1887224326350635382, 1737126160402665958, 1741871612960600471

2026-08-26 11:40:11 -04:00
1 parent 5ff36e91e5
commit ba3674cbd7
@@ -0,0 +1,57 @@
+---
+title: "@ChipsandCheese9 In theory compiler virtual wave size could have perf affects on modern Intel arch."
+type: archive
+source: twitter
+source_url: "https://x.com/NOTimothyLottes/status/1741876953345405010"
+author: "NOTimothyLottes"
+handle: NOTimothyLottes
+post_id: "1741876953345405010"
+date: 2024-01-01
+archived: 2026-08-26
+draft: false
+tags:
+  - archive
+  - twitter
+  - ChipsandCheese9
+description: "@ChipsandCheese9 In theory compiler virtual wave size could have perf affects on modern Intel arch."
+in_reply_to: ""
+parent_post_id: "1741871612960600471"
+---
+
+## Source
+
+- URL: https://x.com/NOTimothyLottes/status/1741876953345405010
+- Author: NOTimothyLottes (@NOTimothyLottes)
+- Posted: 2024-01-01 17:40:02
+
+## Branch
+
+**1/** **@NOTimothyLottes** ^1741876953345405010
+
+**@ChipsandCheese9**
+
+In theory compiler virtual wave size could have perf affects on modern Intel arch. I always try for 16-wide on Intel.
+
+**2/** **@lamchester** ^1742027618818683331
+
+**@NOTimothyLottes** **@ChipsandCheese9**
+
+Intel's compiler used wave16 for most of the main test loop. Unfortunately I don't know of a way to tell the compiler to use a certain wave size.
+
+![](https://pbs.twimg.com/media/GCzslNKaEAAzNmg?format=png&name=orig)
+
+**3/** **@0x22h** ^1742464332439634405
+
+**@lamchester** **@NOTimothyLottes** **@ChipsandCheese9**
+
+You can set the wave width through HLSL 6.6. However, Intel's new driver has removed Wave8 support for GPUs before DG2, and older drivers can still use Wave8.
+
+**4/** **@lamchester** ^1742578005405507676
+
+**@0x22h** **@NOTimothyLottes** **@ChipsandCheese9**
+
+HLSL is a DirectX thing, not OpenCL. Also there's no way to guarantee a particular wave size is available on a particular GPU. Different architectures have different supported wave sizes
+
+## Related
+
+- Spine: [[archive/threads/ChipsandCheese9/2024-01-01-hello-you-fine-internet-folks]]
@@ -0,0 +1,37 @@
+---
+title: "@ChipsandCheese9 Wish I understood half of wtf you were talking about? I do, however, make it to the conclusions."
+type: archive
+source: twitter
+source_url: "https://x.com/SuperbPhoenix/status/1741887737467224472"
+author: "Superb⚡Phoenix"
+handle: SuperbPhoenix
+post_id: "1741887737467224472"
+date: 2024-01-01
+archived: 2026-08-26
+draft: false
+tags:
+  - archive
+  - twitter
+  - ChipsandCheese9
+description: "@ChipsandCheese9 Wish I understood half of wtf you were talking about? I do, however, make it to the conclusions."
+in_reply_to: ""
+parent_post_id: "1741871612960600471"
+---
+
+## Source
+
+- URL: https://x.com/SuperbPhoenix/status/1741887737467224472
+- Author: Superb⚡Phoenix (@SuperbPhoenix)
+- Posted: 2024-01-01 18:22:53
+
+## Branch
+
+**1/** **@SuperbPhoenix** ^1741887737467224472
+
+**@ChipsandCheese9**
+
+Wish I understood half of wtf you were talking about? I do, however, make it to the conclusions. Good story and right direction. Happy New Year and keep up great work.
+
+## Related
+
+- Spine: [[archive/threads/ChipsandCheese9/2024-01-01-hello-you-fine-internet-folks]]
@@ -0,0 +1,37 @@
+---
+title: "@ChipsandCheese9 I wonder if the throttling is due to local current limits, or temperature limits from unexposed sensors."
+type: archive
+source: twitter
+source_url: "https://x.com/eraser1_/status/1741884551050195168"
+author: "eraser1"
+handle: eraser1_
+post_id: "1741884551050195168"
+date: 2024-01-01
+archived: 2026-08-26
+draft: false
+tags:
+  - archive
+  - twitter
+  - ChipsandCheese9
+description: "@ChipsandCheese9 I wonder if the throttling is due to local current limits, or temperature limits from unexposed sensors."
+in_reply_to: ""
+parent_post_id: "1741871612960600471"
+---
+
+## Source
+
+- URL: https://x.com/eraser1_/status/1741884551050195168
+- Author: eraser1 (@eraser1_)
+- Posted: 2024-01-01 18:10:14
+
+## Branch
+
+**1/** **@eraser1_** ^1741884551050195168
+
+**@ChipsandCheese9**
+
+I wonder if the throttling is due to local current limits, or temperature limits from unexposed sensors.
+
+## Related
+
+- Spine: [[archive/threads/ChipsandCheese9/2024-01-01-hello-you-fine-internet-folks]]
@@ -0,0 +1,39 @@
+---
+title: "@ChipsandCheese9 Is there a reason why you use the texture sampling instructions? How are they different to \"normal\" loads?"
