Video yükleniyor...

Video Yüklenemedi

Bu video yüklenirken bir sorun oluştu. Bu geçici bir ağ sorunundan kaynaklanıyor olabilir veya video kullanılamıyor olabilir.

Ana Sayfaya Dön

Stress testing Metropolis 1998! - Population: 53,500 - CPU FPS: 80 - GPU FPS: 10-80 - RAM: 3,100 MB So: At the moment, the game is GPU bound. I know why and there's a lot of room for improvement. It will involve developing or customizing an existing GPU framework.... show more

Yesbox - Metropolis 1998

9,816 subscribers

84,873 görüntüleme • 1 yıl önce •via X (Twitter)

Oyun Bilim & Teknoloji

Anya Rossi• Live Now

Private livecam show

11 Yorum

John Heritage profil fotoğrafı

John Heritage1 yıl önce

This looks fantastic! Is it a .. "if you liked the original Simcity you'll like this"?

Yesbox - Metropolis 1998 profil fotoğrafı

Yesbox - Metropolis 19981 yıl önce

It's a mix of SimCity, Cities Skylines, and Rollercoaster Tycoon. So yes!

NetMind.AI profil fotoğrafı

NetMind.AI2 yıl önce

Get access to a wide range of GPUs like H100, A100, 4090, 3090 and save over 90% at NetMind Power. Rent Now!

Ben Throop profil fotoğrafı

Ben Throop1 yıl önce

Ooo curious what the bottleneck is. Is it purely the people? Or just the size of city that comes along with that population

Yesbox - Metropolis 1998 profil fotoğrafı

Yesbox - Metropolis 19981 yıl önce

Its the number of draw calls, inability to do a depth pass, and prob. other things I'm not aware of due to my inexperience with all things GPU. <not sure if GPU terms are correct> Every object in the game is separated into parts (so each part can be custom colored). I have to draw each part individually since I am not able to expand the buffer stride. e.g. GPUs allow 16 additional vec4 fields that I could store all this information in. So essentially every tile/object in the game is being drawn 5 times >_< Im forced to use OpenGL 1.0 or 2.0 (not sure which) functions CPU side.

Bewildered Beast profil fotoğrafı

Bewildered Beast1 yıl önce

This is very much near the top of my wishlist

Carson profil fotoğrafı

Carson1 yıl önce

holy moly this is incredible

Thomas Fossgard profil fotoğrafı

Thomas Fossgard1 yıl önce

That’s awesome..

Zsky profil fotoğrafı

Zsky1 yıl önce

Keep improving, I wanna play that piece of gold 😝

Ritik profil fotoğrafı

Ritik1 yıl önce

very cool!

Marcelo Correia profil fotoğrafı

Marcelo Correia1 yıl önce

OMG! Looks like a Transport Tycoon Deluxe with city construction!

Benzer Videolar

GTA V's splash screen at 60 FPS in a PS5 emulator?! SharpEMU can now boot GTA V to its splash screen at a smooth 60 FPS, marking another exciting milestone for PS5 emulation. Current stats: • 60.1 FPS • 16.7 ms frame time • ~731 MB RAM • ~393 MB buffer • ~3% CPU usage • 1+ minute runtime There's still plenty of work ahead, but the progress is undeniable. Credit: Discord/marcelmedia

GTA V's splash screen at 60 FPS in a PS5 emulator?! SharpEMU can now boot GTA V to its splash screen at a smooth 60 FPS, marking another exciting milestone for PS5 emulation. Current stats: • 60.1 FPS • 16.7 ms frame time • ~731 MB RAM • ~393 MB buffer • ~3% CPU usage • 1+ minute runtime There's still plenty of work ahead, but the progress is undeniable. Credit: Discord/marcelmedia

Techjunkie Aman

676,803 görüntüleme • 1 ay önce

PowerInfer - a high-speed inference engine for deploying LLMs locally. Just came across this super interesting project on speeding up inference. It's not MoE but it's a simple approach that exploits the high locality in LLM inference to design a GPU-CPU hybrid inference engine. Hot-activated neurons are preloaded onto the GPU for fast access, while cold-activated neurons (the majority) are computed on the CPU. This approach significantly reduces GPU memory demands and CPU-GPU data transfer. It achieves an average token generation rate of 13.20 tokens/s, with a peak of 29.08 tokens/s, across various LLMs on a single NVIDIA RTX 4090 GPU. It's on only 18% lower than that achieved by a top-tier server-grade A100 GPU. It also significantly outperforms llama.cpp by up to 11.69x while retaining model accuracy. There is a lot more innovation around inference that's coming fast. Really encouraged by the study on sparse computation to enhance the computational efficiency of LLMs. It's now possible to use PowerInfer with Llama 2 and Faclon 40B. Mistral-7B support is coming soon!

