Surprised nobody seems to be talking about it yet so either I suck at search or I guess I’ll start:
I have an original, first-batch 13" from back in 2021. It still works well - power could be better, and I put a little ding in the display, but neither rise to the level of buying replacement parts yet. True to the promise, it’s kind of boring, keeps chugging along, and I have no need to replace it yet.
BUT, that AI 300 does look very nice. So what has changed over the years since then? Can I just buy a new mainboard when they’re available and slap it in? Will I have to upgrade the display to the higher-res one from last year? Would it require a new battery?
I guess - is there a guide maintained anywhere about deprecated parts or upgrade configurations that work and don’t?
With the original 11th gen Framework Laptop the pre-built models shipped with the Intel AX201 Wi-Fi card, which is Intel-proprietary and will need to be replaced. If you bought a DIY variant it came with the AX210 (or no card at all, IIRC the DIY model of that generation had an option to bring your own Wi-Fi card) which is a universal card that you can probably reuse. All newer models (except the Chromebook) shipped with universal Wi-Fi cards (regardless of DIY or prebuilt).
If he has the AX201 (ex. If he bought it prebuilt) then the old card won’t work.
Officially AMD recommends using AMD-branded Wi-Fi cards with AMD CPUs (just like Intel recommends using Intel Wi-Fi cards with Intel CPUs). Framework parrots that recommendation.
IIRC Framework had that same text on the Ryzen 7040 series mainboards and plenty of community members are using Intel Wi-Fi cards (mainly the AX210).
The only cards that cause issues are cards that rely on CNVio2 (Intel’s proprietary protocol) such as the AX201 and AX211 (but not the AX210) or Intel’s Wi-Fi 7 cards (which seem to have some firmware incompatibility with modern AMD systems).
So it sounds like I just upgrade the motherboard only (if at all) and chill until I actually have wifi 7 to bother upgrading that.
Honestly, my biggest problem with the framework 13 is that it’s kind of just a boring little workhorse, and I’ll never have any excuse to upgrade to the latest and greatest unless it’s actually justified.
And RAM. The newer CPUs use DDR5 (instead of DDR4).
From what I’ve seen WiFi 7 cards do get a moderate (~10-20%) bump to bandwidth when on Wi-Fi 6/6e networks compared to WiFi 7 cards. So there is still a benefit, but not huge.
As a user of the original 11th-gen Intel into the AMD upgrade, it was significantly worth it. I have very high hopes for this upcoming AMD upgrade over my current AMD 7840U. If I were in your position upgrading from the 11th-gen would be a certain increase. (My work Framework is still 11th-gen Intel and I see a night-and-day difference already.)
Curious, I’d imagine the battery life would benefit from the lower resolution screen (compared to the new higher resolution ones). Can anyone comment to that?
I have the FW13 12th gen intel, and really the battery last 2-3 hours on wifi. Even when new, May 2023, it never last more than 4 hours. I know that expansion cards could drain the battery, but I’ve been through these tweaks already.
What would be the expected battery life? Would it be woth taking the occasion of buying a new battery, along with the AMD AI 300 mainboard upgrade?
This. The AI300 isn’t built to be faster than a 7840U EXCEPT in AI tasks. The AI300 370 is 8 skinny cores and 4 full fat cores, which is probably a great idea, but is performance terms it is a sidegrade. The real upgrade is the 80 TOPS, but if you’re waiting on software to use that upgradeuntil a new hardware generation comes out you didn’t benefit. I very much want a local AI/LLM, but software is still all over the place on linux. JeffSer’s Alpaca and IBM’s push to get their LLMs into Fedora aren’t user friendly yet, so there’s still a big hurdle before most of us can really use the extra power of the AI300.
The main contributor for performance is RAM bandwidth, as long as it’s DDR5 5600 there isn’t an upgrade there.
Software side, AMD is really, REALLY bad at it. I have given up on ROCm, Torch or ONNX support for thr 760m. Vulkan does work, but only llama.cpp has Vulkan binaries. Supposedly AI300 series has support.
I’m likely going to upgrade for the Intel 300 series motherboard if Framework makes one. Intel 200 series is pretty decent, and Intel is the ONLY company other than Nvidia that makes drivers for their accelerators. To this day the NPU inside my 7640u is literally dead silicon. I root for AMD, but so far making useable acceleration has eluded them…
Intel 300 series is rumored to have Xe3 iGPU, and more importantly a NPU that works with OpenVINO. It should allow me to do LLM on the go with much lower power consumption and possibly run diffusion on my Laptop thanks to having lots of system memory.
I know that RAM I/O is really important, but the ~70 TOPS from the TPU should be able to outperform the 8/16 full fat CPU side of a 7840U, particularly in performance per watt (i.e. maybe overall performance is pretty close, but the R9 AI 370 could theoretically match the 7840’s performance with just the TPU but much less power draw AND use the TPU and 8+4 CPU cores together for even more total performance, presuming that the RAM doesn’t bottleneck it). However, where we sit now, the TPU is a bunch of wasted silicon real estate as getting even GPU cores involved isn’t really feasible, much less TPU. I think that kind of speaks to my intial post, wherein we’re still in a software gray zone kind of like the 32 to 64 bit transition: there’s clearly more performance to be had, but it’s case specific and heavily dependent upon software development that is still in the works.
The same goes for everyone’s hardware…..I want to see it working in my exact use case (local LLM on Fedora) before I am satisfied that it’s a real upgrade.
The other upgrade in the HX 370 is the 16 CUs in the iGPU. Between having more CUs and the upgrade to RDNA 3.5, it’s a significant improvement in graphics performance.
The CPU is a step up from the 7840U, though not as much. The 50 TOPS NPU is a big upgrade from the 8 TOPS NPU in the 7840U, if you can find anything that actually uses it.
That’s probably by far the biggest one (especially the 16cus, rdna3.5 itself seems pretty situational).
And the much faster avx512 stuff if you are using that. 8 full fat zen4 cores are just pretty hard to beat.
That’s the big if
At least the 370 while also being a lot more expensive is a clear upgrade, the 340 is pretty much entirely a downgrade and the 350 is more of a sidegrade depending on what you use it for.
Of course if some killer feature use for the npu shows up that is a different story.