I currently have the AMD Ryzen 5 7640U CPU and want to upgrade my mainboard to Ryzen AI 9 HX 370 (preferred, but worried about noise & heat) or Ryzen AI 7 350 (a decent upgrade from my current mainboard). For folks who have gone through this upgrade, can you please speak to how the two CPUs compare on (a) noise, (b) heat, (c) performance, and (d) battery life?
I’m not interested in benchmarks, just observation and perception based on real-world use (of folks who have upgraded from 7040 series to the AI series). Real-world use that involves software development, VMs, Video/Audio editing, and even local AI would be awesome.
You’re likely on-point here. My understanding was that the Ryzen AI 7 350 was closer to, and an improvement over, 7840 except in raw GPU perf where it’s a slight downgrade. The AI 9 HX 370 is the true upgrade, and that’s what I want to get, but am concerned about its perf, battery life, noise and heat signatures in the FW 13 chassis.
In cpu wise the ai has the same number of cores but that is made up of 4 better zen5 cores and 4 likely worse zen5c cores vs 8 full fat zen4 ones and the gpu is quite a bit worse. It is not as bad as the 340 vs the 7640u though (unless someone really has an application for the npu idk why anyone would pay for the 340 as long as the 7640u is available).
Even the 370 isn’t that much better compared to what you have. If you really need more multicore performance or play stellaris 24/7 go for it, otherwise maybe stick with what you have for now.
Personally even if I did want the 370 after the experience of early adopting the 7480u I’d still wait a couple more moths before even considering it. Battery live and idle power was pretty bad there too early on.
I upgraded from the R7 7840U to the R9 AI HX 370 and I have found that gaming performance improved (7840U was still borderline, HX370 is playable); however the power usage and heat output of the HX370 is quite a bit more while doing simpler tasks (apps open but otherwise “idle”).
This will hopefully improve in time. (I run Linux, and 7840U took a little while to optimize.)
This may get better with time, the 7480u also had a pretty rough start in that department. Would be nice if someone would start shipping stuff in a semi finished state but it is what it is.
The 7640u may have it beat in that department, especially if we are talking mainboard only prices. In the 13 we are still more power limited than core limited so having 4 fewer gpu cores doesn’t hurt as much as if it was unconstrained.
All great points, thank you! Based on the discussion so far, I’m leaning towards skipping the current gen . Frankly, the 7640u handles my devel- and VM-heavy workflow perfectly , but FOMO is a powerful driver (not out of the woods yet ).
Coming from an M1 Pro, I’ve been pretty surprised at the thermals on this HX 370. It takes a lot to get the fans to come on with the M1 Pro, but simply watching a YouTube video at 1080p (not even 4K) gets the fans spinning on this device. Running Fedora 42 with kernel 6.14.8-300.
Have also been surprised by relatively poor performance. On paper it looks like the 370 should generally beat the M1 Pro. But with an Ollama RAG based project I am working on, same code on the M1 turns out around 37 tokens/s processing, and this 370 is around 12. On the Mac Ollama is using Metal GPU support, and I was able to get Ollama on this lappy to report it is using the Radeon.
But as others have said, it’s early days for this chipset on Linux, so I am sure things will only get better.
While I can’t speak to the thermals of the new chips, I imagine it wouldn’t be that different from 7840u given that they’re both 4nm chips.
Regarding the performance in LLM tasks, no amount of optimization will make it come close to M1 Pro, not to mention M4 Pro/Max, M3 Ultra.
This is due to LLM tasks being bottlenecked by the memory throughput.
FW laptops use 2 channel SO-DIMM DDR5 5200MT/s memory which achieves total throughput of 41.6GBps*2 = 83.2GBps.
MacBooks use soldered memory, which is not upgradable but much faster.
M1 Pro has 200GBps, M4 Pro has 273GBps, M4 Max 410GBps and M3 Ultra 800GBps.
So I’d expect, in LLM inference:
M1 Pro to be 2.4x faster
M4 Pro to be 3.3x faster
M4 Max to be 4.9x faster
M3 Ultra to be 9.6x faster
LPCAMM2 will only net 120GBps, so not much improvement on that front either…
At the moment you either pick repairability or speed, not both. This is why Framework Desktop has soldered memory (I believe around 256GBps).
I don’t have numbers but I remember it being quite disappointing when I first got it and I was nowhere near the first batch.
At least I learned my lesson early adopting XD, we’ll see if I remember when zen6 stuff comes out.
Then again I bought a 7900xtx relatively close to launch so idle power wise I am used to much worse, 120w just for having the display on, sounds reasonable right.
On my HX 370 running Ubuntu 25.04, if I watch youtube or f1tv with the power cable attached the fan spins up to an annoying but bearable degree, but it does not spin up when my power is disconnected. Anybody got an idea on how to keep the fans quieter when my power is connected?
installed fw-fanctrl and configured to always run the fan at low RPM, but slightly faster when plugged in, it’s now never audible under normal conditions, you can fine tune the RPM curve to whatever works for you
in fedora changed via “tuned” the profile for when charging to be the same as when discharging ie “balanced-battery” (I think that’s the name)
. Frankly, the 7640u handles my devel- and VM-heavy workflow perfectly 
, but FOMO is a powerful driver (not out of the woods yet 
).