PowerInfer - a high-speed inference engine for deploying LLMs locally. Just came across this super interesting project on speeding up inference. It's not MoE but it's a simple approach that exploits the high locality in LLM inference to design a GPU-CPU hybrid inference engine. Hot-activated neurons are preloaded onto the GPU for fast access, while cold-activated neurons (the majority) are computed on the CPU. This approach significantly reduces GPU memory demands and CPU-GPU data transfer. It achieves an average token generation rate of 13.20 tokens/s, with a peak of 29.08 tokens/s, across various LLMs on a single NVIDIA RTX 4090 GPU. It's on only 18% lower than that achieved by a top-tier server-grade A100 GPU. It also significantly outperforms llama.cpp by up to 11.69x while retaining model accuracy. There is a lot more innovation around inference that's coming fast. Really encouraged by the study on sparse computation to enhance the computational efficiency of LLMs. It's now possible to use PowerInfer with Llama 2 and Faclon 40B. Mistral-7B support is coming soon!

elvis

261,622 görüntüleme • 2 yıl önce

Monad Testnet Visualized (Pt 3) I think Claude and I have fixed the freeze/pause issue. Works a lot smoother and most of the jitter is now just due to my GPU/CPU begging for mercy trying to handle the data visualization

Monad Testnet Visualized (Pt 3) I think Claude and I have fixed the freeze/pause issue. Works a lot smoother and most of the jitter is now just due to my GPU/CPU begging for mercy trying to handle the data visualization

Ray J

18,673 görüntüleme • 1 yıl önce

The fact that it starts up *immediately* when tabbing into the game feels like useful info. Not a thing that ramps up or down. I capped the FPS so stop asking about it. GPU is at well under 80% load. I'm only drawing 50% of the power my PSU can handle.

The fact that it starts up immediately when tabbing into the game feels like useful info. Not a thing that ramps up or down. I capped the FPS so stop asking about it. GPU is at well under 80% load. I'm only drawing 50% of the power my PSU can handle.

Theo - t3.gg

52,664 görüntüleme • 1 yıl önce

GPU by hand ✍️ I drew 42 frames to show how a GPU speeds up an array operation of 8 elements in parallel over 4 threads in 2 clock cycles. Below is an overview: CPU • It has one core. • Its global memory has 120 locations (0-119). • To use the GPU, it needs to copy data from the global memory to the GPU. • After GPU is done, it will copy the results back. GPU • It has four cores to run four threads (0-3). • It has a register file of 28 locations (0-27) • This register file has four banks (0-3). • All threads share the same register file. • But they must read/write using the four banks. • Each bank allows 2 reads (Read 0, Read 1) and 1 write in a single clock cycle.

GPU by hand ✍️ I drew 42 frames to show how a GPU speeds up an array operation of 8 elements in parallel over 4 threads in 2 clock cycles. Below is an overview: CPU • It has one core. • Its global memory has 120 locations (0-119). • To use the GPU, it needs to copy data from the global memory to the GPU. • After GPU is done, it will copy the results back. GPU • It has four cores to run four threads (0-3). • It has a register file of 28 locations (0-27) • This register file has four banks (0-3). • All threads share the same register file. • But they must read/write using the four banks. • Each bank allows 2 reads (Read 0, Read 1) and 1 write in a single clock cycle.

Tom Yeh

101,788 görüntüleme • 1 yıl önce

Crazy how Action Quake 2 and The Specialists still inspire a lot of FPS games. I grew up with those. I still don't know what the subgenre is called, there's so little FPS games in that category, I doubt there is even a name for them I need to make a list

Crazy how Action Quake 2 and The Specialists still inspire a lot of FPS games. I grew up with those. I still don't know what the subgenre is called, there's so little FPS games in that category, I doubt there is even a name for them I need to make a list

ZooL

32,611 görüntüleme • 3 ay önce

My take on Searle's Chinese Room: It's just milking a misleading visual metaphor. If you *actually* visualize what's going on in that room, the guy is using the book to process information at lightspeed, much like the operation of an H100 GPU or a brain.

