I’m looking forward to buy some FW16’s but there’s something that is holding me back, and it is the fact that Framework does not provide an Oculink expansion bay module. I would definitely order them right away otherwise. I’m in need of new machines and ideally, if someone can confirm me the following, I could proceed with the order.
Does anyone with an FW16 successfully connected to the M2 SSD sockets a desktop GPU via Oculink and made that work?
I don’t mind having the laptop open for now (while a solution like Josh Cook’s is available OcuLink Expansion Bay Module - #177 by Josh_Cook) - so making that work is enough for me.
I’m also interested in knowing if anybody got a desktop GPU plugged through OcuLink to one of the M.2 slots for SSD drives.
Ideally the expansion bay would be preferable, but meanwhile that solution could work.
However, considering the initial delays for shipping FW16s and the apparently stalled state(for now) of the OcuLink expansion bay project I think it could be a good idea to wait a bit since new AMD processors will be released later this month with some manufacturers announcing new models with them.
Sure, FrameWork’s bandwidth is not comparable, but after the good job done with FW16’s first AMD mainboards I think there’s reasons to believe a new board with an AMD Ryzen AI 9 HX 370 might happen sooner than expected.
My question is about if someone has been able to have an eGPU setup without the expansion bay, just having the laptop open and connecting to the M2 SSD sockets
When the AMD 8040 series was announced, Nirav said something along the lines of “For such a minor change, we aren’t going to be releasing any products that use this chip line, and will stick to 7040.” And something disparaging current trends to slap AI into everything, even when it doesn’t have a general use yet. The “AMD Ryzen AI 9 HX 370” and its siblings are, again, just a tiny update over the 8040 series. I don’t think we’ll see an updated model announced soon (I’d love to be wrong!).
The 890M iGPU is a bigger deal.
But, the biggest issue that I feel is an impediment to Framework adopting newer AMD (and Intel) APUs/CPUs; all of the announced laptops use LPDDR5X memory soldered to the motherboard. Both AMD and Intel have made comments that it is faster and cheaper for them to only support soldered on memory for these chips. Which I feel, for Framework to do an about-face turn and start offering only options with soldered memory is the antithesis of their stated goals and company motives.
I might be basing this off of pre-release rumors, I’ve been busy for the past couple months and haven’t had my ear to the tech news as closely as usual.
Theoretically, Framework could potentially switch to LPCAMM2 memory modules, but .
Personally I don’t think a transfer from a unified CPU config with 8 Zen 4 cores to a hybrid CPU config with 4 Zen 5 and 8 Zen 5c should be considered “tiny”. That is a very big change, with obvious pros and cons.
Pros include dramatically better energy efficiency especially at low TDPs. The 28W HX 370 in Zenbook S 16 punches well above its class, surpassing the 45W (with liquid metal) 7840HS in Framework 16 in many CPU-heavy workloads, according to a review from Phoronix.
Cons is between the Zen 5 cluster and Zen 5c cluster there is a VERY high core to core latency (~150ns), according to MLID and Geekerwan. Geekerwan also have found this latency affected many games that need more than 4 cores, and dragged the gaming performance quite a bit. And the iGPU will be bottlenecked even by LPDDR5X-7500 if the game is really memory-bandwidth intensive (looking at you, PUBG and ZZZ).
It’s up to you if you still want to call these changes “a tiny update”.
As Mark points out there have been substantial improvements/changes.
For the particular use I’m looking after, improvement in power efficiency, and better iGPU (hopefully LPCAMM2, but that might be too much for Framework as of now) make the price a less bitter bite so to speak.
Basically what I’m looking for in Framework for our tiny company is powerful machines(within reason) as a desktop substitute, but energy-independent enough as to be able to work a full day/1.X days when going around.
The fact that Framework was eager to announce they would be skipping 8040 series but haven’t released any kind of statement regarding strix makes me believe they have something in the oven, but of course, without official confirmation that’s just castles in the sky.
Every company has its own rules and procedures and that’s ok. Although Framework is on good track they still have to gain more reputation to reach a wider customer segment and the last thing they need is undelivered promises, but I can’t see how informing of a (very) likely thing to happen under a big question mark so that customers can plan accordingly could ever be harmful.
And going back to the topic, the reason for the interest in eGPU compatibility through Oculink is basically for the possibility of using the laptop as a desktop replacement.
PassMark PerformanceTest [3D only] → Score : 32221 (Average 4070Ti score : 31772) (sorry I was unable to upload my screenshots)
I had to run the error 43 fixer script which took me 1 second and then Nvidia Geforce Experience was able to detect and install the needed drivers, a restart and all running perfectly.
I haven’t tested any games yet, but I’ll do it soon.
I did use the 2280 slot as it’s above the 2230. And removing the input deck plate
If using the 2230, both slots would have been used for the Oculink m.2 adapter, which means my OS should be installed somewhere else (USB stick or storage module)
Thank you so much for the test once more, you should go now to egpu and post the build and maybe link here so that it gets some visibilty haha. I’m sure others will be glad to know about this setup as well
Just to add some impressions from yesterday evening, I was able to run Dead by Daylight at highest settings, 4K, 120hz without any freeze or issue at all.
I played for about 2 hours, the GPU was 65% usage and 160W power consumption on average.
That said, DbD isn’t the most demanding game and the gpu wasn’t used at full load.
So I didn’t achieve to see if the 4x Oculink bandwidth reduces performances a lot or not.
Nonetheless, 4K 120hz freezeless is outstanding from my pov ! (I used the 780m since april)
.