+type: archive
+source: twitter
+source_url: "https://x.com/te42kyfo/status/1742164070722077056"
+author: "Dominik Ernst"
+handle: te42kyfo
+post_id: "1742164070722077056"
+date: 2024-01-02
+archived: 2026-08-26
+draft: false
+tags:
+  - archive
+  - twitter
+  - ChipsandCheese9
+description: "@ChipsandCheese9 Is there a reason why you use the texture sampling instructions? How are they different to \"normal\" loads?"
+in_reply_to: ""
+parent_post_id: "1741871612960600471"
+---
+
+## Source
+
+- URL: https://x.com/te42kyfo/status/1742164070722077056
+- Author: Dominik Ernst (@te42kyfo)
+- Posted: 2024-01-02 12:40:56
+
+## Branch
+
+**1/** **@te42kyfo** ^1742164070722077056
+
+**@ChipsandCheese9**
+
+Is there a reason why you use the texture sampling instructions? How are they different to "normal" loads?
+
+By the way: GCN/CDNA gets 64B/cycle only for 4B or 16B loads (so float or float4/double2). With 8B loads, its only 32B/cycle, because of only 4 addresses per cycle.
+
+## Related
+
+- Spine: [[archive/threads/ChipsandCheese9/2024-01-01-hello-you-fine-internet-folks]]
@@ -0,0 +1,38 @@
+---
+title: "Hello you fine Internet folks,"
+type: archive
+source: twitter
+source_url: "https://x.com/ChipsandCheese9/status/1741871612960600471"
+author: "Chips and Cheese"
+handle: ChipsandCheese9
+post_id: "1741871612960600471"
+date: 2024-01-01
+archived: 2026-08-26
+draft: false
+tags:
+  - archive
+  - twitter
+  - ChipsandCheese9
+description: "Hello you fine Internet folks,"
+in_reply_to: ""
+---
+
+## Source
+
+- URL: https://x.com/ChipsandCheese9/status/1741871612960600471
+- Author: Chips and Cheese (@ChipsandCheese9)
+- Posted: 2024-01-01 17:18:49
+
+## Thread
+
+**1/** **@ChipsandCheese9** ^1741871612960600471
+
+Hello you fine Internet folks,
+
+Happy New Year to all of y'all and with a new year comes some new GPU tests focusing on the L1 and Local Memory (LDS/Shmem) bandwidth of various AMD, Intel, and Nvidia GPUs.
+
+Hope y'all enjoy!
+
+https://chipsandcheese.com/2024/01/01/a-new-year-and-new-tests-gpu-l1-cache-bandwidth/
+
+Branches: [[archive/threads/ChipsandCheese9/2024-01-01-hello-you-fine-internet-folks/2024-01-01-NOTimothyLottes-in-theory-compiler-virtual-wave-size-could-have]], [[archive/threads/ChipsandCheese9/2024-01-01-hello-you-fine-internet-folks/2024-01-01-eraser1_-i-wonder-if-the-throttling-is-due-to-local]], [[archive/threads/ChipsandCheese9/2024-01-01-hello-you-fine-internet-folks/2024-01-01-SuperbPhoenix-wish-i-understood-half-of-wtf-you-were-talking]], [[archive/threads/ChipsandCheese9/2024-01-01-hello-you-fine-internet-folks/2024-01-02-te42kyfo-is-there-a-reason-why-you-use-the-texture]]
@@ -0,0 +1,12 @@
+---
+title: ChipsandCheese9
+type: note
+draft: false
+description: Archived threads by ChipsandCheese9.
+tags:
+  - archive
+  - twitter
+  - ChipsandCheese9
+---
+
+- [[archive/threads/ChipsandCheese9/2024-01-01-hello-you-fine-internet-folks]]
@@ -0,0 +1,37 @@
+---
+title: "@NOTimothyLottes queue to queue syncs via cpu need to go"
+type: archive
+source: twitter
+source_url: "https://x.com/SheriefFYI/status/1737163114670358622"
+author: "Sherief, FYI"
+handle: SheriefFYI
+post_id: "1737163114670358622"
+date: 2023-12-19
+archived: 2026-08-26
+draft: false
+tags:
+  - archive
+  - twitter
+  - SheriefFYI
+description: "@NOTimothyLottes queue to queue syncs via cpu need to go"
+in_reply_to: ""
+parent_post_id: "1737162074172637506"
+---
+
+## Source
+
+- URL: https://x.com/SheriefFYI/status/1737163114670358622
+- Author: Sherief, FYI (@SheriefFYI)
+- Posted: 2023-12-19 17:28:56
+
+## Branch
+
+**1/** **@SheriefFYI** ^1737163114670358622
+
+**@NOTimothyLottes**
+
+queue to queue syncs via cpu need to go
+
+## Related
+
+- Spine: [[archive/threads/SheriefFYI/2023-12-19-always-always-always-use-an-async-copy-queue]]
@@ -0,0 +1,37 @@
+---
+title: "@NOTimothyLottes @SheriefFYI Yup, right now I async copy mesh and textures data, uniform/constat buffer are just a memcpy via a mapped ptr."
