Hi there all!
So, Ive been keeping a close eye on CPU prices while building my mobile studio, and I would like to open this discussion with a quote from the funtime show (not literal)
- There is no intention to optimise for Ryzen Zen+ until there is a clear performance/price difference as intel has worked great till now.
A bit of history: I would like to emphasize here that from a technical standpoint, Zen first gen had latency issues which heavily influenced Vmix performance compared to Intel in the same price point. this reflects in all benchmarks as well. With Zen+ release, the specific items Vmix has been looking at haven't specifically been addressed, preferring the status quo because intel works (as said in the october funtime show).
My goal here is to move forward and get an actually informed thread with news about the CPU market.
Price comparisonSo a quick pricewatch comparrison, in the netherlands
18/09
i7 8700k - 400 euro (average)
Ryzen 2700x - 350 euro (average)
14% difference
18/10
i7 8700k - 486 euros (cheapest)
Ryzen 2700x - 318 euros (cheapest)
35% difference
This does not take into account the higher cost of getting the i7 to run, the Ryzen comes with a pretty good stock cooler, the intel doesnt (+-30 euros for a mediocre cooler) and the intel motherboards are slightly more expensive too.
Performance and OverclockingI have no way of getting Vmix performance metrics (well, maybe next week if I have spare time, Ive got 2 identical systems bar the CPU) but I can look at gaming and desktop performances.
Taking a quick average (not looking at specific games or workloads for now) I get with
https://cpu.userbenchmark.com/i7 8700k - 108% of average
2700x - 99% average
So thats a 10% difference
Looking at Multicore/singlecore performance:
i7 8700k - 137/1080 points
2700x - 119/1295
For a -13% / +20% performance difference
To quickly address overclocking, I personally dislike overclocking for Vmix, since I prefer ultimate stability. These scores include a mass of stock benchmarks mixed with OC benchmarks, always looking at the spread. That said, the i7 does overclock slightly higher, but the 2700x responds a lot more to memory clock.
The recent rebranding btw gave minimal performance upgrades for intel for a doubling of price.
Vmix and core count vs clock speedThis is something Id love to see a more technical explanation of if available.
All Ive heard was that Vmix does benefit more from higher clock speed compared to core count. This means that the i7, and possibly older gen intels with higher clocks will outperform the 2700x and maybe even TR (which should be compared to i9 or x series CPU really) in pure render times.
And this is where my own workflow differs from these guidelines. I run modular productions with plenty of external stuff. So I benefit a lot from having more cores to run other programs without affecting Vmix too much. On my main rig Ill be running several NDI monitors in addition to Vmix, some control software, and ingesting plenty of NDI feeds.
On my Sattelites I often run Vmix as well as NDI re-encoding/distribution, Content, etc etc, where multi threading is a real boon.
Several questions to the devs- Does vmix decoding NDI or in general make use of multithreading in some form?
- Are there plans (Vmix 22) to improve Vmix multicore/thread usage, seeing as both intel and AMD are moving to more cores over pure clock speeds
- Are you planning on taking another look at AMD Zen+ and Threadripper performance if intel shortages stay as they are? Even if only to get some metrics out to us.
So why this threadI invite you all to share your thoughts on the current developments on the CPU front.
Intel production shortages are a real thing now, with AMD going strong on sales and looking better on developments for the next generation compared to intel. If anything, it will be an incentive for intel to step up and/or lower prices, which is great for us all.
I am fanboy agnostic, I use what works for the system Im building, but I dont mind taking a risk here and there to see if something works, thats just who I am (hence my TR2 system). I got a mac for Qlab and Windows for Vmix as an example. But I would like to gather some ideas and metrics out in the open if possible.
If anyone has a way of metering Vmix render times + usage statistics in a chart or combined number, let me know!