Marvin

daveinpublic said:

So if someone offers a game for free, and it gets 1,000,000 downloads, and it made the developer $200K in a year....

That developer went from only giving Unity $20K total, to now $200K per month. In one year it will cost them 120x more. It will bankrupt them.

That's the fastest change in price structure I've ever seen. Your game went from making you $180K in a year, to costing you $200K each month.

I think they've messed up on their pricing structure. If you were making money from the games, you'd likely have a Pro license, which is about $2k/year.

The Pro license gives you 1,000,000 free installs. If you get more downloads than this and make more than $1m, like say 100m downloads, the charge per download goes down to $0.02 for most of the other 99m downloads. That's still crazy though at ~$2m because it could be more than you make if it's a free-to-play game.

These changes happen when a company has been mismanaged and is struggling financially. Unity has only recently made a profit and has been losing money all the time until then and not a small amount - billions in losses.



Just like what happened with Reddit, they come up with a crazy plan on how to make money and are prepared to burn every bridge because they are in survival mode.

Reddit was in the position to do it and will be ok. Unity isn't in the position to do this. They aren't the leading game engine developer. It won't be easy for every developer to switch to another engine. Genshin Impact is built on Unity for example and cost hundreds of millions of dollars to make but a lot of the cost is asset creation and code, which doesn't have to be remade from scratch.

Unity earnings are here:

https://investors.unity.com/financials/sec-filings/default.aspx
https://finance.yahoo.com/quote/U

The worst part isn't so much the pricing but that they can change the terms whenever they want, to arbitrarily harmful terms and game devs investing so much time and capital have to deal with it. Once that trust is broken, devs will abandon the platform. The question is to what? While Unreal is an obvious choice, it's a very heavy engine for Indie developers and is mainly designed for high-end AAA games with a steep learning curve.

Maybe this will be a wake up call for game studios that there needs to be an engine separate from an individual company's or even an individual company owner's (Sweeney) problems. A lot of companies already make their own engine, they could easily all share an engine and contribute to it. It would be best to be a minimal core engine that companies can build out with their own extras.

The core engine would mainly need to cover the renderer and the target platform deployment, which are the hardest parts. Once you can handle realistic lighting and shading and deploy to multiple devices, the rest can be handled by small 3rd party code libraries.

AppleInsider said:

The A17 Pro's new graphics capability is enabling a new generation of mobile games including Resident Evil Village, Resident Evil 4 and Death Stranding, all expected later this year. Also on deck is Assassin's Creed Mirage, expected in the first half of 2024.

These are good titles to get ported over. Nice to see Resident Evil 4, hopefully this will make it to Mac too.

A16 is around half the performance of M1 and about the same as the Steam Deck so A17 will be somewhere between A16 and M1.

M1 can run Resident Evil Village at 60FPS on medium quality:



Steam Deck can run Resident Evil 4 at 40-60FPS at 720p:



Given that there is hardware raytracing now, that's likely to make it to Mac too with M3. That will help beyond games for raytraced rendering. They noted the speedup as 4x over software rendering so this will close the gap between the M-series Macs and higher-end Nvidia GPUs (M2 Ultra on page 2 at 1/4 the NVidia 4090):

Blender opendata

This won't be until later in 2024 but it's a very useful hardware feature. Nvidia had examples of how much their hardware RT improved rendering speed vs software:



It doesn't apply to the whole frame time but it reduces the lighting rendering part significantly, leaving an overall 2-3x speedup over software raytracing. Software raytracing isn't used much in games but for games, hardware raytracing increases the realism of the lighting, reflections and shadows:



Bringing AAA games to mobile is also a good strategy. The Mac doesn't have the userbase for gaming but iOS does so if AAA devs will be willing to port to iOS, it's a lot easier to target Mac after the iOS port is done. The full desktop titles are around 60-80GB so they will likely lower the texture quality for mobile and this will help with the lower memory available.

charlesn said:

Hey Andrew: since they launched, it has been reported (and you cover this in your article) that Apple Silicon machines use memory much more efficiently than their Intel predecessors so that less memory is needed than in the past. Are there any rough estimates of how M-series memory compares to Intel... meaning: XGB of memory in an M-series machine is roughly equivalent to having XGB of memory in an Intel machine. Would appreciate your thoughts about this. 

