Didn’t Expect This: Apple vs Threadripper + NVIDIA [YouTube Release]

Youtube Releases
1

Additional Resources:

Connect With Us

Lawrence Systems Shirts and Swag

►👕 Lawrence Systems

AFFILIATES & REFERRAL LINKS

Amazon Affiliate Store
:shopping_cart: Lawrence Systems's Amazon Page

UniFi Affiliate Link
:shopping_cart: Ubiquiti Store

All Of Our Affiliates help us out and can get you discounts!
:shopping_cart: Partners We Love – Lawrence Systems

Gear we use on Kit
:shopping_cart: Kit

Use OfferCode LTSERVICES to get 10% off your order at
:shopping_cart: Tech Supply Direct - Premium Refurbished Servers & Workstations at Unbeatable Prices

Digital Ocean Offer Code
:shopping_cart: DigitalOcean | Cloud Infrastructure for Developers

HostiFi UniFi Cloud Hosting Service
:shopping_cart: HostiFi - Launch UniFi and UISP in the Cloud

Protect your privacy with a VPN from Private Internet Access
:shopping_cart: https://www.privateinternetaccess.com/pages/buy-vpn/LRNSYS

Patreon
:money_bag: https://www.patreon.com/lawrencesystems

Chapters
00:00 Mac M4 Max Studio VS Threadripper and Nvidia
00:55 My Test Setup
01:45 Davinci Resolve Render Results

2

I’m wondering if Resolve isn’t using the A6000 card fully, would be interesting to see one of their recommended cards which might include an AMD. Would also be interesting to include Windows, but I’m guessing the Mac will be faster than most any combination of Windows.

3

Works the same on WIndows and it pins the GPU which I measured using nvtop.

4

And Windows Arm is still only (officially) with Qualcom, missing the boat again.

5

Interesting Video Tom! I’d like to share some experience regarding Resolve as well.

Two years ago we did some testing editing and rendering 8K Arriraw in Davinci Resolve which is very resource intensive.

Originally we started out on Windows but got fed up with various issues with the 25Gb Networking etc., so we moved to RHEL.

Since we were able to cache the entire project in ZFS Arc and to some extent on the local machines (shared via NFS) the GPUs were the main bottleneck.

Moving to RHEL we saw a nice increase in utilization on the Dual 4090s, performance was measurably better than under Windows.

But at one point we decided to try out an AMD RX 7900 XTX for fun and found we could get faster render times with a single one of those cards running ROCm than with two 4090s running CUDA.
Entire System Power Consumption was also much lower of course.

We compared our stats to some other Resolve Users and our 4090s were not under-performing compared to what performance they were seeing.

Unfortunately ROCm Support in Resolve is pretty lackluster and not all features fully support it so we had to stay with Nvidia.

But it really illustrated how much the performance is down to software optimization rather than pure compute performance.

6

The optimization on Apple are very good. Also the video is full of people saying I should have compared the M4 to a more modern GPU but ignoring that any PC build with a modern GPU would cost far more.

This video even shows how a 4090 can’t beat a laptop running an M2 Max

7

I suspect shenanigans. The card should not be pegged for encoding files. The Ada generation of Nvenc chip does not support 4:2:2 encoding. It’s possible that DaVinci tried to encode or decode 4:2:2 and had to fallback to software. Or maybe the format being converted isn’t supported by the Nvenc or Nvdec chips.