Question Very basic Question about 32 vs 64 GB RAM

[Caveman]

Currently building a RIG with the following core components:

1) ASUS ROG STRIX B650E-E GAMING WiFi Socket AM5 (LGA 1718) Ryzen 7000 gaming motherboard(16 + 2 power stages, DDR5, 4x M.2 slots, PCIe® 5.0, WiFi 6E, 2.5G LAN, USB 3.2 Gen 2x2 Type-C® rear I/O port, Aura Sync RGBc)

2) AMD Zen 4 7700X 5th Gen 8-core, 16-threads

3) 4090 FE

4) CORSAIR Vengeance RGB 32GB (2 x 16GB) 288-Pin PC RAM DDR5 5600 (PC5 44800) Desktop Memory Model CMH32GX5M2B5600Z36K
16-16-16-76 CAS 36

After reading around, I'm not convinced that adding another 2 x 16GB of RAM will make this new build faster or slower and if 2 x 32 GB would have been better for a 64GB goal. I remember that AMD DDR4 had specific configs that worked better than others but haven't found any clear direction for DDR5. Can someone help or point me to some resources that are eaasy to understand and confident in their assessment? i.e. 2 banks of 2 x 16 is better than 1 bank for 2 x 32, etc... Benchmarks would be nice.

Steve from Gamers Nexus had a perfect video. Wondering if these rules still hold for new AMD series...

 

[Turbonium]

Short answer: just get 2x16GB and you are good.

I get that, but I'm really just curious.

I'm not as knowledgeable as many of you on this forum (not even close), but I enjoy learning about the technical stuff.

 

[igor_kavinski]

DDR5 Demystified - Feat. Samsung DDR5-4800: A Look at Ranks, DPCs, and Do Manufacturers Matter?

www.anandtech.com

Doesn't make a huge difference.


Unless you use a VERY memory sensitive app, like WinRAR.

 

[kschendel]

Wait, so what's "better", single-rank or dual-rank sticks?

DDR4 shows a definite, if small (5% give or take) improvement with dual-ranking, whether it's done with 1 dual-rank stick per channel, or 2 single-rank sticks per channel. The former is easier on the memory controller and you might, maybe, be able to hit higher rates / better timings.

DDR5 is sufficiently different that I dunno. The above post suggests that it's no big deal for DDR5, although I'd have to RTFA to see if they are really testing single vs dual rank.

 

[HardWarrior]

32GB is enough for 95% of all normal applications(excluding scientific, complex mathematics, database, graphics image rendering,etc…).

Ultimately this depends on HOW you run these potentially memory-hungry apps. For me (hardcore gaming and recreational browsing) 32GB has been, and will be enough for the foreseeable future.

 

[q52]

DDR5 Demystified - Feat. Samsung DDR5-4800: A Look at Ranks, DPCs, and Do Manufacturers Matter?

www.anandtech.com

Doesn't make a huge difference.

View attachment 74184
Unless you use a VERY memory sensitive app, like WinRAR.

This. If you're already waiting 33 seconds for a task to complete who gives a crap about waiting 4 more seconds?


Not sure I've ever seen a real world situation where memory speed makes any noticeable difference.

 

[q52]

2 vs 4 depends a lot on the memory controller built into the CPU. It also depends to a certain extent on the motherboard: trace quality, routing, shielding, BIOS timing setup. 4 sticks is a more difficult load to drive, electrically.

For DDR4, Intel CPU's up to 11th gen have robust memory controllers and seem quite happy with 4 sticks. 12/13th gen, I've heard conflicting stories. The first 2 Ryzen generations did not care for 4 sticks at all; Zen 2 (Ryzen 3000) is kind of hit or miss, and Zen 3 (Ryzen 5000) is mostly OK with pushing 4 sticks.

Current DDR5 memory controllers don't like 4 sticks AT ALL from what I hear; you need to drastically drop transfer rates to get them to work, whether Intel or AMD. For your DDR5 setup I would definitely advocate sticking to 2 sticks even if it means replacing/selling what you have. Of course, if you have a dire immediate need for 64GB and are money constrained, 64GB at a reduced speed may well be better than nothing.

This sounds like a bunch of hand-waving nonsense. What the heck is that even supposed to mean, that some CPU "likes" or "doesn't like" some memory configuration? Either it works or it doesn't. Excessively high memory clock speeds have always been a niche thing with variable support and questionable benefit. I'm having a hard time believing that it makes any real world difference between single and dual rank and double or quad or single memory sticks. Just get normal speed memory in a configuration that fits your budget and upgrade path and be done with it. Who cares if it takes you six extra seconds to unzip an archive once every three months? You think you're actually gonna notice a visual difference between gaming at 150fps and 140fps especially when your monitor only runs at 60Hz or 120Hz?

One of the things that actually will bite you in the butt later is deciding to upgrade and realizing you filled all the memory slots with low capacity sticks and now instead of adding two more modules you need to replace all four to get your target memory capacity. In all my years of upgrading and building computers that's about the only thing that memory configuration has affected.

 

[kschendel]

This sounds like a bunch of hand-waving nonsense. What the heck is that even supposed to mean, that some CPU "likes" or "doesn't like" some memory configuration? Either it works or it doesn't.

That's true for a specific sample. For a class such as Zen+ you have to talk statistics. It's pretty well understood by most engineers that "like" means something along the lines of "has a reasonably good chance of working". I don't have exact statistics and I doubt anyone else does, not even within AMD. Yes, it's handwaving to a degree, but no, it's not at all nonsense.

