[-]

LordBiscuits@reddit

Fucking thirty two slots?

All OEMs charge for RAM. Dell will charge you a crisp $7K to go 8x64GB on a Precision workstation.

[-]

KerbodynamicX@reddit

512 GB on a Mac? how? The most RAM I can configure on Mac Pro is 192GB, and that thing costs like $10k

[-]

How to calculate the ram requirement for a computer that would primarily used for fem ?

[-]

Madness_Reigns@reddit

By looking at the budget for the project usually.

[-]

SloppyCandy@reddit

You ask yourself two questions:

What's the max amount of RAM this computer supports.
Who is paying for it.

[-]

agvuk@reddit

I once had Matlab tell me it needed 705GB of RAM to multiply 2 arrays but I can't imagine ever needing more than 64GB for something that isn't work related, and even 64GB is high.

[-]

Used 128gb for four years. Never regretted it. Was a heavy gamer but I used so much more. My setup had 5 monitors, audiophile quality sound for music, cam and mic for streaming, mixer and dac, stream deck, ran windows parent OS with various Linux distro’s in vm, docker.

I used basically all of this at once at different times and it was always smooth as butter. Please note that I went crazy with this pc at the time and bought the ryzen 9 5950x which has 16 cores and 32 threads; it’s a work horse. I also used an RTX 4090 founders edition. I spent enough money on gear in 2020 to buy a motorcycle so… yeah.

used photoshop and Illustratior
streamed on twitch / recorded video
played video games at 1044 200fps
listed to audiophile quality music
ran dev box environment with various docker containers.
ran discord and browser with hundreds of tabs open at all times.
watched streaming services
ran desktop fusion and various other GUI mods and monitor software 24/7
routed various console video and audio to be recorded and played through my pc
played VR games and space/flight/tank sims with proper flightstick/thrust/instrument setup.

We use some EEG acquisition/analysis software that's only available for mac, although in the past five years I haven't actually even installed in on this machine.

Other than analysis, I also use it to code experiments, where stimuli are presented. Timing matters to us even more than to video games, because EEG can record brain responses with ~ms accuracy, so even a frame of latency (16.6ms) is a real problem. We use a matlab toolbox to expose opengl, and to do fancy timing-related things like track the beam position of a CRT monitor as it refreshes, giving us sub-frame timestamps of when an image actually started being drawn (depending on its horizontal position).

These days, we don't use CRTs and the toolbox is no longer recommended for macs anyway, but there was a time when a mac was the best option for stimulus presentation, and so I standardised all my code to run on mac a decade ago and it's just easier* now to keep buying them than to mess with it.

The other thing is that this machine is really nice. For a company that was one of the first to solder/glue components into laptops, Apple engineers did a really nice job of making a modern desktop PC. It's also amazingly quiet. It can suck down over 400W without making a sound. Meanwhile, my 3090, and even the dark rock cooler on my 5800x3D, make a hell of a racket.

* not really, cos it becomes death my a thousand mac hardware upgrades, but I can limp along without having to really bit the bullet at any one time

[-]

I'm not sure how much the RAM draws, but must be an appreciable amount just to provide current for 384GB.

even going that route, 32 to maybe 64 GB is more than enough

Divini7y@reddit

That's it - that's how RAMS works. You are correct.

[-]

S1rTerra@reddit

But I thought you shouldn't use that much ram? It ruins your performance. /s

[-]

Solonotix@reddit

It can, though, for slightly technical reasons.

So, the fastest usage of memory is to not have any, and only use your CPU cache. This isn't possible for most use cases, so we will see memory allocations happen, even if it's just loading the instruction set for the application you are running.

So, we're using memory. How do we best utilize it? Well, for one, the fewer calls you need to make, the better. That's where you see things like SIMD (Single Instruction Multiple Data). This often requires organizing memory in the order you will process it, and structuring your code in a way that works with SIMD optimizations. It is at this point I would say allocating more memory than what you strictly need is good...

