Will I ever need or even be able to use 32GB of memory?

[whm1974]

In my i5-4670 based rig, I have 2x8GB DDR3-1600 modules and I'm wondering if I will ever need to add more RAM before I build another system? Personally I think the answer is no, and given that I have 4c/4t CPU, I think it is very likely I will run into needing more cores and threads before wanting for more memory.

What do you guys think? I know right now the sweet spot of memory is around 16GB as that seems to be plenty for most users with the majority of use cases.

I'm right about this?

 

[BonzaiDuck]

In my i5-4670 based rig, I have 2x8GB DDR3-1600 modules and I'm wondering if I will ever need to add more RAM before I build another system? Personally I think the answer is no, and given that I have 4c/4t CPU, I think it is very likely I will run into needing more cores and threads before wanting for more memory.

What do you guys think? I know right now the sweet spot of memory is around 16GB as that seems to be plenty for most users with the majority of use cases.

I'm right about this?

Probably. And I wouldn't spend more money on RAM for a Haswell system -- unless you 'really need it'. but what does that mean?

I started configuring my Skylake system in fall 2016. I thought about 32GB versus 16GB. I settled for the latter, knowing it was more than enough for anything I do. But I was also using the program PrimoCache to cache slow source disks to NVME and RAM. So, victim to my own curiosity, I bought the second kit.

It all works great, and I don't regret the purchase. Even so, the performance gains are questionable for the expense. It wasn't a practical decision. It was a "let's see and find out" decision. Or an "I was thinking about 32GB at the beginning anyway" decision.

Enthusiast-tinkerers will always spend more than they have to. Even if they are trying to avoid it.

 

[dave_the_nerd]

I'd agree with your supposition - probably not with this rig, it's already 4-5 years old.

But eventually yes. Progress happens, and it usually ends up needing MOAR RAMS and MOAR COARS.

 

[whm1974]

I'd agree with your supposition - probably not with this rig, it's already 4-5 years old.

But eventually yes. Progress happens, and it usually ends up needing MOAR RAMS and MOAR COARS.

I'm thinking for my next build I will go for 32GB RAM and a 8c/16t CPU.

 

[BonzaiDuck]

I'm thinking for my next build I will go for 32GB RAM and a 8c/16t CPU.

Well -- I built my basic Skylake rig in October, 2016, and finished it July of last year. It only has 4 cores and 8 threads. You will certainly beat me to the octo-core, because I'm happier than a pig in sh-poop anyway. And I probably think that an i7-8700K (6c/12t) is the limit of my ambition.

 

[Mr Evil]

Ever? Definitely. Consider that in 2003 people complained when a web browser used up 60MB of RAM. By 2016 complaining about a browser using >1GB was common. That's more than a 20x increase in 13 years, and there's no reason to think that trend will change in the next 13.

Over the same time period, mainstream CPUs have increased in core count by about 8x. Though we will surely get moar cores in the future, there is good reason to believe that the need for more will slow down.

 

[whm1974]

Well -- I built my basic Skylake rig in October, 2016, and finished it July of last year. It only has 4 cores and 8 threads. You will certainly beat me to the octo-core, because I'm happier than a pig in sh-poop anyway. And I probably think that an i7-8700K (6c/12t) is the limit of my ambition.

Well I'm planing on build another system about three years or so from now, so I imagine that 8 cores will be mainstream as i5 are right now.

 

[dave_the_nerd]

Well I'm planing on build another system about three years or so from now, so I imagine that 8 cores will be mainstream as i5 are right now.

Quad cores were Intel's mainstream from 2007 (8xxx series C2Q) until like ten minutes ago. I think the move to 8-core consumer chips will take a little longer than that.

 

[whm1974]

Quad cores were Intel's mainstream from 2007 (8xxx series C2Q) until like ten minutes ago. I think the move to 8-core consumer chips will take a little longer than that.

Well if AMD can continue to improve as they have been doing, maybe moving over to 8 core processors will not take as long.

 

[dave_the_nerd]

Well if AMD can continue to improve as they have been doing, maybe moving over to 8 core processors will not take as long.

They had hex cores back in the Phenom II days, which was also 8-10 years ago. The 8-core 'Dozers weren't even really an improvement on that.

You're doing the extrapolating thing and not thinking long-term enough.

 

[whm1974]

They had hex cores back in the Phenom II days, which was also 8-10 years ago. The 8-core 'Dozers weren't even really an improvement on that.

You're doing the extrapolating thing and not thinking long-term enough.

Well yes but Ryzen is a great improvement over the 'Dozers. Hopefully AMD won't drop the ball when I get ready to build.

 

[BonzaiDuck]

Well I'm planing on build another system about three years or so from now, so I imagine that 8 cores will be mainstream as i5 are right now.

Yes -- that makes sense. Everyone has their own build and retire cycle, depending on different sets of factors. If it's as much about "want" as "need," I think of it as something like surfing: People look for the right wave, and they have different perceptions about which wave is the right one. So we anticipate new generations of processors, motherboards and other things; we look at the expenses and our resources; we order our parts and go forward.