On a system with split memory, there's some duplication of memory objects when both CPU and GPU work on the same data. There sometimes still has to be duplication on unified memory because the CPU and GPU programs don't always address the same data formats.

There's not a consistent percentage gain from unified vs split, it depends on the data. The main benefit of unified is that copying data is faster than copying between two different memory chips.

Intel systems also used to have more memory. The MBPs shipped with 16GB system memory + 8GB video memory = 24GB total. I would say it's reasonable that 16GB unified memory can perform in an equivalent way to 16GB+8GB split.

It's still the case that 8GB total memory is only suitable for basic use cases like document editing. Roughly 4GB is allocated to the system so that only leaves 4GB for all other applications.
16GB is ok for heavier usage like software development, photo editing, film editing.
32GB is best for VFX work, any kind of high resolution image/video work and light AI.
64GB+ is best for AI and running multiple heavy pro apps together. Basic image AI can use 24GB RAM on its own.

Some upcoming smartphones are rumored to have 24GB+ RAM due to using local generative AI models:

https://www.xda-developers.com/24gb-ram-smartphones/

ooloo said:

Side-note:  Have you guys checked out Baldur's Gate 3 for macOS?  It's awesome!  It even runs great on our old intel Macs lol.  Maybe Apple should buy Larian Studios since they missed the boat on Bungie and Blizzard...

Buying a game studio is a hard way to make inroads into gaming. Baldur's Gate 3 is an adult-rated game and took 6 years to make. Larian also wants to stay independent:

https://www.eurogamer.net/baldurs-gate-3-dev-says-its-not-interested-in-being-acquired-but-finds-speculation-flattering
https://gamingbolt.com/microsoft-considered-acquiring-cd-projekt-red-larian-studios-fromsoftware-and-more-in-2021

When game studios are funded by big companies, they tend to lower quality. This has ruined a lot of big studios like Irrational Games and Bioware.

Microsoft has bought a few studios and the studios would probably rather be owned by them due to having a console audience.

Microsoft, Sony, Nintendo have around 15-20 studios each.

When a AAA studio takes around 3 years to make a game, to release one every other month, it would take nearly 20 studios working in parallel and you still have a lead time of 3 years.

A single studio can work if they make a hugely popular franchise like GTA or Call of Duty but buying a studio first and making a new franchise is very difficult and time-consuming and that popularity comes from deploying on multiple platforms.

Apple has the cash to do it. A major studio of 1000 people (x $100k per employee) = $100m per year. Apple could easily run 10-20 major studios if they wanted to. The hard part is finding that many talented people and to make the return on investment (for Mac), they have to port to multiple platforms.

The current AAA game model has been described by many as unsustainable long-term because big budget games have to sell 5-10 million copies at $60 to break even. If a big studio takes 5 years to make a game, the cost can be as much as $300m and marketing on top. This is why Square Enix sold off major franchises like Tomb Raider:

https://www.eurogamer.net/tomb-raider-finally-achieved-profitability-by-the-end-of-last-year

I don't think targeting old games is a bad thing because a lot of Mac users never had access to those games so old game ports helps build a gaming audience. New games are best to have available but it requires regular investment. Apple managed to get Resident Evil Village + DLC but not the newer Resident Evil 4.

https://www.imore.com/mac/resident-evil-4-skips-mac-even-after-village-headlined-apples-gaming-lineup-its-not-good-enough

If the Mac could get a Steam Deck level game library, that would build a good gaming audience. Then publishers would have a reason to make native ports.

