Question How far should I bother pushing RAM? - Noob

[SyCoREAPER]

Id consider myself a noob when it comes to understanding over clocking for RAM, CPU I'm good, here * poof *.

Ryzen 5900HS (Asus Strix G15)

I've got two sets of RAM, timings are similar:
2x 18 - 3200
2x 32 - 2933

Here is the hitch, the laptop has a limited BIOS. I found a tool on GitHub, code is legit, seems to manipulate the BIOS into thinking faster RAM is installed. Confirmed Laptop and Windows are tricked.

I ran Cinebench Multi, the score for the overclocked 64GB is a tad faster (only 40 points). Right now I'm running a stress test in CB for stability.

At what point, if any, am I at the risk of damage or going to definitely run into issues? Should I try beyond 3200 or am I pushing my luck?

 

[Lafong]

What is the ultimate purpose of the machine?

Getting the highest benchmarks?

Something else? What?

If the former, I assume you'd keep pushing as you see fit as long as that remains your primary purpose. Continual experimentation, which I guess is what you are doing now.

 

[SyCoREAPER]

@Lafong

It's used for some older and moderately newer games and video transcoding (hence 64GB) . It only has a AMD 6800m GPU (roughly 6700 desktop) @ 1080 w/300hz display. If I can push it further and enough people think it's worth it I'll do it but I'm leary when it comes to RAM and how much difference it really makes with DDR4. Maybe I'm thinking too much in CPU terms but 300mhz seems like substantial jump on something thats XMP speed was already 2933.

But again, I'm not a RAM connoisseur.

I was/am happy at the 3200 mark but was reading that 3200 is the minimum for Ryzen (not sure how true that is). I'm not looking to flex with scores, just real-life PERCEIVABLE gains. If we are talking 1-2 frames that's not worth it to me. While the tool worked, I'm not sure how it's modifying the BIOS (despite reading the code) but am not sure how safe repeated rewrites are either.

 

[hotaru.hino]

There's no real life perceivable performance improvement going with faster RAM. Plucking one example from a site that looked at throwing in DDR4-3200 and DDR4-3600 and DDR4-3733:

AMD Ryzen RAM scaling - performance effect in games

The impact of memory timings and frequency on AMD Ryzen 3000 systems in games has been a topic of discussion. In this artilcle we'll zoom in on specifically that. See, AMD made a change in 3rd gener... Games performance – the influence of the Command Rate (1T vs 2T)

www.guru3d.com

AMD Ryzen RAM scaling - performance effect in games (Page 5)

The impact of memory timings and frequency on AMD Ryzen 3000 systems in games has been a topic of discussion. In this artilcle we'll zoom in on specifically that. See, AMD made a change in 3rd gener...

www.guru3d.com

There's almost zero improvement. It's well within the margin of error. When people say DDR4-3200 is the "minimum" for Ryzen, it's probably corrupted from the notion that Ryzen 5000 CPUs officially support up to DDR4-3200. So why settle for anything lower?

 

[SyCoREAPER]

When people say DDR4-3200 is the "minimum" for Ryzen, it's probably corrupted from the notion that Ryzen 5000 CPUs officially support up to DDR4-3200.

That's probably exactly it. BS Bandwagon that picks up steam, that's why I dont trust what I read and rather ask the community.

Thanks for those links. That is indeed pretty negligible. I ran the laptop (nothing heavy besides the stress test) and also have no WHEA's. I think I'll just keep it at the overclocked 3200 and be happy with that and not tinker or obsess over it further.

 

[hotaru.hino]

That's probably exactly it. BS Bandwagon that picks up steam, that's why I dont trust what I read and rather ask the community.

Thanks for those links. That is indeed pretty negligible. I ran the laptop (nothing heavy besides the stress test) and also have no WHEA's. I think I'll just keep it at the overclocked 3200 and be happy with that and not tinker or obsess over it further.

Ryzen 5900H's are also officially specc'd for DDR4-3200 (https://www.amd.com/en/products/apu/amd-ryzen-9-5900hs)

Note that while it says it supports up to LPDDR4-4266, this is a different type of RAM.

 

[SyCoREAPER]

Ryzen 5900H's are also officially specc'd for DDR4-3200 (https://www.amd.com/en/products/apu/amd-ryzen-9-5900hs)

Note that while it says it supports up to LPDDR4-4266, this is a different type of RAM.

Yea I stay away from Low-Power RAM. I'm not concerned about battery life and definitely don't want the performance penalty

 

[Karadjgne]

Ram speeds are selective. There are some programs like AutoCAD that can use higher speed ram to it's advantage, and games like Microsoft Flight Sim can as well, but for the most part there's no difference worth noticing.

Ryzen can benefit from faster ram, but after a certain point it exceeds the F-clock and moves from a 1:1 ratio to a 2:1 ratio, and is a detriment to performance.

What's generally a boost to performance is speed combined with timings, for instance there's no real discernable difference between 3200/14 and 3600/16, the faster speed is offset by the looser timings and vice versa.

So it's possible if the 2933 has tighter timings than the 3200, it'll match the performance with the side benefit of having double the capacity except for some ram speed specific programs.

 

[SyCoREAPER]

Ram speeds are selective. There are some programs like AutoCAD that can use higher speed ram to it's advantage, and games like Microsoft Flight Sim can as well, but for the most part there's no difference worth noticing. Ryzen can benefit from faster ram, but after a certain point it exceeds the F-clock and moves from a 1:1 ratio to a 2:1 ratio, and is a detriment to performance. What's generally a boost to performance is speed combined with timings, for instance there's no real discernable difference between 3200/14 and 3600/16, the faster speed is offset by the looser timings and vice versa. So it's possible if the 2933 has tighter timings than the 3200, it'll match the performance with the side benefit of having double the capacity except for some ram speed specific programs

Does something like CPU or anything in Windows show the ratio it's running at?

 

[Karadjgne]

No. It's a knowledge thing. Zen wasn't an issue, if you could actually get ram speeds to 3200MHz you were doin something right. With Zen+, it was 3466MHz, Zen2 was 3733MHz, and not sure what Zen3 was. That's half the reason speeds like 3200MHz and 3600MHz were so popular and 3433/3800MHz wasn't used much, even if fclock could do those speeds, as at those speeds you lost performance and really didn't get the boost back until hitting 4200MHz.

The f/m/u clocks have to do with data transmission speeds between the i/o die and cores die. That's why fclock generally reads at ½ ram speed, it's the speed of there and back. So ram at 3600 on a Zen2 has an fclock of 1800, one way, 1:1. 3800MHz has a possible effective fclock of 950, (1900 at a 2:1 ratio). On a Zen2 cpu.

Not sure, but I believe that was one of the things that got fixed for Zen3, so may not apply. But either way, dumping higher voltages and heat through ram for gains so small they don't register on anything but a bench test doesn't do you any good.

 

[hotaru.hino]

Does something like CPU or anything in Windows show the ratio it's running at?

HWiNFO can show the IF clock (FCLK) and memory controller clock (UCLK), but not the ratio. But these should be the same number anyway.

 

[Karadjgne]

FCLKs best kept at 1900 or lower if ram speeds are 3800 or faster. Amd has issues with 2000+ and stability. I think that was resolved for 7000 series, not entirely sure, haven't looked.

 

[SyCoREAPER]

Right now I am letting the laptop set everything. The faux BIOS you boot into has all those settings but I honestly have now idea what im doing overclocking RAM.

I just set it to 1600mt and left everything else on Auto. I guess the Laptop is managing it fine and the RAM is a good batch