Uhhhm I’m actually not entirely sure VRAM will be a benefit either, apart from speed
I know some iGPUs (like the 680M) can use up to 16GB of the shared system memory on some systems.
Shows it relies on 64-bit PCI-E encoding for 4GB+, so may not be available on the Framework
So we’d need an engineer to confirm if more than 4GB will be available to us
Also this shows lower performance for 8GB in SOME games, but in that case it ALWAYS uses nearly 8GB of system memory, so that just means you would need to get more than 16GB of total RAM, which you likely do anyway for video editing
I don’t know the mechanics of it, and I believe it depends on the program you use, but Hitfilm definitely seems to benefit from VRAM. It was borderline unusable on my old desktop without a GPU despite having 16GB of RAM, and adding a simple 2GB R7 240 was night and day for 1080p videos. Then I built a new PC and started working with 4K, and it definitely seems to use VRAM rather liberally for the editing process, though I think final export and encoding is all CPU. Still, if I can’t easily scrub through clips to get the final product right, I’ll never get to encoding!
I’m not a pro, though. I just like making videos to tickle friends and family (and perhaps most importantly, myself). I could be misinterpreting what I’m seeing in various resource monitors as I work, but for now my understanding is for video editing VRAM is even more important than TFLOPS, and that dedicated VRAM is always faster than shared system RAM. Maybe one day I’ll try unplugging my desktop’s GPU and see what the effect is?
Hitfilm didn’t benefit from the VRAM, but from the GPU. VRAM without a GPU is useless.
If it used VRAM, it used the GPU, probably for the compositing. Encoding depends, hardware encoders (whether on the GPU or CPU, e.g. AV1 on either dGPU or iGPU) are traditionally not always the BEST for optimised video encoding where quality and small size is more important than encoding speed. I suspect for AV1 using the hardware encoders might be best though.
Well it will have to start doing everything on the CPU, which obviously sucks. GPU is better for data crunching, a good amount of VRAM may just help reducing bottlenecks.
dedicated VRAM IS faster, but considering moving between CPU and GPU portion of the shared ram is probably faster, I wouldn’t be surprised if the iGPU holds up pretty well. I couldn’t tell you, would need actual tests.
But you missed my point: You wanted VRAM, then chances are the iGPU will be able to get more than the dGPU. dGPU will likely be better for video editing, but not sure it’s essential. Only way to know is to get the dGPU and compare!
Definitely, and I think that’s what makes what I said sound dumb. I don’t actually believe the VRAM alone is what makes a difference! I think what I was really trying to say is that it’s a more important spec of a GPU for video editing, since I do notice when my 8GB of 5700XT fills up.
Though that doesn’t justify the 7700S, since as you point out it only has 8GB of VRAM while the iGPU will have a lot more at it’s disposal (I plan to get at least 32GB, probably 64). If anything, I’ve put myself on the side of the OP in wanting a “better” GPU, specifically one with more VRAM! Like, if they offered a RX 5000-series with 24GB, I’d probably be perfectly happy with that.
I don’t actually need that though. An 8GB 5700XT seems to work just fine for what I do. What’ll be interesting is will the performance without the 7700S installed be in fact already be better than what I’m used to? I’m getting the impression the performance is on par, if not better, and the 780M will have tons more RAM at its disposal, even if it’s marginally slower.
It really seems to give the CPU most of the work already. I was curious what my GPU was contributing to the process, so I started monitoring via Task Manager and Adrenalin as I worked, and it really doesn’t seem like the GPU contributes much except for generating, maintaining, and displaying the preview of the pending product. Encode seems to just pin the CPU, and system RAM seems to just hold the raw clips. Not to minimize the importance of being able to move through the working file fluidly, and you can tell when the VRAM fills up because when you move to an un-rendered (I guess?) portion of the video, you can see it catching up to where you are now and dumping where you were before.
Again, at least I think that’s how it works. I don’t really know. I should probably stop talking. Nonetheless, I’m pretty sure the 7700S will be a nice-to-have in my use case, though I am certainly going to see what life is like without it given I’m perfectly happy with a 3700X and a 5700XT. I’ll actually be quite happy to know I don’t have to bring my GPU module on the road when space and power are precious, though hopefully it makes enough of a difference that I do use it!
Fair enough. Truly a difficutl decision to make for you, since it relies on some other stuff.
It is NOT certain the 780M will be able to address more than 4GB of the system memory. Given it’s an AMD advantage system I would HOPE AMD made sure Framework supports that, but it is not necessearily a given.
On the other hand, looking at a 7600 vs 5700XT, it seems the 780M, which is expected to be slighly slower than a 7600, is slightly better or on par to the 5700XT in terms of compute - but of course, that doesn’t always translate to media processing. But 780M has a much better encoding engine (AV1) which means encoding doesn’t necessarily need to use the CPU anymore (as long as the platform / target system you deploy to has AV1 support).
Hey, same here. But searching for hitfilm gpu usage it seems you can tell it to use the GPU hardware encoders, others have much higher GPU usage than CPU. That depends on you requirements, a CPU encoder will generally be just a little better in keeping file size down or quality high.
OH! Sorry, silly assumption by me. Alright, another thing I’ll be interested in finding out.
A little, for sure. I think we’ve generally deliberated your point that the 7700S may or may not be a worthwhile improvement and it sounds like it’s a bit TBD, at least for my use case. Nonetheless, I thank you for all your insight, and I yield back the thread!
If these leaks come to pass, I’d be shocked if they didn’t offer the 7800m XT, or at least the lower power 7900s for the framework laptop since it is part of the AMD advantage program. That would be a nice 16GB option, though the raw performance is still going to fall behind the 4090 laptop from nvidia I suspect. Maybe we’ll even see something official from AMD at gamescom. It does seem more likely that nothing official from Framework will come before the official launch of the Framework 16 though.
Same issue here. When I heard that the Framework 16 was going to have a GPU option, I was briefly excited. Unfortunately, when I checked it out, all it included was a low-end Radeon. At what would end up being roughly twice the price of a decent gaming machine, Framework remains in the “not worth it” category for me.
Desktop and mobile GPUs have different specs. At least AMD denotes them with an M or S at the end. For example, the 7700S shares specs with the desktop 7600.
What ends up being the desktop 7800 XT will likely be called the 7900M on laptops, if AMD plans on having a -900 SKU for laptops like NVIDIA.
Note that Framework hasn’t stated for how many cycles their custom connector is rated. So we don’t really know if they’re rated also for 50 cycles or maybe 200. Or maybe 1000. But in any case, it’s still not a hot-swappable connection and the screws and the input module assembly probably can and will wear out over time. Even if Framework told us that their connector is rated for 1000 cycles, I still wouldn’t trust the complete laptop assembly to survive that many cycles and in any case, it’s not really practical to constantly disassemble and reassemble your laptop, and we have seen obvious seams between the input modules in the prototypes, and I assume they can only get worse with repeated assembly cycles.
It’s just providing an easy upgrade function. Cool. I’m happy already.
That said, perhaps the marketing video showing a feller upgrade-racing with the lady sporting the afro might have given a slightly false impression that this is somehow a routine exercise.
…or NRP’s team announces it’s fine for 10k cycles. Eh, whatever, it’s an upgradeable laptop. That’s the main thing for me.
Please read the quote from Nirav again. He was specifying a different connector, a pogo connector, got rated for 2000 cycles, not the Framework Expansion Bay Module connector.