+type: archive
+source: twitter
+source_url: "https://x.com/_woookie_/status/1737165821925818879"
+author: "Łukasz Bogaczyński"
+handle: _woookie_
+post_id: "1737165821925818879"
+date: 2023-12-19
+archived: 2026-08-26
+draft: false
+tags:
+  - archive
+  - twitter
+  - SheriefFYI
+description: "@NOTimothyLottes @SheriefFYI Yup, right now I async copy mesh and textures data, uniform/constat buffer are just a memcpy via a mapped ptr."
+in_reply_to: ""
+parent_post_id: "1737162074172637506"
+---
+
+## Source
+
+- URL: https://x.com/_woookie_/status/1737165821925818879
+- Author: Łukasz Bogaczyński (@_woookie_)
+- Posted: 2023-12-19 17:39:41
+
+## Branch
+
+**1/** **@_woookie_** ^1737165821925818879
+
+**@NOTimothyLottes** **@SheriefFYI**
+
+Yup, right now I async copy mesh and textures data, uniform/constat buffer are just a memcpy via a mapped ptr. I also have an "immediate" copy for any data upload into storage buffers per frame  (e.g. draw info). I gotta try rebar for the last ones though
+
+## Related
+
+- Spine: [[archive/threads/SheriefFYI/2023-12-19-always-always-always-use-an-async-copy-queue]]
@@ -0,0 +1,81 @@
+---
+title: "@NOTimothyLottes @SheriefFYI Wait, is this an issue with VK only or D3D12 too?"
+type: archive
+source: twitter
+source_url: "https://x.com/lectem/status/1737171397477106038"
+author: "Clément Grégoire"
+handle: lectem
+post_id: "1737171397477106038"
+date: 2023-12-19
+archived: 2026-08-26
+draft: false
+tags:
+  - archive
+  - twitter
+  - SheriefFYI
+description: "@NOTimothyLottes @SheriefFYI Wait, is this an issue with VK only or D3D12 too?"
+in_reply_to: ""
+parent_post_id: "1737161181524095352"
+---
+
+## Source
+
+- URL: https://x.com/lectem/status/1737171397477106038
+- Author: Clément Grégoire (@lectem)
+- Posted: 2023-12-19 18:01:50
+
+## Branch
+
+**1/** **@lectem** ^1737171397477106038
+
+**@NOTimothyLottes** **@SheriefFYI**
+
+Wait, is this an issue with VK only or D3D12 too?
+
+**2/** **@SheriefFYI** ^1737171797991190978
+
+**@lectem** **@NOTimothyLottes**
+
+both, it’s part of WDDM architecture.
+
+**3/** **@lectem** ^1737172605260443945
+
+**@SheriefFYI** **@NOTimothyLottes**
+
+Oh damn, I was under the impression that it was not really expensive as Unreal does use them quite a bit and I never saw issues with that...
+How big of an issue can it be?
+
+**4/** **@SheriefFYI** ^1737173083008487777
+
+**@lectem** **@NOTimothyLottes**
+
+depends on how long your wait is usually. if the fence is signaled early enough the wait can become a no-op.
+
+**5/** **@lectem** ^1737173826050392448
+
+**@SheriefFYI** **@NOTimothyLottes**
+
+In the scenario I've measured it was basicly copies or compute done at the beginning of frame so the gfx queue was definitely waiting and signal was after work.
+Didn't really measure but it didn't seem that big of an issue. Maybe because it was at beginning / end of cmd exec?
+
+**6/** **@lectem** ^1737174096947954046
+
+**@SheriefFYI** **@NOTimothyLottes**
+
+Or by early enough do you mean that if the wait is short (under millisec or something ) it's OK?
+
+**7/** **@SheriefFYI** ^1737174479967568185
+
+**@lectem** **@NOTimothyLottes**
+
+please measure and then we can talk.
+
+**8/** **@lectem** ^1737175062585696506
+
+**@SheriefFYI** **@NOTimothyLottes**
+
+I'll try to find a few captures during the week.
+
+## Related
+
+- Spine: [[archive/threads/SheriefFYI/2023-12-19-always-always-always-use-an-async-copy-queue]]
@@ -0,0 +1,37 @@
+---
+title: "@NOTimothyLottes @SheriefFYI I don't do async textures/buffers in my gpu abstraction yet, still not sure if I want to make it the default or optional at all."
+type: archive
+source: twitter
+source_url: "https://x.com/pATjako/status/1737173438417035568"
+author: "Peter Jakobs"
+handle: pATjako
+post_id: "1737173438417035568"
+date: 2023-12-19
+archived: 2026-08-26
+draft: false
+tags:
+  - archive
+  - twitter
+  - SheriefFYI
+description: "@NOTimothyLottes @SheriefFYI I don't do async textures/buffers in my gpu abstraction yet, still not sure if I want to make it the default or optional at all."
+in_reply_to: ""
+parent_post_id: "1737162074172637506"
+---
+
+## Source
+
+- URL: https://x.com/pATjako/status/1737173438417035568
+- Author: Peter Jakobs (@pATjako)
+- Posted: 2023-12-19 18:09:57
+
+## Branch
+
+**1/** **@pATjako** ^1737173438417035568
+
+**@NOTimothyLottes** **@SheriefFYI**
+
+I don't do async textures/buffers in my gpu abstraction yet, still not sure if I want to make it the default or optional at all.
