latencies on ECC because that's not the point of ECC. It's not about "speed and ECC cannot coexist", it's that they don't because nobody bothers making fast ECC modules because, again, that's entirely missing the point of ECC.

I assume you're also unaware that ECC memory support on Zen 4/5 motherboards is also hit and miss?
 
Yet this is the first mention you've made of it in this thread, and it changes the whole equation, because I'm utterly baffled why you're interested in high clocks and low latencies on ECC because that's not the point of ECC. It's not about "speed and ECC cannot coexist", it's that they don't because nobody bothers making fast ECC modules because, again, that's entirely missing the point of ECC.

I assume you're also unaware that ECC memory support on Zen 4/5 motherboards is also hit and miss?
Aaaaaand there we have it. This is exactly what I was trying to avoid. If you at least had anything productive to contribute. Lecturing me on ECC memory isn't it. Sorry to be so blunt.
 
What about 6400 MT and higher latency? Would that be vastly preferable to 5600 MT and slightly lower latency?
You do understand to compare latency you should calculate the timing in nanoseconds so you're comparing apples to apples. If the 6400MT/s is even slightly slower in latency I would still pick this option for the greater throughput although I have doubts AM5 will actually drive them at that speed, but 6000MT/s should be doable. Whether or not it's worth the extra premium in price is up to you. A downside is if you decided to fill all 4 dimms your speed will likely need to be reduce dramatically (making the high speed dimms possibly a waste of money) but I think you know that already.
For something like a 25% markup? I wonder if they're the same modules or if the 6400 MT ones are actually much higher binned. Right now I'm trying to decide between this kit and this kit. And yes, they're vastly more expensive than regular DDR5 kits. :)
Whoa plot twist but I admit I was scratching my head about your OP since you didn't mention ECC on the first post. Anyway I would never have thought to look at vcolor for ECC thanks for the links. On e-bay I've been eyeballing Mem-Store for ECC modules for a bit cheaper ($488 for 128GB at 4800MT/s advertised with Asrock MB compatibility) but am unfamiliar with the quality of the vendor.
 
Last edited:
Nice attitude ... ECC .... .... ....

I doubt I read ECC yesterday whne I read the hole stuff. And than this behaviour.

For some reason I found recently the Unwatch or Watch button on top right on a topic. Worth testing the unwatch feature now
 
For me? Right from the start. I just didn't want this thread to get tainted by comments on how speed and ECC cannot coexist. I never buy anything else.
In the past I OC'ed my DDR4 2666MT/s modules to successfully 3200MT/s for my Zen3 chips but it can be hit or miss per module since their not binned for that. I got lucky in that my Microns only needed a voltage bump to do it with minor timing adjustments. I don't bother with it anymore these days because the time it takes to test and the added uncertainty when something goes wrong isn't worth it. Having said that the OC method is what I used to determine if the ECC is working and the OS is logging errors.
 
Aaaaaand there we have it. This is exactly what I was trying to avoid. If you at least had anything productive to contribute. Lecturing me on ECC memory isn't it. Sorry to be so blunt.
Here's something blunt in return:

You asked a question, deliberately omitted information thereby invalidating the input that other people provided and thus wasting their time, and you have the temerity to get all high and mighty when you get called out for your BS.

Over and above that, the question itself in the context of ECC is itself a waste of time because asking about ECC and overclocking in the same sentence is just plain stupid, because overclocking ECC memory defeats its whole point. It's like wanting to buy an airplane based on how fast it drives on the runway.

I'm out, and you can go on my ignore list for being literally insane.
 
Buy what ever you can afford now and in 6 months you wont even remember what you have
 
The general consensus seems to be that Hynix is best and Micron is worst for DDR5, with Samsung somewhere in the middle. I bet for stock JEDEC speeds it's probably a toss-up.

Yeah for JEDEC speeds its never going to matter. For OCing or EXPO just forget about Micron entirely.
What gets me is that months later, Micron has still not fixed their shit. How?! Why?!

They’re one of the largest memory manufacturers in the world, how have they not fixed their issues?
 
Nice attitude ... ECC .... .... ....

I doubt I read ECC yesterday whne I read the hole stuff. And than this behaviour.

For some reason I found recently the Unwatch or Watch button on top right on a topic. Worth testing the unwatch feature now
What issue do you have with my attitude? Like I wrote, I specifically did not state I was looking for ECC memory because of the inevitable comments about how speed and ECC functionality are mutually exclusive. And as you can plainly see I was proved right. Sadly.

What I'm looking for is the best memory I can get. I had already boiled it down to Hynix. Whether or not I'm interested in ECC is completely beside the point.

ECC memory is usually about 2-3% slower than non-ECC memory at the same specs. Latencies aside, 6400 MT/s at 1.1v for the most expensive kit seems quite speedy to me. AFAIK not many "gamer" kits (can?) do that.

Here's something blunt in return:

You asked a question, deliberately omitted information thereby invalidating the input that other people provided and thus wasting their time, and you have the temerity to get all high and mighty when you get called out for your BS.

Over and above that, the question itself in the context of ECC is itself a waste of time because asking about ECC and overclocking in the same sentence is just plain stupid, because overclocking ECC memory defeats its whole point. It's like wanting to buy an airplane based on how fast it drives on the runway.