But then we get to the other side of that decision. What happens if you over-allocate memory. Well, fortunately, uninitialized memory is basically free to request. However, uninitialized memory is risky, as it requires that the rest of the code operate on a basis of checking the data for initialization before usage. The more times you need to access the memory to check that it's initialized properly, or initialize it if not, is going to make your application run slower. Failure to do these checks will often result in undefined behavior, and that can lead to system instability.

And lastly, we get to the point of how this plays with other applications. Take into account that every application operates under the same rules as I mentioned above. Now, consider that for proper SIMD optimizations, memory layouts often need to be contiguous. Larger allocations become harder to optimize this way as more RAM is allocated because there will be fewer large chunks remaining to use. This is another way it can be problematic to over-provision memory, since the kernel will do its best to give you what you ask for, but it may only be able to give you a virtual allocation that looks contiguous while the physical layout is actually disorganized and performs poorly.

And lastly, if every application asked for all the RAM available at all times, the system would rapidly run out of resources and be unable to handle requests for new allocations. As such, it is generally a best practice to only use as much memory as you need, and to be sparing in your allocation of additional memory

[-]

homelaberator@reddit

So, what you're saying is I should get more RAM.

[-]

GuardianOfFeline@reddit

There are a lot of misconceptions here.

There is no risk involved when you allocate uninitialized memory. You simply can’t read it and that’s it. If you read it, you are simply writing a bug, same as any kind of other bug that you can write. Modern code analysis tool can also easily catches these so they don’t really easily make to production code. There are often use cases where you simply need to allocate a lot of memory and not use it: e.g. to store the result of large matrix multiplications.

There is also no need to check if a segment of memory is initialized or not at runtime because it is the programmers responsibility to know if they initialize it or not.

Pre-allocation actually reduces fragmentation.
Even when the physical memory is fragmented. As long as data is continuous in the virtual memory the SIMD can still operate very effectively. The TLB will hide the physical address from the SIMD unit.
Yes, you don’t want to malloc a chunk of memory for no reason. But things like Video editing either uses more of your scratch disk or more of the memory. If you have a large memory it makes senses to use more. So it is actually a very good reason here.
General rule of thumbs are dumb. Considering trade offs is very important in any kind of Engineering.

It's likely Photoshop, when I have to edit a few high res images at once I'll start dipping into swap on 32GB. If you work with seriously high res photos, I could easily see 64GB not cutting it.

[-]

gatornatortater@reddit

You doing a roadside billboard at 1200dpi or something?

[-]

Putrid-Flan-1289@reddit

No shit Mr Technical. Ever hear of a joke?

[-]

buildapc-ModTeam@reddit

Hello, this has been removed. Please note the following from our subreddit rules :

Rule 1 : Be respectful to others

Remember, there's a human being behind the other keyboard. Be considerate of others even if you disagree on something - treat others as you'd wish to be treated. Personal attacks and flame wars will not be tolerated.

Consider your life choices mang.

[-]

xRealVengeancex@reddit

You are wasting money if you got for 128 lol

[-]

Poppa_Mo@reddit

Dig into the settings a bit, you can tweak this so it's used optimally for what you have/need and not just blanket grabbing what isn't already claimed. Almost all the Adobe Apps have this, but it's not a surface level setting you just toggle, it's buried a bit and in different locations in each app.

Actually there is one game that could probably do this called BeamNG a softbody car simulation game.

Not that it’s poorly optimized or anything but you can if your cpu is strong enough spawn so many cars that it will definitely hit 128gb.

With ~14 non simplified cars I already completely fill my 32gb.

[-]

Riaayo@reddit

No Game I know of uses anywhere close to that.

I feel like I could have filled that up with Cities Skylines mods to be honest lol.

But yeah no unmodded game is going to need that much ram in the next two decades at a minimum. Hell, the custom pc market will probably be dead and everything will be single-board bullshit before any game would need that much.