+
+## Related
+
+- Spine: [[archive/threads/SheriefFYI/2023-12-19-always-always-always-use-an-async-copy-queue]]
@@ -0,0 +1,67 @@
+---
+title: "@NOTimothyLottes @SheriefFYI What’s the best way of doing that if rebar is not supported? Say on WebGL / DX11?"
+type: archive
+source: twitter
+source_url: "https://x.com/Meetem4/status/1737448166683713602"
+author: "Meetem"
+handle: Meetem4
+post_id: "1737448166683713602"
+date: 2023-12-20
+archived: 2026-08-26
+draft: false
+tags:
+  - archive
+  - twitter
+  - SheriefFYI
+description: "@NOTimothyLottes @SheriefFYI What’s the best way of doing that if rebar is not supported? Say on WebGL / DX11?"
+in_reply_to: ""
+parent_post_id: "1737162074172637506"
+---
+
+## Source
+
+- URL: https://x.com/Meetem4/status/1737448166683713602
+- Author: Meetem (@Meetem4)
+- Posted: 2023-12-20 12:21:37
+
+## Branch
+
+**1/** **@Meetem4** ^1737448166683713602
+
+**@NOTimothyLottes** **@SheriefFYI**
+
+What’s the best way of doing that if rebar is not supported? Say on WebGL / DX11?
+
+**2/** **@NOTimothyLottes** ^1737478162009801189
+
+**@Meetem4** **@SheriefFYI**
+
+For DX11, use Vulkan instead. On pre-large-bar, AMD still supported the 256 MiB device_local+host_visible, NV one has to fallback to slower bus crossing just host_visible buffer.
+
+**3/** **@NOTimothyLottes** ^1737479360867688545
+
+**@Meetem4** **@SheriefFYI**
+
+As for Web* - Root problem is more like the idea that if someone has a broken arm, that all arms should be broken too, so until the controlling parties can be convinced that it is ok to support both people with broken arms and non-broken arms, good luck
+
+**4/** **@Meetem4** ^1737515940055527782
+
+**@NOTimothyLottes** **@SheriefFYI**
+
+Thank you. Does large bar require support for ReBAR, so 30xx gpus?
+
+**5/** **@NOTimothyLottes** ^1737519841341321657
+
+**@Meetem4** **@SheriefFYI**
+
+Probably more complex, as in dependent on {motherboard chipset, BIOS/UEFI settings, GPU chipset that the IHVs greenlight enable which might be driver dependent}?
+
+**6/** **@Meetem4** ^1737528676198039713
+
+**@NOTimothyLottes** **@SheriefFYI**
+
+Thank you very much, I'm extemely appreciate that!
+
+## Related
+
+- Spine: [[archive/threads/SheriefFYI/2023-12-19-always-always-always-use-an-async-copy-queue]]
@@ -0,0 +1,46 @@
+---
+title: "Always always always use an async copy queue"
+type: archive
+source: twitter
+source_url: "https://x.com/SheriefFYI/status/1737126160402665958"
+author: "Sherief, FYI"
+handle: SheriefFYI
+post_id: "1737126160402665958"
+date: 2023-12-19
+archived: 2026-08-26
+draft: false
+tags:
+  - archive
+  - twitter
+  - SheriefFYI
+description: "Always always always use an async copy queue"
+in_reply_to: ""
+---
+
+## Source
+
+- URL: https://x.com/SheriefFYI/status/1737126160402665958
+- Author: Sherief, FYI (@SheriefFYI)
+- Posted: 2023-12-19 15:02:05
+
+## Thread
+
+**1/** **@SheriefFYI** ^1737126160402665958
+
+Always always always use an async copy queue
+
+**2/** **@NOTimothyLottes** ^1737161181524095352
+
+**@SheriefFYI**
+
+To be fair, that is only the case on PC if one can afford to delay waiting for transfer completion for a few frames (aka via cmd completion fence on VK), because queue to queue syncs (aka via VK semaphores) are perf-death-style CPU interrupt based ...
+
+Branches: [[archive/threads/SheriefFYI/2023-12-19-always-always-always-use-an-async-copy-queue/2023-12-19-lectem-wait-is-this-an-issue-with-vk-only-or-d3d12-too]]
+
+**3/** **@NOTimothyLottes** ^1737162074172637506
+
+**@SheriefFYI**
+
+... so for latency sensitive stuff, either stuff it via CPU writes to DEVICE_LOCAL+HOST_VISIBLE (or large BAR), or a limited kernel size copy shader if needing image stuffing
+
+Branches: [[archive/threads/SheriefFYI/2023-12-19-always-always-always-use-an-async-copy-queue/2023-12-19-SheriefFYI-queue-to-queue-syncs-via-cpu-need-to-go]], [[archive/threads/SheriefFYI/2023-12-19-always-always-always-use-an-async-copy-queue/2023-12-19-_woookie_-yup-right-now-i-async-copy-mesh-and-textures-data]], [[archive/threads/SheriefFYI/2023-12-19-always-always-always-use-an-async-copy-queue/2023-12-19-pATjako-i-dont-do-async-textures-buffers-in-my-gpu]], [[archive/threads/SheriefFYI/2023-12-19-always-always-always-use-an-async-copy-queue/2023-12-20-Meetem4-whats-the-best-way-of-doing-that-if-rebar-is-not]]
@@ -0,0 +1,12 @@
+---
+title: SheriefFYI
+type: note
+draft: false
+description: Archived threads by SheriefFYI.