As for the rest of it, dual ranking has been shown to result in performance increases typically around 5%, give or take, depending on what you're doing. I passed no judgement on whether that is real world noticeable.

 

[BonzaiDuck]

I'd say the advice here is pretty accurate. There's only one exception, or the reason I went beyond 16GB (2x8) in the first place. I have to use all four slots for DDR4, and someone here said that DDR5 isn't up to the task yet.

So I've socketed 64GB or 4x16GB into my board. For most applications I need or use, it's still going to be fine with 16GB, but I use the remainder for caching NVME and SATA storage to RAM. The forums on Romex (Primocache) seem to indicate that the program continues to work well with Optane in the mix.

I wouldn't know what is or isn't obsolete with recent chipsets like Z690, but I'm not salivating over latest-gen NVME (7,000 Mb/s sequential throughput) or desperate for the latest hardware with my systems, primarily because of the PrimoCache deployment and bonus in using lots of RAM for it. Most stable systems I've ever had, barring a static-charge disaster that borked a motherboard.

I"m even using the PrimoCache server version on my Windows 2012 R2 box . . . .

 

[sdifox]

I'd say the advice here is pretty accurate. There's only one exception, or the reason I went beyond 16GB (2x8) in the first place. I have to use all four slots for DDR4, and someone here said that DDR5 isn't up to the task yet.

So I've socketed 64GB or 4x16GB into my board. For most applications I need or use, it's still going to be fine with 16GB, but I use the remainder for caching NVME and SATA storage to RAM. The forums on Romex (Primocache) seem to indicate that the program continues to work well with Optane in the mix.

I wouldn't know what is or isn't obsolete with recent chipsets like Z690, but I'm not salivating over latest-gen NVME (7,000 Mb/s sequential throughput) or desperate for the latest hardware with my systems, primarily because of the PrimoCache deployment and bonus in using lots of RAM for it. Most stable systems I've ever had, barring a static-charge disaster that borked a motherboard.

I"m even using the PrimoCache server version on my Windows 2012 R2 box . . . .

True geek cred. This man spent more on his ram than on his car.

 

[BonzaiDuck]

True geek cred. This man spent more on his ram than on his car.

Not really. I think for the second or "twin" system with the workstation board, I bought two kits of 2x16 Ripjaws more than a year ago. I'd have to look at the prices I paid, but I think the total was under $270. It was still worth it to me to have the extra expanded caching to RAM.

I've mentioned my 28-year-old Trooper in these threads as opposed to "Social->Garage". I budget $1,000/year for maintenance and repair, but I expect to get by with less. It all hinges on the longevity of subsystems and parts. I keep a spreadsheet going back to 2002. The latest little "crisis" was a dead starter-motor -- easy to diagnose. They still have OEM parts available -- two of four different manufacturers. The Trooper came back yesterday with a $650 bill for parts and labor.

Some people would lose patience with this sort of thing. I ordered the best part -- Denso. The cost for replacement in 2003 was $533, and it's been 20 years. Figure an annual inflation-rate of 1.5%. 1.015^20 x 533 = $718. I'm good with that. What else could happen this year? probably nothing. Better than a car-payment on low-mileage pre-owned Prius/RAV4 hybrid, less the $120/mo gas savings, plus the full insurance!

What leaks in the wallet do I really worry about? Tooth repair! And I have dental insurance, even so.

Always Blush and Fross daily.

 

[coercitiv]

DDR5 is sufficiently different that I dunno. The above post suggests that it's no big deal for DDR5, although I'd have to RTFA to see if they are really testing single vs dual rank.

For DDR5 we're no longer interested that much in DR vs SR, since DDR5 has more bank groups than DDR4 in their respective worst DIMM configuration. A quick comparison and explanation can be found here.

AFAIK the basic rule of thumb for the average consumer looking to build a DDR5 system is to buy 2x16GB as the minimum and probably optimum configuration. Those who want more than 32GB of RAM should increase DIMM size and aim to stay with 2 sticks of memory. Those who need even more memory are probably seasoned professionals who don't need advice on this and who understand the tradeoffs involved.

 

[kschendel]

For DDR5 we're no longer interested that much in DR vs SR, since DDR5 has more bank groups than DDR4 in their respective worst DIMM configuration. A quick comparison and explanation can be found here.

I don't suppose you could point me at a transcript or equivalent text somewhere? I don't have enough time left on earth to sit through a buildzoid video, as informative as they may be.

 

[coercitiv]

I don't suppose you could point me at a transcript or equivalent text somewhere? I don't have enough time left on earth to sit through a buildzoid video, as informative as they may be.

https://www.micron.com/-/media/client/global/documents/products/white-paper/ddr5_new_features_white_paper.pdf

https://www.micron.com/-/media/client/global/documents/products/white-paper/ddr5_more_than_a_generational_update_wp.pdf

[LATER EDIT] One notable graph from the white paper bellow.

DDR5 can have more banks & groups in SR than DDR4 has in DR. It's worth noting the performance improvement when comparing DDR4-3200 DR vs. DDR5-3200 SR. There's some more gains to be had from going dual rank DDR5, but they'll just be offset by the loss in max memory clocks (and higher mem speed is easy to get in the consumer world).




All that being said, you might as well watch the Buildzoid video, the relevant part for our conversation is surprisingly fast, from 0:27 to 5:07, so less than 5 minutes total