[-]

Hitchie_Rawtin@reddit

Video Editing, 3D modeling & virtual machine hosting is the only thing I can think of that is capable of using that much RAM

ButterCup955@reddit

Cheat Engine. in order to find the static address of a value, u ve to conduct something call a pointer map search. that ll eat up as much ram as u got.

[-]

leffronte@reddit

I'm doing CT-scan segmentation and softwares like 3D Slicer recommend having 10x more RAM than the amount of data to load (typically 8 to 12 Gb in my case).

[-]

blearx@reddit

Machine learning. Though not always maxing it out

[-]

wowcorny@reddit

16 is the bare minimum now, 32 is ideal, 64 is for productivity, 128 is for very specific cases only or if you are rich and have 1000 chrome tabs open

[-]

txmail@reddit

or if you are rich

Prices have come down so much, 128GB of DDR4 is only like $150 for a server or $250 for a desktop (or $180 on sale). No way I would build a new right without a minimum of 64GB and probably just max it out from the start which is usually 128GB or 256GB.

[-]

Criss_Crossx@reddit

64gb is still a reach for most people. Definitely cheap however, I may pick up a kit eventually for my 7900x. Hopefully it will be a long time before 64gb is the budget capacity. That's a lot of space for most uses and applications should be optimized, not bloated to fit.

I upgraded CAD workstations to 128gb and there was a noticeable difference. Went from 32gb->64gb->128gb. And the Autodesk software still appeared to want more, including GPU resources. Those two systems had rtx 5000's.

The newer workstations I worked on this year were HP Z4's with 16 core xeons and a4500's. These systems were buttery-smooth in rendering in Inventor and I think they could be optimized further. I enjoyed watching my supervisor run every piece of software he needed on it. He was impressed too.

Incorrect - if the LLM is that big that it takes 256 GB, the required GPUs to run that would be $50k+ especially since inference is relatively cheap

[-]

dank_imagemacro@reddit

Macross_zero@reddit

128 GB of RAM doesn't just have to be for gaming. You will never use that much. At least not yet. But as others have said. Video editing, 3D modeling and Fluid simulations can easily approach and even go past that on a regular basis though. Another thing to think about as well is music production. I do this as a hobby and even with 64 GB of RAM if I'm working with a lot of samples it can be a struggle.

[-]

TheGreatTurning@reddit

Yes. For fluid and pyro sims in Houdini.

[-]

I have sniper elite v2 saved to a ram disk on my server so I can just load it into my gaming machines ram and play it that way. I did it just for fun, but I think I would need 10gbit networking to make it more practical for larger games.

[-]

BurrowShaker@reddit

What's the point if nvme is faster than the fastest reasonable networking. The latency advantage is going to dwarves by network latency anyway.

[-]

PinchCactus@reddit

I have a ton of storage in my server, I don't have nearly as much storage on my PC. Rereading this.... I don't think you understood exactly what I'm doing. Step 1. Create ram disk, step 2 install game to ram disk, step 3 play game. Step 4, image ram disk and save to server. Step 5 Close ram disk. When I want to play, I start up a ram disk and load the image to it. So the game runs entirely from ram once the image is loaded. 10gb would make the loading the image to the ram disk much faster. It's not super practical but it's fun and doesn't use storage on my main PC.

[-]

BurrowShaker@reddit

Oh, I thought the ramfisk was on server, and that made zero sense to me.

The ramdisk image makes more sense.

My bad

[-]

PinchCactus@reddit

No but that could be fun. Time to run the ram disc on a virtual machine on my server lol

[-]

BurrowShaker@reddit

Not worth the hassle I'd say.

I'd probably would not bother with the ramdisk and have the server storage over NFS, FS cache in ram should achieve much of the same performance without the loss of ram, I'd expect.

[-]

I do. Only when playing Minecraft.

[-]

AtomicNixon@reddit

Yes. Houdini.

[-]

appetrosyan@reddit

I use 96GiB and intend to upgrade to 192 when the time comes.