+tags:
+  - archive
+  - twitter
+  - SheriefFYI
+---
+
+- [[archive/threads/SheriefFYI/2023-12-19-always-always-always-use-an-async-copy-queue]]
@@ -16,6 +16,7 @@ Example of the note shape (single spine, no branches): [[archive/threads/How to
 - [[archive/threads/AgileJebrim]]
 - [[archive/threads/BOENSAW]]
 - [[archive/threads/CharlieMQV]]
+- [[archive/threads/ChipsandCheese9]]
 - [[archive/threads/Dispatch_Graph]]
 - [[archive/threads/FFmpeg]]
 - [[archive/threads/Jonathan_Blow]]
@@ -23,6 +24,7 @@ Example of the note shape (single spine, no branches): [[archive/threads/How to
 - [[archive/threads/MyNameIsMJP]]
 - [[archive/threads/NOTimothyLottes]]
 - [[archive/threads/SebAaltonen]]
+- [[archive/threads/SheriefFYI]]
 - [[archive/threads/TheGingerBill]]
 - [[archive/threads/Triang3l]]
 - [[archive/threads/VPCOMPRESSB]]
@@ -7,7 +7,7 @@ author: "Jebrim"
 handle: AgileJebrim
 post_id: "1887249524051353899"
 date: 2025-02-05
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -7,7 +7,7 @@ author: "PBrrtrn"
 handle: BrrtrnP
 post_id: "1887236840194998670"
 date: 2025-02-05
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -57,6 +57,172 @@ I mean, maybe checking for actual gpu caps would make more sense, and just maybe

 or vulkan version, if you are lazy

+**5/** **@GustavSterbrant** ^1887259761294594524
+
+**@simplex_fx** **@BrrtrnP**
+
+It does that. But you’re too lazy to ask about it or to read the code. Instead you jump on the insult train immediately, while calling other people lazy. Not a great look.
+
+**6/** **@simplex_fx** ^1887260479594332463
+
+**@GustavSterbrant** **@BrrtrnP**
+
+it literally compares fucking strings too.
+
+**7/** **@GustavSterbrant** ^1887260618526478339
+
+**@simplex_fx** **@BrrtrnP**
+
+Yes and? Oh yeah I forgot you’re like an Amish person but for coding.
+
+**8/** **@simplex_fx** ^1887260914757607829
+
+**@GustavSterbrant** **@BrrtrnP**
+
+maybe it's a fucking retarded thing to figure out gpu caps based on vendor string?
+
+go back sucking off your pal
+
+**9/** **@GustavSterbrant** ^1887261419856343342
+
+**@simplex_fx** **@BrrtrnP**
+
+He’s not. I asked him if he checks capabilities and he said his engine does that first, this is just for the warning. But you jumped to conclusions thinking it’s so easy to write a Vulkan renderer from scratch.
+
+**10/** **@AgileJebrim** ^1887261780059009035
+
+**@GustavSterbrant** **@simplex_fx** **@BrrtrnP**
+
+It is easy to write a Vulkan renderer from scratch.
+
+**11/** **@GustavSterbrant** ^1887261914969063450
+
+**@AgileJebrim** **@simplex_fx** **@BrrtrnP**
+
+Not even remotely.
+
+**12/** **@GustavSterbrant** ^1887262512258924686
+
+**@AgileJebrim** **@simplex_fx** **@BrrtrnP**
+
+What you are talking about is an extremely narrow, compute only ray marcher. That’s not a renderer, that’s a laboration. A renderer is capable of handling content, materials, shaders etc. Be efficient with uploading memory, visibility, submitting frames and synchronization.
+
+**13/** **@NOTimothyLottes** ^1887284591754440819
+
+**@GustavSterbrant** **@AgileJebrim** **@simplex_fx** **@BrrtrnP**
+
+In my generation the now vintage machines didn't really have enough memory to handle the associated code complexity we have today, and well one can still author things in that way on modern machines, which makes it still approachable for one person :)
+
+**14/** **@simplex_fx** ^1887287012287226246
+
+**@NOTimothyLottes** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+Not that vintage, if you worked on WinCE 4-5-6 crap. Most didn’t even had fpus 😅
+
+**15/** **@NOTimothyLottes** ^1887288475671576874
+
+**@simplex_fx** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+My dad tried to force me off games by throwing away the Atari and C64, so I had cut my teeth mostly on ASM on a 8086 at first.
+
+**16/** **@simplex_fx** ^1887289402415579625
+
+**@NOTimothyLottes** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+We were too poor, never got hands on PC.
+
+First pc was then top of the line 166mhz pentium, not sure how my father managed to buy it, it was ultra expensive.
+Considering it was in mid 90s eastern europe, I suspect something semi-illegal at minimum😅
+
+**17/** **@simplex_fx** ^1887291056959443001
+
+**@NOTimothyLottes** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+All the c64 had was basic-assisted asm 😅😅
+
+I was furious when I saw shit like PLOT xy on an Atari or something 😅 We had to hit mem addresses with a hammer until it looked good (or froze the computer, probably destroying a day of work)
+
+**18/** **@simplex_fx** ^1887294440026616229
+
+**@NOTimothyLottes** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+Tbh perfect graphics api to me is still that.