This whole Mac gaming topic is like an endless déjà vu. Every time Mac gaming makes a little progress it gets stunted. Other big companies have supported gaming for over 30 years and on Windows, it's possible to play games from 30 years ago. Apple has made a few breaking changes and a lot of developers aren't willing to commit to the platform. Bioshock was on iOS and got discontinued:

https://appleinsider.com/articles/17/01/31/bioshock-for-ios-is-officially-dead-developer-2k-reveals

""In June 2015, BioShock became incompatible with iOS, following Apple's version 8.4 update. As a result, the game was removed from the App Store and is no longer supported," the company's support page now reads."

This is why a compatibility layer can be a better option for older titles but at least have a long-term support option for a game API.

It would be nice to see Apple form an internal game group that oversees how they handle game ports and marketing and have the group work with publishers and they can report at events about new games coming to the platform. 5-10 major game ports per year would be a reasonable target.

BMac23 said:

Thanks for the advice - much appreciated!

However, my question remains: What type of flash -- triple-level cell (TLC) or quad-level cell (QLC) does each Mac Mini model use?

Your 2014 Mac was probably using multi-level cell (MLC) flash that stores two bits per cell. The endurance is worse with TLC and even worse with QLC. The performance goes down as well from MLC to TLC to QLC.

Apple dropped the price of the base model 2023 M2 Mac Mini, so I'm worried it's QLC flash. If so, I can add RAM and bump up to 512 GB of flash storage, and that may not help me much. If the M2 Pro Mac Mini uses TLC, that may be the only way to go, but I'm loathe to spend the extra money.

If I can't get an answer to the TLC-vs.-QLC flash question, I'm probably going to look at non-Apple mini computers. At least I'll know what I'm getting. I've looked already, and the HP mini specs say TLC NAND flash. I really wanted a Mac, but Apple is making it difficult to buy one.

There have been teardowns of some M-series Macs that show they use Toshiba (now called Kioxia) 3D NAND TLC:

https://www.techinsights.com/blog/teardown-apple-macbook-pro-16
https://forums.anandtech.com/threads/apple-silicon-soc-thread.2587205/page-124
https://www.techpowerup.com/267018/gigabyte-announces-updated-ud-pro-series-high-endurance-sata-ssds

""vendor-name"="Toshiba" and "nand-marketing-name"="itlc_3d_g4_2p_256" would indicate that we're looking at 256 Gbit 4th generation 96-layer 3D TLC BiCS NAND flash dies from Kioxia (formerly Toshiba). All of the ≤ 1 TB Q series SSDs in M1 Macs and ≤ 2 TB R series SSDs in M1 Pro/Max/Ultra Macs that I've seen thus far use those same dies."

"96-layer KIOXIA 3D TLC NAND flash memory, with increased endurance rating across the board. The 1 TB variant offers a staggering 750 TBW endurance (for perspective, the NAS-optimized WD Red SA500 1 TB is rated with 600 TBW); while the 512 GB variant offers 370 TBW; and the 256 GB variant 175 TBW."

The drives use an SLC cache that reduces wear, they won't necessarily have worse endurance than older MLC.

Even if the entry mini used QLC, it would likely still have TBW around 100. If there was 1TB written every month, it would last about 8 years and the entry price is $600. Losing $600 in 8 years isn't a huge loss.

The M2 Pro model with 16GB/1TB is $1799 and will have SSD endurance well beyond 20 years for average use.

For people working with 4K ProRes film (~0.5TB/hour), maybe heavy virtual machine creation, docker images, software builds etc, there could be more than 1TB written per month. Writing 10TB per month could wear out the higher endurance SSDs in 6 years. That's why it's best to use external drives for write caches for Photoshop/After Effects/Da Vinci.

If someone was working on a 15 minute After Effects animation, it can write over 100GB per render. Render 10 changes in a day and that's 1TB of writes per day, 20TB/month. Those heavy use cases should use external/disposable drives.