My take on Searle's Chinese Room: It's just milking a misleading visual metaphor. If you actually visualize what's going on in that room, the guy is using the book to process information at lightspeed, much like the operation of an H100 GPU or a brain.

Liron Shapira

14,986 görüntüleme • 1 yıl önce

Palette Pmonday Palette: Memory's Journey This kind of palette is an xbit classic, so of course I had to include it! Also I know I record in this room a lot. It's a good room okay? Wishlist Silent Santicado on Steam! Link in bio #metroidvania #indiegame #pixelart #ドット絵

Palette Pmonday Palette: Memory's Journey This kind of palette is an xbit classic, so of course I had to include it! Also I know I record in this room a lot. It's a good room okay? Wishlist Silent Santicado on Steam! Link in bio #metroidvania #indiegame #pixelart #ドット絵

Nix Pix 🐰 Wishlist Silent Santicado Now!

18,045 görüntüleme • 2 yıl önce

Dylan Patel of SemiAnalysis says a worse GPU with better storage and memory now beats the best chip without them, so buying the newest GPU alone no longer wins inference. So, an AMD GPU with more memory can outperform Nvidia in some cases. "So what we have is we have over $80 million of compute, GPUs from Nvidia, AMD, TPUs from Google, Trainium from Amazon, and we run this benchmark constantly on the newest inference engine, newest drivers, newest PyTorch version, whatever it is." "Every day it runs on an automated CI, and we run it on all the latest Chinese models, from GLM, Zhipu, Moonshot, Kimi, Alibaba, all these models we run." "Initially, when we were benchmarking the difference between these chips and different engines, different schemes for parallelism, we were just running it fixed context length." "But now with Agent X, we've analyzed over $5 million worth of Claude Code traces. This is real production traffic that people have donated to us as well as internally generated. Now we know what the actual agent workload looks like." "And then as we implement that and run those benchmarks, it turns out yes, the chip you're using is very important, but now even more important is how are you handling this memory offload?" "And so while an Nvidia GPU is faster than an AMD GPU in most cases, because AMD GPUs have more memory, they actually end up outperforming in some cases." "Or you can have a worse GPU, but a much better storage solution, and now you can outperform what the best GPU can do without those solutions. So just buying the newest and latest GPU alone doesn't get you the best inference economics." "Actually, you need to layer in all these other innovations including storage and memory." [ Who's the top player on your chart? ] "That really is a difficult multivariable problem. And generally that means you need to have, yes, you need to have the best GPU, a GB300, but you also need to have the best storage solutions. And so I won't spoil who's the best right here, but I will say that storage solutions matter a lot and memory solutions matter a lot, as does your front-end networking. That matters a lot."

Dylan Patel of SemiAnalysis says a worse GPU with better storage and memory now beats the best chip without them, so buying the newest GPU alone no longer wins inference. So, an AMD GPU with more memory can outperform Nvidia in some cases. "So what we have is we have over $80 million of compute, GPUs from Nvidia, AMD, TPUs from Google, Trainium from Amazon, and we run this benchmark constantly on the newest inference engine, newest drivers, newest PyTorch version, whatever it is." "Every day it runs on an automated CI, and we run it on all the latest Chinese models, from GLM, Zhipu, Moonshot, Kimi, Alibaba, all these models we run." "Initially, when we were benchmarking the difference between these chips and different engines, different schemes for parallelism, we were just running it fixed context length." "But now with Agent X, we've analyzed over $5 million worth of Claude Code traces. This is real production traffic that people have donated to us as well as internally generated. Now we know what the actual agent workload looks like." "And then as we implement that and run those benchmarks, it turns out yes, the chip you're using is very important, but now even more important is how are you handling this memory offload?" "And so while an Nvidia GPU is faster than an AMD GPU in most cases, because AMD GPUs have more memory, they actually end up outperforming in some cases." "Or you can have a worse GPU, but a much better storage solution, and now you can outperform what the best GPU can do without those solutions. So just buying the newest and latest GPU alone doesn't get you the best inference economics." "Actually, you need to layer in all these other innovations including storage and memory." [ Who's the top player on your chart? ] "That really is a difficult multivariable problem. And generally that means you need to have, yes, you need to have the best GPU, a GB300, but you also need to have the best storage solutions. And so I won't spoil who's the best right here, but I will say that storage solutions matter a lot and memory solutions matter a lot, as does your front-end networking. That matters a lot."