+
+Unified memory, and mostly just dropping shit into random mem addresses instead of calling fuckton of api crap
+
+**19/** **@NOTimothyLottes** ^1887297722815783269
+
+**@simplex_fx** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+I know of at least one other game dev (besides myself) doing compute-only (no vertex/pixel) all GPU-side game logic. And minus init-time setup, the runtime is mostly exactly as you described, API-free.
+
+**20/** **@simplex_fx** ^1887302551155941632
+
+**@NOTimothyLottes** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+I still see too many problems related input and sound. Mostly latency related ones.
+
+**21/** **@NOTimothyLottes** ^1887304954836099411
+
+**@simplex_fx** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+It is substantially lower input latency. More so if the majority of render logic can be view-independent. You read input on GPU then only right before updating player+camera before view-dependent render. Below is how I do it ...
+
+![](https://pbs.twimg.com/media/GjEM3AMWUAAmFo0?format=png&name=orig)
+![](https://pbs.twimg.com/media/GjENxwBXkAA9grx?format=png&name=orig)
+
+**22/** **@simplex_fx** ^1887306831485190594
+
+**@NOTimothyLottes** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+I mean gameplay vs input not cam vs input
+
+**23/** **@NOTimothyLottes** ^1887308171569459242
+
+**@simplex_fx** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+Gameplay logic is done on the GPU too.
+No CPU involvement.
+So yeah, it is about as real low latency as possible.
+
+**24/** **@simplex_fx** ^1887308790103437322
+
+**@NOTimothyLottes** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+Ok but input devices don’t grow on the gpu.
+
+Not to mention sound card and nic
+
+**25/** **@NOTimothyLottes** ^1887310996022112697
+
+**@simplex_fx** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+Incoming UDP packets (I prefer fixed size) can be fed to the GPU using a similar method. Modern GPUs from AMD/NV have quite capable scalar cores which run in parallel with the vector ones, and GPU scalar cores might even be lower power ...
+
+**26/** **@simplex_fx** ^1887311684047397364
+
+**@NOTimothyLottes** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+I mean, is there some direct nic-gpu stuff?
+
+I recently maybe saw something in an AI related context, not sure of that’s a thing, or how that works.
+
+Kinda like nvme->gpu stuff?
+
+My problem is more like the other way, aka fetching data from gpu
+
+**27/** **@NOTimothyLottes** ^1887316887219925286
+
+**@simplex_fx** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+It would be nice if one could setup shared memory mappings between devices [storage/sound/NIC/etc] and then completely bypass the CPU, but today's OS/driver/HW engineering fail continues.
+
+**28/** **@NOTimothyLottes** ^1887319803834077606
+
+**@simplex_fx** **@GustavSterbrant** **@AgileJebrim** **@BrrtrnP**
+
+At a minimum IHVs could provide a way to shader store to the DRAM used by the HDMI out audio ring buffer, and also have HW write the [in/out] position to L2|DRAM so it is shader read accessible. Then also you'd need pinned allocations and quality of service on GPU runtime
+
 ## Related

 - Spine: [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in]]
@@ -7,7 +7,7 @@ author: ":3"
 handle: Colonthreee
 post_id: "1887227641905721796"
 date: 2025-02-05
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -7,7 +7,7 @@ author: "DeltaWave0x 🇮🇹"
 handle: DeltaWave0x
 post_id: "1887270110626521366"
 date: 2025-02-05
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -7,7 +7,7 @@ author: "Dev"
 handle: Kapendev
 post_id: "1887242908551569426"
 date: 2025-02-05
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -7,7 +7,7 @@ author: "NOTimothyLottes"
 handle: NOTimothyLottes
 post_id: "1887240771360739590"
 date: 2025-02-05
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -7,7 +7,7 @@ author: "WireWhiz"
 handle: TheWireWhiz
 post_id: "1887287287790080010"
 date: 2025-02-05
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -7,7 +7,7 @@ author: "Can"
 handle: ecoezen
 post_id: "1887230363342090332"
 date: 2025-02-05
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -1,63 +0,0 @@
---
-title: "@simplex_fx Why are you so mean and negative though?"
-type: archive
-source: twitter
-source_url: "https://x.com/kulkalkul/status/1887267961594876245"
-author: "Bora"
-handle: kulkalkul
-post_id: "1887267961594876245"
-date: 2025-02-05
-archived: 2026-08-25
-draft: false
-tags:
-  - archive
-  - twitter
-  - simplex_fx
-description: "@simplex_fx Why are you so mean and negative though?"
-in_reply_to: ""
-parent_post_id: "1887224326350635382"
---
-
-## Source
-
- URL: https://x.com/kulkalkul/status/1887267961594876245
- Author: Bora (@kulkalkul)
- Posted: 2025-02-05 22:32:00
-
-## Branch
-
-**1/** **@kulkalkul** ^1887267961594876245
-
-**@simplex_fx**
-
-Why are you so mean and negative though?
