I dread the idea of rebuilding my DAW/Dorico computer. It runs very well, is completely reliable, and still has plenty of memory and SSD capacity. But it is now 10+ years old, running Windows-10, which is at end-of-life. And although it has a video card with GPUs they don’t seem to be usable with Spectralayers.
I have 3 monitors on this machine (2 large ones plus another smaller one where I stash miscellaneous windows that I want to keep open but don’t need right in front of me.
I am a little confused about video cards. It seems some system suppliers quote the Nvidia chipset (e.g. 3060, 4060, 4070, 4080) as opposed to quoting a specific card. The cards don’t always come with 4 ports, so I’m a little uneasy about how to assure I can connect at least 3 monitors.
But apart from that, my real question relates to the GPUs. I use SL unmixing a lot. I use it for programs that may run an hour or two. On my current system it might take 8 hours to unmix one WAV file at the middle quality setting, and I might have 4 or 5 WAV files from the live program that need to be unmixed. I can use batch processing to run while I sleep, but some of these jobs could take well over 24 hours.
If I am going to go through the trouble of rebuilding my studio computer, I’d like to get a GPU package that will speed things up a lot. However, the difference in price between a 3060 and a 4090 is almost $2000 just for the video card. How can I figure out what will be most cost-effective? I don’t want to spend an extra $2000 if it would really make much difference. What are reasonable expectations for these various GPUs running unmix with SL?
According to this chart, the best performance/price would be the RTX 4060:
But if you can afford it, I would recommend going slightly higher with the RTX 4070, or even a RTX 4080. The RTX 4090 is not worth the perf/price if you want to balance your budget.
I’m going to go out on a whim and tell you something contrary to @Robin_Lobel and that is to go “Arm”. The new “Arm Qualcomm snapdragon X Elite” chips are fantastic and it will get better if more consumers(such as yourself) help to support that architecture.
The TRUTH is that you don’t need high end Nvidia gpu chips for heavy gpu tasks. “Arm” chips are just as capable as a Nvidia RTX 4090 and the secret (that they don’t want you know) of “Arm” chips being just as capable to Nvidia RTX 4090 chips is optimization… “Final cut Pro” runs well on the new m4 chips because it is well optimized. The problem with Nvidia chips are that it is way overpriced while you’re receiving little-to-nothing in return, another problem is that Nvidia has become so complacent that they’ve stopped innovating and creating better technology for the consumer. If you(the consumer) continue to keep supporting Nvidia (and their lack of progression) then you are essentially supporting a corporation to stay wealthy/rich and do nothing to improve. If you(and the rest of the world) support the new “ARM” architecture (and stop buying Nvidia chips) then you’re forcing Nvidia to stay competitive and innovate and create new technologies. So far (in the past 10 years) Nvidia hasn’t innovated anything interesting other than Ray-tracing and their neural networks, and therefore I don’t believe it is right for you (or anyone) to continue to support them.
The TRUTH is that all those SDK’s and Cuda Architectures from Nvidia can be ported/bridged over to the ARM architectures (and with all these deep learning/machine learning technologies there is no excuse for Nvidia not to use those A.I. technologies to do it). If you (and 90% of consumers) stop supporting Nvidia it would force Nvidia to start porting/bridging over to the “ARM” architecture and force them to stay competitive and keep innovating and force them to optimize specifically for “ARM” chips.
It should be given the fact how accepted it is already and how many consumers have already adapted to it.
From my basic presumption of the op (and the post/topic in-general) I get the impression that the op doesn’t have an unlimited budget (based off the fact that the op’s pc is 10 years old). Therefore that is why I recommend the “ARM” architecture because they are 3 times less then the price of a Nvidia gpu and provide the same quality as any chip (in-terms of operating applications and executing calculations/functions). I cannot and will never recommend getting a Nvidia graphics card because honestly they are not that good. Nvidia has subtlely fooled the pc world into believing that you need high end gaming cards to do ray-tracing and that is simply a marketing ploy in order to get you to make Nvidia richer. Also (honestly speaking) the Nvidia gpu’s have not improved nor evolved over the past 10 years, for the past 10 years Nvidia has made SMALL incremental improvements in their gpu processing power while having the audacity to ask the consumer for twice the money. If Nvidia made any sort of effort to improve their architecture to where their chips output twice the processing power for half the tdp then I wouldn’t mind advocating for Nvidia, but nothing hasn’t improved about their chips and for you(the consumer) to continue to support Nvidia would mean to support a complacent corporation that stifles innovation.
I appreciate the insights you shared about ARM. Of course, I don’t want to waste money. But I also don’t want to buy something that may not actually perform well in a variety of cases. My immediate concern is getting faster processing on the SL AI processes (especially unmixing).
But of course, many software products in the music tech space are leaning heavily on AI. As a practical matter, I’d expect to get 5-7 years of good service from my next system, and that probably revolves around the GPU. Is ARM even a real thing yet? DO you think it will run SL (and other music software that uses AI) fast right out of the box, or would we be looking at years before the vendors might optimize for ARM?
It all depends on what you choose in Preferences > System > AI Processing Device.
If you set it to CPU (default choice) all is computed on the CPU.
If you set it to GPU, all is computed on the GPU.
I have a Intel ARC A750 and get close to realtime when unmixing stems. Not too shabby for $250 or less. So Nvidia a is not a must, and it will certainly get better as drivers/AI processes get better in the future.