I'm out, and you can go on my ignore list for being literally insane.
Nice. At which point did I *ever* state I was looking to OC the memory? Show me where. For the record: I'm not sure how old you are, but I've been using ECC memory for the last 30 years. Back when it was mostly referred to as parity memory.

Buy what ever you can afford now and in 6 months you wont even remember what you have
To be honest, I can afford the priciest kit. Just not 100% convinced if it is worth the money. Seems like prices for the few ECC DDR5 UDIMM modules/kits are almost triple that of "regular" DDR5 memory. That's in stark contrast to ECC DDR4. Of course DDR4 is a much older iteration at this point, but even DDR5 is no longer brand-spanking-new.
 
What about 6400 MT and higher latency? Would that be vastly preferable to 5600 MT and slightly lower latency? For something like a 25% markup? I wonder if they're the same modules or if the 6400 MT ones are actually much higher binned. Right now I'm trying to decide between this kit and this kit. And yes, they're vastly more expensive than regular DDR5 kits. :)
I have a feeling the 6000CL26 and 6400CL28 are the same quality bin. If your into memory overclocking, get really cheap 5600 A-Die KD5AGUA80-56G460D and drop it to CL24. It will need a fan and 1.5v, but it will be half the price.

AMD rules is highest frequency possible and lowest latency in Gear 1 mode. That means for 99% of users the limit is 6400 for 7000/9000 series.

Will you see the difference from 6000CL26 to CL32? Probably not. Mostly depends on the game and what is acceptable to you in 1% lows.

One of the best kits I have tested so far on X670E Gene with 9950X is the G.SKILL 2x16GB A-Die Z5 Royal Neo DDR5-6000CL26. The kit works fine in EXPO, or can be overclocked depending on your CPU IMC to 6400, 6600 and 8000, to 8400 with tight timings on my system.


View attachment 385442
I just got the RGB NEO version Friday. On my test bench now. Reminds me of DDR4-4000 CL15. High bin, but pricey.
 
To be honest, I can afford the priciest kit. Just not 100% convinced if it is worth the money. Seems like prices for the few ECC DDR5 UDIMM modules/kits are almost triple that of "regular" DDR5 memory. That's in stark contrast to ECC DDR4. Of course DDR4 is a much older iteration at this point, but even DDR5 is no longer brand-spanking-new.
If you have the money and you're only going to populate 2 dimms I would go for the faster kit to get the more optimal CPU utilization by being in the MT/s sweet spot for Zen 4/5.
 
What gets me is that months later, Micron has still not fixed their shit. How?! Why?!

They’re one of the largest memory manufacturers in the world, how have they not fixed their issues?
I think their focus is the OEM market more now so not a big priority for then.
 
I have a feeling the 6000CL26 and 6400CL28 are the same quality bin. If your into memory overclocking, get really cheap 5600 A-Die KD5AGUA80-56G460D and drop it to CL24. It will need a fan and 1.5v, but it will be half the price.

AMD rules is highest frequency possible and lowest latency in Gear 1 mode. That means for 99% of users the limit is 6400 for 7000/9000 series.

Will you see the difference from 6000CL26 to CL32? Probably not. Mostly depends on the game and what is acceptable to you in 1% lows.
I'm not really into memory overclocking (or overclocking in general). That doesn't mean I don't want as much speed as possible, even if it's ECC memory. And since I plan to upgrade to Zen 6 eventually (unless that turns out to be impossible) I would like for the memory not to be outdated specwise. Assuming the Zen 6 IMC is at least as capable as the current ones.

Thanks for the info about AMD memory sweet spots. I'm still speculating whether or not the 5600 CL 46 and the 6400 CL 52 are the same bin, considering the latter costs $100 more ($499 opposed to $399).
 
I'm not really into memory overclocking (or overclocking in general). That doesn't mean I don't want as much speed as possible, even if it's ECC memory. And since I plan to upgrade to Zen 6 eventually (unless that turns out to be impossible) I would like for the memory not to be outdated specwise. Assuming the Zen 6 IMC is at least as capable as the current ones.

Thanks for the info about AMD memory sweet spots. I'm still speculating whether or not the 5600 CL 46 and the 6400 CL 52 are the same bin, considering the latter costs $100 more ($499 opposed to $399).

6400 CL 52 is slightly faster but that's only taking into account 1 timing. Also don't forget memory training may optimize your timings a bit. For example my RAM was supposed to be CL46 but UEFI optimized it to CL44 after picking the profile.

Snag_24258fed.png
 
Last edited:
6400 CL 52 is slightly faster but that's only taking into account 1 timing. Also don't forget memory training may optimize your timings a bit. For example my RAM was supposed to be CL46 but UEFI optimized it to CL44 after picking the profile.

View attachment 385502

The full specs are these for the interesting two 2x 48GB kits. Price difference is $100, as stated above. I'm still baffled that they've managed to eke out 6400 MT/s at 1.1v.
5600MHzPC5-44800CL-46-45-45-901.1V
6400MHzPC5-51200CL-52-52-52-1031.1V
 
Status
Not open for further replies.
Back
Top