Discussing Game Optimization, Zen 5 X3D, Arrow Lake, RDNA 4, and PS6 w/ Taylor Hadden
Added 2023-09-02 03:51:43 +0000 UTCNext Broken Silicon will see the return of Taylor Hadden, a Senior Tools Engineer who just swapped from working for Obsidian within XBOX Studios to working for NetEase Games...and he has A LOT to say about game development and how hardware is changing. Planned subjects include:
- Studios Struggling to Optimize for BOTH 16GB Consoles and 8GB GPUs
- Implications for Zen 5 X3D vs Intel Rentable Units in Gaming
- Concerns about AI affecting Developer's and Artist's Jobs
- Thoughts on the PS5 Pro & PS6
- What's required for a "Next Gen" VR Experience
- Upcoming and RDNA 3, Lovelace, RDNA 4, and Blackwell Releases
...and of course anything else you all want to ask him! As a Tools Engineer Taylor is someone in a unique position to talk about just about anything that goes into game development - whether software or hardware - he interacts with EVERYONE to get games out the door! Do not hold back on your questions, just be sure to be respectful, concise, and use good grammar to have them considered! You have ~24 hours to submit below!
Comments
What could Sony add to the PS5 Pro that’s not just additional Compute power(+16 units), or added CPU/Ram speed increases. If I was AMD, Id be looking at Nvidia and their DLSS adoption about to become significantly increased due to the switch 2. The developers and sales that the PlayStation brings to the table, what would you two like to see AMD push to have more optimized adoption to games. Developers coding with VCache in mind giving them a leg up on Intel? Hardware accelerated FSR3? Or something else entirely? I’d like to see something out of the left field and if the Pro does have the rumored 60CU variant, to run 52CU and customize 2 WGPs to be able to do mixed precision arithmetic and have partner with Nvidia for a custom software version of DLSS. Question 2 - with the significant advancements ARM has made with the processors over the last few years, will there come a time where(especially with consoles) they drop the CISC CPUs for RISC CPUs?
Travis Gooding
2023-09-04 00:33:41 +0000 UTCHey Tom and Taylor. I was curious, right now there's a big rift between indies and triple a studios. Indies seem to have the most exciting concepts and gameplay. But triple a the polish and budget. Specifically on the indie side, how do you think indies can get closer to indies on their own? Reaching double A status or something along those lines
yoda king
2023-09-03 08:36:06 +0000 UTCCan you describe how optimisation takes place during different phases of the game development? Is it something often done from the ground up or is it usually done later in the pipeline? How do engine developers approach optimisation?
qhfreddy
2023-09-03 08:34:54 +0000 UTCHow fine grained are CPU optimisations nowadays, high level Vs low level? Compiler level? What algorithms and processes are often the subject of extensive optimisation effort?
qhfreddy
2023-09-03 08:32:41 +0000 UTCHey Tom and Taylor! I am just getting into developing audio for games, sound design/audio programming and had a question in mind: How much hardware power should I expect to be working with in regards to everything audio? I understand that it won't be remotely close to graphics, but I just want to have the right expectations for what I could deal with in the future.
Money Man Mike
2023-09-03 08:32:02 +0000 UTCTo what extent do you feel that the trend over the last decade of emphasizing high frame rates (above 60 fps) has posed challenges for developers in expanding the level of detail for in game art (in terms of geometry, shaders, etc.)? Would games be more detailed in the PC market if developers could target 30 or 60 fps as they often do on consoles, or is this less of an issue than people think?
Per Lichtman
2023-09-03 08:31:46 +0000 UTCDevs always seem to like to have more ram, is that just because it makes memory management easier? How much RAM is more than enough in that sense nowadays?
qhfreddy
2023-09-03 08:29:05 +0000 UTCHas V-Cache been at all impactful in the direction game optimisation has taken? What about e-cores?
qhfreddy
2023-09-03 08:27:06 +0000 UTCHow would you say optimising for GPUs and CPUs is different? Can you explain why AMD sponsored games don't perform better on AMD CPUs just as often as they do, but on the GPU side it's usually much more clear cut?
qhfreddy
2023-09-03 08:25:46 +0000 UTCHey Tom and Taylor! I know that Tom has been concerned about the consolidation of gaming with Microsoft buying up many large and popular game studios. I see that Taylor is working at Netease one of the largest game companies coming out of China (along with Tencent). I know that Netease and Tencent themselves have been acquiring many studios themselves as China has been approving fewer video games to be made. Firstly, are you two concerned about Chinese influence in gaming? And secondly, how has netease treated you versus Microsoft?
yoda king
2023-09-03 00:58:53 +0000 UTCGames have historically used FP32. For larger/very large games (and in game engines), FP64 has slowly become the main base to place objects in the world with acceptable precision. Will we see a need for more FP64 throughput in GPUS in the future, or is this just for global positioning and FP32 will remain the main workhorse?
KingHarkinian
2023-09-03 00:02:40 +0000 UTCOhaio Tom-san and Taylor-san 1.How does a studio/ dev pick the minimum target cut-off point for a game? As game requirements goes up, more people on the lower end would be cut-off from playing that certain game. Alienating part of your potential audience is bad on one hand, but on the other hand, It's quite clear bring in 2023 gaming experience to a GT 710 is "impractical". I noted that game development takes 5-6 years, or more, and a lot can change, for better or worse, in between the start and the end. How and when would the studio predict and pick the minimum specs, and would a studio goes back and retroactively "fix" the game if the future hardware prediction turns untrue (Eg: people still holding on to old hardware for longer than expected or future hardware did not improve as fast as expected) ? 2.Would game start using more cores beyond 8C16T and the use of "E cores" It seems like most game still does not care much about E-cores, but as of now, when an i5 has upto 8 E-cores, would devs starts actively using them to offload task from P-cores? What pop up in my mind are on the fly decompression of texture for example, substitute for the dedicated hardware in consoles. If there is any other potential uses of E-core in games, I would love to hear about it.