-
-**2/** **@simplex_fx** ^1887268929195671807
-
-**@kulkalkul**
-
-eastern european privilege :D
-
-how long are you following me? :D
-
-I'm not negative, I'm telling people to don't be fucktarts like people I repost.
-
-**3/** **@kulkalkul** ^1887270503515107409
-
-**@simplex_fx**
-
-Year, yeah I know this is a general theme of you.
-
-I don't get what's so wrong about OP's post that it is deserving insults. He just shared that he won't be supporting almost 10 year old GPUs in his personal project.
-
-I am pretty sure you can make the same point without insults.
-
-**4/** **@simplex_fx** ^1887271068760441102
-
-**@kulkalkul**
-
-seeing the actual detection code was the trigger I think :D
-
-## Related
-
- Spine: [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in]]
@@ -1,49 +0,0 @@
---
-title: "@simplex_fx I'M USING MY FUCKING 1080TI UNTIL IT FUCKING EXPLODES IT HAS BEEN MY BEST FRIEND FOR SO LONG, FUCKING LEGENDARY! GODS CARD!"
-type: archive
-source: twitter
-source_url: "https://x.com/p1xelHer0/status/1887262571369275777"
-author: "p1xelHer0"
-handle: p1xelHer0
-post_id: "1887262571369275777"
-date: 2025-02-05
-archived: 2026-08-25
-draft: false
-tags:
-  - archive
-  - twitter
-  - simplex_fx
-description: "@simplex_fx I'M USING MY FUCKING 1080TI UNTIL IT FUCKING EXPLODES IT HAS BEEN MY BEST FRIEND FOR SO LONG, FUCKING LEGENDARY! GODS CARD!"
-in_reply_to: ""
-parent_post_id: "1887224326350635382"
---
-
-## Source
-
- URL: https://x.com/p1xelHer0/status/1887262571369275777
- Author: p1xelHer0 (@p1xelHer0)
- Posted: 2025-02-05 22:10:35
-
-## Branch
-
-**1/** **@p1xelHer0** ^1887262571369275777
-
-**@simplex_fx**
-
-I'M USING MY FUCKING 1080TI UNTIL IT FUCKING EXPLODES IT HAS BEEN MY BEST FRIEND FOR SO LONG, FUCKING LEGENDARY! GODS CARD!
-
-**2/** **@simplex_fx** ^1887262735551111219
-
-**@p1xelHer0**
-
-rich asshole, we only have 1060 :D
-
-**3/** **@p1xelHer0** ^1887447576699425162
-
-**@simplex_fx**
-
-Gotta spend money to make money! Or something...
-
-## Related
-
- Spine: [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in]]
@@ -1,37 +0,0 @@
---
-title: "@simplex_fx In most cases it doesn't make sense to do so."
-type: archive
-source: twitter
-source_url: "https://x.com/Bernatmago/status/1887407288786280594"
-author: "Bernat | ベルナットゥ"
-handle: Bernatmago
-post_id: "1887407288786280594"
-date: 2025-02-06
-archived: 2026-08-25
-draft: false
-tags:
-  - archive
-  - twitter
-  - simplex_fx
-description: "@simplex_fx In most cases it doesn't make sense to do so."
-in_reply_to: ""
-parent_post_id: "1887224326350635382"
---
-
-## Source
-
- URL: https://x.com/Bernatmago/status/1887407288786280594
- Author: Bernat | ベルナットゥ (@Bernatmago)
- Posted: 2025-02-06 07:45:38
-
-## Branch
-
-**1/** **@Bernatmago** ^1887407288786280594
-
-**@simplex_fx**
-
-In most cases it doesn't make sense to do so.
-
-## Related
-
- Spine: [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in]]
@@ -1,37 +0,0 @@
---
-title: "@simplex_fx 2016 tech."
-type: archive
-source: twitter
-source_url: "https://x.com/Charcharo/status/1887389094306615522"
-author: "Charcharo"
-handle: Charcharo
-post_id: "1887389094306615522"
-date: 2025-02-06
-archived: 2026-08-25
-draft: false
-tags:
-  - archive
-  - twitter
-  - simplex_fx
-description: "@simplex_fx 2016 tech."
-in_reply_to: ""
-parent_post_id: "1887224326350635382"
---
-
-## Source
-
- URL: https://x.com/Charcharo/status/1887389094306615522
- Author: Charcharo (@Charcharo)
- Posted: 2025-02-06 06:33:20
-
-## Branch
-
-**1/** **@Charcharo** ^1887389094306615522
-
-**@simplex_fx**
-
-2016 tech. Obsolete. Don't support them.
-
-## Related
-
- Spine: [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in]]
@@ -7,7 +7,7 @@ author: "Sebastian Aaltonen"
 handle: SebAaltonen
 post_id: "1887404173710258607"
 date: 2025-02-06
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -7,7 +7,7 @@ author: "Karabo 🇿🇦"
 handle: karabo_x1
 post_id: "1887576466973507924"
 date: 2025-02-06
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -1,43 +0,0 @@
---
-title: "@simplex_fx GPU poor no longer."
-type: archive
-source: twitter
-source_url: "https://x.com/piotroxp/status/1887567193883484542"
-author: "piotroxp"
-handle: piotroxp
-post_id: "1887567193883484542"
-date: 2025-02-06
-archived: 2026-08-25
-draft: false
-tags:
-  - archive
-  - twitter
-  - simplex_fx
-description: "@simplex_fx GPU poor no longer."