Solana takes a lot of RAM, and there are benefits to running some things inside a RAMDisk.

Programming can give you a lot of mileage out of lots of RAM, if you know how to put it to work.

[-]

AbdDjamil_27@reddit

Congratz, you can now open 2 tabs of chrome

[-]

marcuseast@reddit

I have 128GB on my gaming rig, but I never use more than 64GB in reality. It’s not necessary.

[-]

rbardy@reddit

When your sistem gets close to use 64gb?

It's been like this for at least 15 years. Take a look at most modern games and tell me they're somehow better in improvement than the 15 year segment prior to that.

Gameplay advancements in the AAA industry crawled once we hit XB1 and PS4 era. It was already stagnating in the PS3 360 era.

Just about everything gameplay wise we do with gaming today can be done on hardware made 15 fucking years ago, some sims notwithstanding. But AAA games aren't sims.

[-]

fuckandstufff@reddit

Games are better today than they ever have been. Anyone who says otherwise is clouded by nostalgia. You're allowed to like old games. We all do. But try and be objective broseph. Look at demons souls to elden ring or baldurs gate 1/2 vs baldurs gate 3. Hell, even Diablo 2 vs Diablo 4.

[-]

winterkoalefant@reddit

In their defence, you're always going to find more adventurousness in indie games because they have to innovate to get any recognition. AAA games cost a lot to make (by definition) so they have to be more cautious. New ideas from indie games usually make it into big budget games, either by their success resulting in higher-budget sequels or by getting copied. Gamers benefit.

As for hardware requirements, Microsoft Flight Simulator 2020's earth-size world, Fortnite's building and combat mechanics, No Man's Sky's seamless interplanetary travel, A Plague Tale's swarms of rats, etc., I would say their gameplay absolutely justifies their hardware requirements.

Even if there's little gameplay evolution, we're paying for new stories, new environments, etc.. Not just graphics, and not that better graphics aren't appreciated.

[-]

HehaGardenHoe@reddit

The 360 Era was great IMO, and the PS4 Era was fine but had most of the innovating near the start (whether it was attempts at using the pad on the controller, or things like No Man's Sky or the nemesis system from Shadow of Mordor).

Mind you, it's your Chrome extensions that take up all that RAM, not Chrome itself.

[-]

KFC_Junior@reddit

just btd6 on a late game run for me eats up all 32gb ram...

[-]

Finetales@reddit

Modded Cities Skylines will eat through even 64GB if you have enough mods.

[-]

marcuseast@reddit

Playing a game like Cyberpunk 2007 in ultra at 4K on my widescreen while live-streaming and/or video editing, streaming to Discord and having 40 tabs open on Chrome at the same time.

[-]

rbardy@reddit

I feel a bit more releaf lol

yeah maxed out DCS thrives on 128gb.

I tried for my am4 system with 2 different kits but both kits generated errors in memtest even at stock speed.

Even just running 2 sticks of 32gb wasn't stable in memtest, with 2 different cpu's.

This was for a truenas build and potentially running vm's and servers on it as well.

[-]

karo_scene@reddit

I don't.

You would only need 128 GB of RAM if you were [1] running many virtual machines or [2] doing physics simulations. For instance Maxwell Render, a physics simulator that simulates physics equations to create 3D scenes using modeled materials to look like aluminum, quartz etc.

[-]

DJDIRTYDAVIE@reddit

I have 32 and it's perfect. I just use it for school and gaming.

[-]

LouBlacksail@reddit

I just literally put together a brand new Ryzen 9 PC with RGB 128GB of RAM. To be fair there is no reason to not go above your need with resources, but I do plan to do video editing (needed) and also do audio production (which is far less RAM taxing).

I have 64gb, 32 for gaming and 32 for all the random bloat apps I leave open.

Unless I was video editing and rendering in blender whilst also playing Minecraft and Skyrim modded simultaneously, I don't think I could hit 128gbs.

Started with 16GB, upgraded this year because I have the slots and I might as well use them. I use it mostly for laughing at the poors who don't have 128GB like I do.

