A new report states that Microsoft has provided baseline specifications for AI PCs and increased memory capacity is a priority. Increased memory requirements for PCs, regardless of platform, are always to be expected as new technology and features get introduced over time so it’s no big surprise that the age of 8 GB of RAM is on its way out. After all, as the jokes go, exactly how many Chrome tabs were you hoping to keep open at one time anyway?
Trendforce has reported on the ongoing commercialization of AI servers and AI PCs with a focus on Microsoft’s Copilot AI features. The new report states that Microsoft has set a new minimum memory requirement of 16 GB. While most pre-built desktops usually have this, or more, the mobile industry typically lags a bit behind in this regard, but devices with larger memory capacities have become easier to find there too thanks to increased marketing towards productivity usage and content creation. As PC manufacturers continue to incorporate these larger memory capacities it too is expected that there will be a higher demand in the memory manufacturing sector.
Per Trendforce:
“Microsoft has set the baseline for DRAM in AI PCs at 16 GB. In the long term, TrendForce projects that AI PCs will catalyze an increase in annual demand for PC DRAM bits, with consumer upgrade trends further boosting this demand.”
Trendforce also shares that Microsoft is requiring 40 TOPS of computational power for AI PCs This is expected to be achieved via a combination of CPU, GPU, or NPU technology which should become more common by the end of 2024 and in mass production by 2025 with the launch of Intel’s Lunar Lake processors. Until then, newly launched AI PCs with Intel’s Meteor Lake processors are said to be just under the 40 TOPS threshold, and as explained by Tom’s Hardware these current processors are “almost” at that level of power but not quite.
Per Tom’s Hardware:
“We said ‘almost’ above, and that is an important caveat, as the combined CPU, GPU, and NPU power within Intel’s current Meteor Lake chips are said to reach 34 TOPS at best.”
AMD and Qualcomm are expected to bridge the gap in their upcoming lineups. The AMD Ryzen 8000 series and Qualcomm Snapdragon X Elite processors are both expected to provide roughly 45 TOPS. The last piece in this particular puzzle comes back to RAM but this time concerning its speed. Since storage technology has already been reaching epic potential speeds via PCIe 4.0 and 5.0 NVMe drives, although with caveats regarding cooling, the next elephant in the room is keeping processors from being bottlenecked by slow memory modules. However, this too has been something that has improved significantly with the introduction of DDR4 and now DDR5. In turn mobile memory such as LPDDR5x should take mobile AI PC computing to the next level at an estimated 7500-8533 Mbps, that is if RAM manufacturers can meet the ongoing increased demand for both.
Per Trendforce:
“Consequently, TrendForce forecasts that LPDDR will represent approximately 30–35% of PC DRAM demand this year, with future increases driven by the specifications set forth by AI PC CPU manufacturers, leading to a further rise in LPDDR adoption.”
Discussion (18 replies)
Join Discussion →I'd argue 16GB is bordering on obsolete at this point...
Yea most of the mid tier laptops today come with 32 gig.
Hmm..
So I would say - ok, we have been at 8 for a long long time. But for most people (not all of us here, obviously) - the OS uses most of that for ... peripheral stuff ... and the rest of it is mostly used up by Chrome. Most things that get run by normal people on a daily basis - they just don't use a lot of RAM.
Office doesn't use a ton. Games require more these days, but it's rare to see one require more than 8G (they do exist, just not common outside of the big AAA releases). Just looking over the "Top Sellers" over at Steam right now - none of them require more than 8G (although a couple are recommending 16).
So. We are probably long past due for an upgrade there.
But.
But.
I'm all for upgrades, so long as they are useful. WTF is Microsoft going to use that RAM for on my PC? My understanding was that most of AI compute was done back in the cloud somewhere - my front end just did basic request processing and passed it on. It doesn't take 16G+ of RAM to do Voice Analysis or biometrics, or other basic AI tasks.
Sure, if you want to run a LLM offline, you need some RAM. But I wouldn't want my average office PC running LLMs - that adds a ton of cost to otherwise commodity PCs that get bought by the hundreds/thousands. Just like I don't need my basic office PC hosting a web server, or SQL database, or any number of other things that are best served being centralized and accessed by slimmer clients.
Yes and no. I’m doing quite a bit of work on my 2GB raspberry pi 4s. Works great for Kodi on Xbian, pihole, etc.
Yeah, but on Windows 64 bit it's far beyond being able to do that. I respect what RPs can do though as I've followed their development quite a bit over the last decade.
Honestly the push for more ram is ai driven. More and more people are going to want to experience AI. Fedora.. no that's not the name. But it just added voice ai response for any llm you're running through it. And those things need ram. As the singers lore feature rich even passing off the work to a remote server farm the ability for it to present and do more with that data requires more and more ram.
If you want to tinker with ai 32 gig is entry level. I'm about to add 64 gig to my system just to get more out of a local AI engine.
I probably should have qualified that for use on Windows systems....
I'd kind of gotten to the point that the computer industry had already gotten to the 16/32 gb point already.
Anything less is super budget.
I've been running 32 GB with my desktops, and most of my laptops for more than a decade so it's probably just a short amount of time until I start moving to 64 GB. Granted, I only game and do relatively minor things but that strategy has worked well so far.
On the flip side, I've kept the workstations at 16 GB for a while now and that has panned out well also but have on some rare occasions seen office apps, and browser tabs, and updates, give a memory error but I can count those occurrences on one hand w/o repeating in 10+ years as well.
The out of memory error (at least with Chrome) doesn't appear to be related to system memory - I'll get that when I'm running at ~40gb of system memory used (out of 64gb). It's almost like there's a "lower memory" limit of around 640KB...
Good to know.
Excel starts to crap out on me when I'm working with datasets over a few Mb.
Not getting anywhere near impacting total system memory. Just old legacy code I suspect
Yeah, I've seen some strange things with various sorts. Basically falls on its face sometimes and I end up having to manually move something and for kicks I'll do the sort again and it keeps whatever change I made even though it couldn't figure it out beforehand.
Excel has been awful on my current system. Not a ram issue I can assure you.
Even with small 1 page sheets used for notes or very simple calculations it will freeze from time to time. I am wondering if having networked drives is a part of this - though even if the file is local it still happens.
Try turning auto save off and see if that helps.