-in_reply_to: ""
-parent_post_id: "1887224326350635382"
---
-
-## Source
-
- URL: https://x.com/piotroxp/status/1887567193883484542
- Author: piotroxp (@piotroxp)
- Posted: 2025-02-06 18:21:02
-
-## Branch
-
-**1/** **@piotroxp** ^1887567193883484542
-
-**@simplex_fx**
-
-GPU poor no longer. Now I don’t play games because I’m running models
-
-**2/** **@simplex_fx** ^1887574040342372502
-
-**@piotroxp**
-
-I feel you, I just ordered 4x sxm2 v100 gpus for my "We have AI at home" setup :D
-
-## Related
-
- Spine: [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in]]
@@ -7,7 +7,7 @@ author: "stamisme"
 handle: stamisme
 post_id: "1887407379395801172"
 date: 2025-02-06
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -7,7 +7,7 @@ author: "Tonči Jukić"
 handle: toncijukic
 post_id: "1887409756274344181"
 date: 2025-02-06
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -1,37 +0,0 @@
---
-title: "@simplex_fx i agree that godot is ran by retarda"
-type: archive
-source: twitter
-source_url: "https://x.com/chri4__/status/1887800757828587940"
-author: "chx"
-handle: chri4__
-post_id: "1887800757828587940"
-date: 2025-02-07
-archived: 2026-08-25
-draft: false
-tags:
-  - archive
-  - twitter
-  - simplex_fx
-description: "@simplex_fx i agree that godot is ran by retarda"
-in_reply_to: ""
-parent_post_id: "1887224326350635382"
---
-
-## Source
-
- URL: https://x.com/chri4__/status/1887800757828587940
- Author: chx (@chri4__)
- Posted: 2025-02-07 09:49:08
-
-## Branch
-
-**1/** **@chri4__** ^1887800757828587940
-
-**@simplex_fx**
-
-i agree that godot is ran by retarda
-
-## Related
-
- Spine: [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in]]
@@ -7,7 +7,7 @@ author: "Simplex"
 handle: simplex_fx
 post_id: "1887224326350635382"
 date: 2025-02-05
-archived: 2026-08-25
+archived: 2026-08-26
 draft: false
 tags:
  - archive
@@ -36,7 +36,7 @@ Open source engines like Godot and this are ran by retarded people.

 ROLL YOUR OWN FUCKING ENGINE!

-Branches: [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-Colonthreee-i-would-rather-spend-time-on-the-gameplay-than]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-ecoezen-lol-by-only-looking-at-the-code-snapshot-its]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-BrrtrnP-context-its-a-mostly-experimental-engine-meant-to]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-NOTimothyLottes-rolling-my-own-i-dont-support-10xx-because-they]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-Kapendev-your-choice-of-words-always-makes-me-laugh-lol]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-AgileJebrim-i-think-its-wise-to-support-older-hardware-tbh]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-p1xelHer0-im-using-my-fucking-1080ti-until-it-fucking]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-kulkalkul-why-are-you-so-mean-and-negative-though]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-DeltaWave0x-if-you-really-are-a-aaa-gamedev-for-10-years-and]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-TheWireWhiz-min-spec-gpu-in-vr-is-a-standalone-headset]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-06-Charcharo-2016-tech-obsolete-dont-support-them]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-06-SebAaltonen-9-year-old-hardware-is-not-ancient-we-had-to]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-06-Bernatmago-in-most-cases-it-doesnt-make-sense-to-do-so]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-06-stamisme-tbh-coding-your-own-engine-from-scratch-is-not]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-06-toncijukic-rolling-your-own-engine-to-production-quality-is]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-06-piotroxp-gpu-poor-no-longer-now-i-dont-play-games-because]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-06-karabo_x1-i-have-thought-about-this-i-am-not-a-game]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-07-chri4__-i-agree-that-godot-is-ran-by-retarda]]
+Branches: [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-Colonthreee-i-would-rather-spend-time-on-the-gameplay-than]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-ecoezen-lol-by-only-looking-at-the-code-snapshot-its]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-BrrtrnP-context-its-a-mostly-experimental-engine-meant-to]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-NOTimothyLottes-rolling-my-own-i-dont-support-10xx-because-they]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-Kapendev-your-choice-of-words-always-makes-me-laugh-lol]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-AgileJebrim-i-think-its-wise-to-support-older-hardware-tbh]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-DeltaWave0x-if-you-really-are-a-aaa-gamedev-for-10-years-and]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-05-TheWireWhiz-min-spec-gpu-in-vr-is-a-standalone-headset]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-06-SebAaltonen-9-year-old-hardware-is-not-ancient-we-had-to]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-06-stamisme-tbh-coding-your-own-engine-from-scratch-is-not]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-06-toncijukic-rolling-your-own-engine-to-production-quality-is]], [[archive/threads/simplex_fx/2025-02-05-i-mean-1050-and-1060-are-just-top20-gpus-in/2025-02-06-karabo_x1-i-have-thought-about-this-i-am-not-a-game]]

 **2/** **@simplex_fx** ^1887224657239285919