Help me.

[-]

Secretx5123@reddit

Data science, constantly I need of more memory. Even 128 is lacking sometimes.

[-]

Hexagonian@reddit

Here. Working on megaprojects tends to do that. One subsection with relevant links loaded can easily go north of 20GB

[-]

Ethan-Hou@reddit

I mainly use it to run my matlab programs, which are some mathematical simulation models

[-]

hardcore_softie@reddit

Hell yeah bro I'm running 128 gigs...across 4 different rigs.

I think most people using that much RAM in one system typically are doing a lot of 3D video editing and stuff. There's also the benchmark guys trying to hit that fighter number just because.

I use 32 GB. At work I use plenty of systems with hundreds of gigs of RAM. They are used for delivering data in datacenters. They connect to like 80 hdds and read/write to a lot of them at once. The ram might be for caching, not really sure.

I have 128.

I got fed up with running out at 32 that i just said "screw it" and went with the highest amount i could fit.

I no longer have any situation where i can run out of RAM.

[-]

Cyber_Akuma@reddit

I no longer have any situation where i can run out of RAM.

fuzzynyanko@reddit

64GB in my desktop. When I'm not running virtual machines, I rarely break 32. The amount of RAM available to Windows to use for caching goes up. A side-effect of this though if that it makes startup REALLY long since I have a mix of SSD and magnetic drives.

Virtual machines and AI

[-]

bs9tmw@reddit

Basically 2 categories of answers:

128GB is way too much (gamers)
128GB is a nice start, but I could use more (homelabbers and self-hosters)

[-]

CanadianKwarantine@reddit

Not yet.

[-]

maw_walker42@reddit

My FreeBSD package build box had 96. I don’t know if necessary but was cool because I never ran out no matter what.

[-]

wraith8015@reddit

Servers will routinely use >128GB for virtual machines and things of that nature (which isn't using the RAM as much as allocating it anyway...)

There's really not much use in a normal system to go beyond 64GB other than future proofing. 128GB would be a practical hard stop in a consumer system. Even these "heavy production" applications in the comments seem out of touch with reality. Most of those apps benefit from hardware acceleration, and care a lot more about how much VRAM you have on your GPU than about system RAM.

I'm sure there's "that guy" with 63,000 chrome tabs open, seventeen different copies of a production app running, all while gaming who would overwhelm the system, but you really would have to use as much RAM as possible for the sake of using RAM for that to really be a thing.

[-]

celestrion@reddit

Does anyone use 128GB RAM on their system? And what do you primarily use it for?

Virtual machines (package builders and simulations of customer environment) and moderately-sized databases on one machine.

Two jobs ago, I worked for a financial transaction monitoring firm and I put together a database of detailed stock market activity for a single day that we could use for regression testing our monitoring code. Searching for patterns of activity (as opposed to hand-crafting something intentional) for real-world test scenarios made a 64GB machine cry. With 128GB of memory, I'd actually get results in hours instead of a couple of days.

The package-building (running poudriere) virtual machine eats 64GB of RAM all by itself. It could do with more, but its VM host only has 192GB, and I do other things on it. More memory for building packages means being able to run more compilation threads at once without swapping to disk. Finding the sweet spot between threads and swapping makes something like a 10x difference in overall runtime.

[-]

markianw999@reddit

My chrome tabs come close witj pre tab sleeping builds. 90 something

[-]

Pyro919@reddit

[-]

xilvar@reddit

On my laptop (an m1 max MacBook Pro) I have 64gb and I use it to run LLM models (AI) at a pretty decent rate due to the much higher memory bandwidth of main memory on macs. However, I cannot run models as large as I would prefer to run with only 64gb.

I have been considering upgrading to an m3 MacBook Pro with 128gb of ram in order to do so. Or paying a modification shop to upgrade my current machine to 128gb by soldering in all new ram.