Dark Side of the Force
2023-09-02 23:49:16 +0000 UTCHello, "Optimisation" is thrown around a lot. But what are the most recurrent optimisation targets? What should be targets, but demand too much time/effort to optimise to a sufficient degree?
KingHarkinian
2023-09-02 21:57:01 +0000 UTCIs software demands outpacing hardware to the point where games are designed with upscaling technologies to achieve playable framerates?
Steven M
2023-09-02 17:43:43 +0000 UTC[Copied from Discord Reader Mail - topic seems especially relevant given the suggested subjects] Hi Tom and Taylor, Hardware Unboxed recently uploaded a video covering Starfield GPU settings optimizations. In this video (https://www.youtube.com/watch?v=40iwgUjBmoA&t=216s) they said that Starfield would typically utilise only about 8GB of VRAM, and even when testing with a RTX 4090 this rarely exceeded 8GB of VRAM utilisation. Buildzoid has uploaded a video referencing an article by PCGH (https://www.pcgameshardware.de/Starfield-Spiel-61756/Specials/cpu-benchmark-requirements-anforderungen-1428119/), showing CPU performance in Starfield and made the speculative observation that the game appeared to be memory bandwidth limited, most obvious when they compared the 12900K vs 13700K, where the 13700K using DDR5-5600 outperformed the 12900K using DDR5-4400 by 28%, which pretty much perfectly correlates with the memory bandwidth increase. (https://www.youtube.com/watch?v=s4zRowjQjMs&t=490s) I have a couple of thoughts/questions on this topic: 1) Is this Bethesda's attempt at Starfield trying to remain compatible with 8GB VRAM GPUs, at the cost of needing to swap resources more frequently between GPU and RAM? The result of which (if true) seems to be exposing memory bandwidth limitations, especially on Ryzen processors due to Infinity Fabric bandwidth limitations. 2) This seems like a pretty fundamental limit in Zen 4's (and earlier) Infinity Fabric die-to-die bandwidth. I know there is speculation about the presence (or lack of) a new Desktop Zen 5 IOD - but if more games end up being bandwidth dependent and if Zen 5 reuses the Zen 4 IOD on desktop, I would assume this issue will persist? Can AMD do anything to rectify this, such as a new Zen 5 IOD for desktop, or is this baked in deep enough to the architecture of Zen 5 that this will be an issue for Zen 5 too? Even with the recent updates to the AGESA, realised memory bandwidth is limited due to the underlying Infinity Fabric bandwidth.
raysin bisket
2023-09-02 17:10:20 +0000 UTCHello Tom and Taylor. I have recently been thinking more and more about how APUs seem to be getting better and better for gaming, and how the Steam Deck's APU still outclasses the Z1 extreme when it comes to low powered gaming. My guess as to why this happens is that the four extra cores on the Z1 extreme are eating extra power that the GPU could use, and thus limiting performance. Thus, I believe that using C-cores might prove to be much more useful that full cores as they would provide better efficiency per core and free some power for the GPU. For example, instead of having a power hungry eight core and a lower powered GPU, we could have a generally more balanced APU that has eight C-cores, and a much more powerful CPU, or maybe a V-cache die instead for more premium products since not only are they more efficent, but also more powerful.
3-Valdion Dreemur
2023-09-02 15:55:06 +0000 UTCHey Tom and Taylor, Intel's new GPU Busy PresentMon statistic seems like a powerful tool for developers looking to optimize their games. What other sorts of tools do you want to see the hardware manufacturers come up with to help game developers with optimization?
Cleansweep
2023-09-02 14:23:08 +0000 UTCHi Tom and Taylor. Is Intel going to rentable units basically a confirmation that their big.LITTLE concept a fail or is it an evolution of that idea? Also what will AMD do? Will they go ZEN-C route or do you think they have their own rentable units in the pipeline?
QuickJumper
2023-09-02 09:17:20 +0000 UTCHey Tom, hey Taylor, when do you think we'll finally embrace/transition to truly utilizing all available cores for games? Working with Unreal Engine 5, it's extremely hard to offload some game logic to other threads, since a lot of available functions from Unreals' framework don't work well or won't even accept anything that's not executed in the main game thread. And currently, game logic gets more and more complicated bottle-necking performance, not just graphics. I could see other engines having the same problem, so what's the best way forward? What are the challenges running logic outside of the game thread?
Dave Scholze
2023-09-02 08:17:13 +0000 UTCWhat are the three biggest sources of a user's game crashing, and do you have some quick tips to indie devs for preventing those crashes?
FrogOnUnicycleVape
2023-09-02 07:13:57 +0000 UTCHow do you think the silicon cost of Blackwell will compare to current gen and do you think NVIDIA will still release a 5090 that offers a massive (maybe even 2x) increase over the 4090 now that it seems like AMD have scrapped N41?
Cloud
2023-09-02 04:10:28 +0000 UTCHey Tom and Taylor, what's the next level of performance that you think game devs are going to shoot for? High refresh with current graphics, higher resolution, more technologies I can't even think of? Cheers!
Xavbeat03
2023-09-02 03:55:25 +0000 UTC