I can imagine retiring my Z170 and 6700K in three years, but I'll need to see what my needs and wants are at that time before I spring into action.

If memory weren't so pricey as it is now, it could make more sense adding RAM to a slightly dated system.

 

[dave_the_nerd]

Well yes but Ryzen is a great improvement over the 'Dozers. Hopefully AMD won't drop the ball when I get ready to build.

If AMD and Intel do their usual bit with a few years in between major micro-architecture changes and more-or-less annual "refreshes" that give a small improvement, then around 2023 you'll be really glad you waited.

OTOH, if you build now, you know that whatever you get won't likely be surpassed by anything significant for several years.

But pity the poor fool who bought a 7700k last August.

 

[whm1974]

If AMD and Intel do their usual bit with a few years in between major micro-architecture changes and more-or-less annual "refreshes" that give a small improvement, then around 2023 you'll be really glad you waited.

OTOH, if you build now, you know that whatever you get won't likely be surpassed by anything significant for several years.

But pity the poor fool who bought a 7700k last August.

Well if I was to build now, it would likely be using the i7-8700 or Ryzen 7-2xxx(?) so I would still be good for another five years anyway.

 

[dave_the_nerd]

Well if I was to build now, it would likely be using the i7-8700 or Ryzen 7-2xxx(?) so I would still be good for another five years anyway.

Right. That's what I said.

 

[cbn]

One advantage to having more RAM is that Windows will use it as cache:

.......but right now RAM is really expensive so I recommend you stay at 16GB if you find your I/O sufficient.

 

[cbn]

Another thing to consider is RAMDisk with a machine learning mode. (Upcoming release of DIMMdrive, etc.)

EDIT: Also Romex Primocache.

Why PrimoCache?
Why do we need PrimoCache since Windows has already implemented a file-system level cache?

• PrimoCache is a block-level cache program which caches data on a logical block basis (offsets within a disk), while windows cache is a file-system level program which caches on a virtual block basis (offsets within a file). Each has strengths and weaknesses, and PrimoCache can be a good supplement to Windows cache.
• Some applications can bypass Windows cache but cannot bypass PrimoCache because PrimoCache runs at a lower level in Windows.
• Windows caches all data, while PrimoCache can cache on behalf of a specified volume in which users are interested. Given same size of system memory, the latter has a higher hit-rate.
• PrimoCache supports persistent SSD caching for mechanical hard disks, improving system boot-up time and applications loading time. Windows cache cannot.
• PrimoCache can customize write-deferred mode, while Windows cache cannot.
• PrimoCache can make use of Invisible Memory on 32-bit Windows as cache, overcoming the Windows limits on amount of system memory.

 

[ch33zw1z]

Depends on usage, if you're not doing anything that consumes 60% or more of your ram on a regular basis, save the bucks

 

[BonzaiDuck]

This is now the 4th year consecutively since I began using the 90-day trial of Primocache on one or more systems. [It is now a 60-day trial.]

There have been virtually no software conflicts using this software, with only a few exceptions that should be noted. More specifically, the benchtest software we use in over-clocking activities seems to collide with the memory set aside by PrimoCache for disk caching. I'm not sure if there were BSOD events associated with it, but it caused the software itself -- such as LinX or Prime95 -- to malfunction. The solution? Remove the Primo caching tasks that use RAM before running these programs. Although I remember no problems with L2 (SSD {SATA, NVME}) cache configurations, this sort of benchtesting for CPU or RAM overclocking doesn't benefit from any sort of cache, and I would turn off L2 caches anyway. After you've found your stable settings, you can set up your PrimoCache tasks.

I advise that any computers -- if possible -- should have UPS battery backup. I especially advise UPS backup when using PrimoCache, mostly because you can safely take advantage of deferred-write caching with much less worry about corruption occurring from a power-outage shutdown. conversely, as I may have implied above, be absolutely confident that your system is stable in CPU clocks and voltage as well as RAM speed and timings. that being said, everyone experiences an occasional BSOD even for a temporary driver conflict. I have yet to experience any corruption or problem with Primocache when these events occur. Once in a blue moon, you may experience a cache reset for some reason most likely with version 2.x.x of Primo. Version 3.x.x seems to have eliminated an episode that has coincided with monthly Windows UPdates.

A combination of NVMe SSD caching and RAM caching provides a level of performance far greater than a human being could detect in comparison to simple NVME performance. The benefits are more noticeable with slower source-disk candidates for caching, particularly HDDs. Caching to RAM would be ideal to the degree of the memory installed -- the more, the better. A minor drawback arises with a delay in boot-time if you elect to save the cache to disk anticipating its reload at boot-time or after a Restart, but boot-time still feels faster than it should. Read the Primo manual about necessary choices when you configure your system to hibernate, but it amounts to enabling a single check-box.

The least frivolous expense in using the software is the price of the software itself. You can create some high-performance configurations with it under choices involving more expense in hardware, such as total memory or SSD speed ( high-end NVME) and size. You have to decide whether or not the hardware is worth the gains. Also, it is now obvious that Primo works with Optane devices. Check the forums at the Romex Software web-site, where you will find threads discussing such options.