That being said, I’m more likely to continue my current path on my pc hardware of obtaining more and more 3090’s to reach my ram target. (Currently at 3x)

[-]

Blu_Falcon@reddit

Pssh yeah… doesn’t everyone run a full OpenShift cluster on their PCs?

[-]

Arcal@reddit

256GB in my workstation

[-]

Past-Catch5101@reddit

ZFS uses most of it 😅

[-]

Savacore@reddit

I have 128 gigs.

So basically, entire genres. Are the most popular genres? No, but there's still a LOT of players in each of them in absolute terms.

Penrosian@reddit

I do! I use it for... my servers!

[-]

Meatslinger@reddit

I know a few people that have 128 GB of RAM, but none of them use 128 GB of RAM. There are a few productivity-related ventures that have already been mentioned in this thread, but generally speaking there is no everyday service or application a person would need that much RAM to handle; you have to be getting into very specialized needs to necessitate that much. Even the ones in regards to video editing are tenuous, as having less RAM simply means a bit more swapping with disk storage. Though this imparts a performance hit, it’s not the kind of impact that would prevent a person from doing it just fine on a system with 64, or 32 GB of RAM.

That said, most modern systems will load the RAM up as much as they can. This is simply because if the space is available, it makes sense to use it. If you have 32 and your program of choice, designed to work on just 8 GB of RAM, can use 28 GB for assets to improve performance (reserving 4 for the rest of the system), it likely will, and will start swapping to disk when it hits your 32 GB limit. Give it 64 to play with, and unless the developer put a hard cap on it, it will use more. So yes, you can fill up 128 GB of RAM, but you’ll pay substantially more for what often equates to sub-1% performance gains.

[-]

Machine learning, loading a 405gb llm model requires 405gb of ram and vram.

[-]

My normal pc is doing fine with 32GB, but I’m considering upgrading my server/nas from 128GB to 256GB RAM. It’s starting to hit the swap file sometimes.

[-]

Tristan_poland@reddit

Unreal Engine game development, Docker containers, And of course, Zen tabs by the hundreds

[-]

dulun18@reddit

probably 1% of PC owners

32GB is plenty for gaming

[-]

Finsceal@reddit

I edit 4k footage for work and regularly get bottlenecked by 64gb when I have premiere and after effects running

[-]

thisoilguy@reddit

I have 96GB of ram and it keeps chrome stable

[-]

HettySwollocks@reddit

Servers: Yes. Throw in some VMs, containers, LLMs, various caching, transcoding etc etc. It soon eats up RAM - I'd argue 128 is on the light side tbh.

On my main machines (which are mostly 64), yeah I can hit that pretty quickly depending on what I'm doing. Sure if it's just every day tasks (browsing, casual gaming) it's overkill - I'm using 23Gig now.

Astrophotography! I basically stack hundreds of uncompressed photos to increase signal to noise ratio of very faint objects, and the net result is beautiful, but sometimes I use ALL of my 128gb and it starts swapping like hell. Just upgraded to 192!

[-]

KirillNek0@reddit

In the next 5 years it is where all new OC should be.

[-]

Naojirou@reddit

Development in Unreal Engine. We end up having bugs that are only replicable in a packaged build. Running the server and the client at once on a BR game and debugging them in ide makes 128 a definite necessity.

[-]

eyetac@reddit

Threadripper and 128gb DDR4 3600mhz. I work on big media time lines and projects in the broadcast space. Going from 64 to 128gb has definitely made for a smoother experience.

[-]

Hax0r778@reddit

The build I was using the last 9 years had 32GB and that was great. I could open so many Chrome tabs.

So when I just built my new system I future proofed it with 96GB. Because I could still do that with 2x48.

I don't really need it, but now I don't have to worry about it for a long time. And I have since picked up a 3d printer and done some modeling. So it's useful for that.

[-]

mrcandyman@reddit

I have 64GB in my system and I've never run out, but I have been pretty close. That's when working on CAD or editing my shitty videos.

[-]

avalanche111@reddit

No, I'm comfortable with my manhood.

[-]

vamadeus@reddit

I have 64 GB in my desktop. Stable Diffusion and sometimes other media editing/rendering can make use of that RAM. Also heavy multitasking.

For gaming 32 GB is more than sufficient. That's what I used for a long time.

[-]

Dukes159@reddit

I did for some programming projects but now that those are finished im back down to 64 for the higher clock speed

[-]

ogie_oglethorpe@reddit

I put in 128gigs of RAM but I also do video editing for my shitty YouTube channel 😂. Honestly it wasn't a big leap to throw in another 64gb with how much I spent to build the PC on the first place.

[-]

Role_Playing_Lotus@reddit

If you're just gaming and doing some streaming and browsing, 32GB is the new minimum recommendation.

This is according to credible sources like Jason of PC Builder, who mentions other sources as well.

Jason explains that while 16GB is still enough for most games, just a few of the newer titles have been recorded using all of 16GB or more. Because 32GB is the next logical step up with what's commonly available on the market, that's the new recommended minimum.

I have 64GB but only because I use my computer for design projects involving a handful of Adobe CC applications (AI, ID, PR, ME, PS) in addition to heavy web browsing and light gaming. I've almost never used more than 32GB with even my most intensive workloads.

Like others have said, if you're a heavy game modder or use Adobe After Effects, you might want to bulk up on RAM (64-128GB). Otherwise, 32GB is going to be just fine for a vast majority of PC users.

It's not unreasonable. Motion tracking footage is still slow as hell. But if you can work pretty freely within the cache, it gets done quicker. bigger is better. Uncompressed video will slay your RAM

I only use 64 tho, I don't do this stuff often or in 4k

i have 256gb of ram, i do 3D stuff. its nice mostly when something bugs tf out and memory leaks. tends to save a crash but its amusing as shit watching something slowly climb and climb

I do! I can allocate a bunch of ram to a VM or some sort of running process like a game server

Yes, for bioinformatics work. Heck, I almost ran out of ram yesterday.... 98 gb used. Crikey.

For personal use i have 64 GB. Enough for everything so far.

[-]

EirHc@reddit

I was trying to do a photogrammetry project last night and my 32gb wasn't enough and the software literally just stopped and told me to get more ram. Thinking of building a machine specifically for 3d modeling that has 128gb.

[-]

The-Foo@reddit

Development work: I run a bunch of VM's and containers (including CUDA / Pytorch stuff loading large datasets and models). Hell, even my GPU has 24GB thanks to the need to run large models. There are days where, even with VM dynamic memory enabled and cgroup restrictions on containers, 128GB isn't nearly enough, and I end up having to burst things over a VPN into my VPC in public cloud (which is stupid expensive, relatively speaking) - and that's with having a set of dedicated virtualization boxes, each with 128GB, to run all kinds of infrastructure, automated local build and test pipelines, and such..

But I've resisted going to something like Threadripper because the cost is brutal and, frankly, the 32 threads of my 5950x are fine from a compute standpoint (the heavy lift is mainly on my 4090). Current consumer boards and CPU's max out at 192GB (on X870 and the like), which isn't enough to get me to go through the hassle of an upgrade. When I can get to 256GB on the consumer side hardware, then I'll upgrade.

I have 192GB (4x48GB) on a 7800X3D, for compiling some kernels things and also machine learning/AI, it helps a lot.

Yes. I built a few "budget workstations" for photogrammetry/LiDAR/GIS with 7950X+4080S (latest one even got a 9950X).

manon_graphics_witch@reddit

128 GB of RAM for compiling large code bases at work is really nice.

[-]

RAM requirements seem to double roughly every 6 years. Right now, 8-9 GB is about the least you'd want for basic daily use (8 feels like it's just below minimum, even on my PC at work).

With that in mind, 128 GB would keep your system running smoothly without excessive swapping until 2047. By that point, everything else will be so slow, you'll have already upgraded your platform twice.

Verzada@reddit