Fireside Alpha

175,586 görüntüleme • 1 ay önce

Every year I use ZERO "Run plays" for the CPU in #NBA2K26 and a lot of people ask why. People think it causes the CPU to never run plays, but as you can see if you can balance the individual player tendencies correctly the result is a smooth authentic CPU offensive experience.

Every year I use ZERO "Run plays" for the CPU in #NBA2K26 and a lot of people ask why. People think it causes the CPU to never run plays, but as you can see if you can balance the individual player tendencies correctly the result is a smooth authentic CPU offensive experience.

Shady 🧊

184,692 görüntüleme • 5 ay önce

Gavin Baker (Gavin Baker) says the disaggregation of inference can extend GPU useful lives from 3-4 years to 10-15. That may single-handedly save private credit and reduce the financing rates for GPUs, which will drive demand and help finance the build-out. "The disaggregation of prefill and inference is going to be amazing for the useful lives of GPU and may single-handedly save private credit. Private credit is in pain from these SaaS loans. But there's a lot of private credit in GPUs too. They were underwriting that to 3-4. The disaggregation of inference means that these GPUs are going to have 10 or 15-year lives. The AI skeptics are like, "Oh, these companies are all cooking their books. The useful life of a GPU is only a year or two. The useful life of a CPU is only four years because the rapid technological change." No. What rapid technological change has done with the disaggregation of prefill and inference is you can put a Cerebras system or Groq LPUs effectively in front of a Hopper or even an Ampere, use that Hopper and Ampere for prefill, and extend the useful life of that GPU until it melts. This is going to be really good for the whole private credit industry. It's gonna help finance the AI build-out. Because if you can start to finance GPUs at 5% or 6% instead of – I think CoreWeave's lowest financing was low sevens – that actually mathematically changes the cost to finance this build-out."

Gavin Baker (Gavin Baker) says the disaggregation of inference can extend GPU useful lives from 3-4 years to 10-15. That may single-handedly save private credit and reduce the financing rates for GPUs, which will drive demand and help finance the build-out. "The disaggregation of prefill and inference is going to be amazing for the useful lives of GPU and may single-handedly save private credit. Private credit is in pain from these SaaS loans. But there's a lot of private credit in GPUs too. They were underwriting that to 3-4. The disaggregation of inference means that these GPUs are going to have 10 or 15-year lives. The AI skeptics are like, "Oh, these companies are all cooking their books. The useful life of a GPU is only a year or two. The useful life of a CPU is only four years because the rapid technological change." No. What rapid technological change has done with the disaggregation of prefill and inference is you can put a Cerebras system or Groq LPUs effectively in front of a Hopper or even an Ampere, use that Hopper and Ampere for prefill, and extend the useful life of that GPU until it melts. This is going to be really good for the whole private credit industry. It's gonna help finance the AI build-out. Because if you can start to finance GPUs at 5% or 6% instead of – I think CoreWeave's lowest financing was low sevens – that actually mathematically changes the cost to finance this build-out."

Invest Like the Best

211,336 görüntüleme • 3 ay önce

Dylan Patel on the importance of memory and storage Two key quotes: "An $NVDA GPU is faster than an $AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads." “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly" Full Quote: “We have over $80 million of compute: GPUs from $NVDA and $AMD, TPUs from Google, and Trainium from Amazon. We constantly run this benchmark using the newest inference engines, drivers, PyTorch versions, and other software. It runs every day through automated CI across the latest Chinese models from GLM, Zhipu, Moonshot, Kimi, Alibaba, and others. Initially, when we were benchmarking the differences between these chips, inference engines, and parallelism schemes, we used fixed context lengths. But with Agent X, we have now analyzed more than $5 million worth of Claude Code traces. This is real production traffic that users have donated to us, combined with internally generated data, so we now understand what an actual agent workload looks like. When we implement those workloads and run the benchmarks, it turns out that the chip you are using is very important, but how you handle memory offload can be even more important. An Nvidia GPU is faster than an AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads. Similarly, you can use a less powerful GPU with a much better storage solution and outperform the best GPU when it lacks those solutions. Simply buying the newest GPU does not necessarily give you the best inference economics. You need to layer in other innovations, including storage and memory.” Interviewer: “Who is the top player on your chart? Can you tell us?” Dylan Patel: “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly.”

Dylan Patel on the importance of memory and storage Two key quotes: "An $NVDA GPU is faster than an $AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads." “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly" Full Quote: “We have over $80 million of compute: GPUs from $NVDA and $AMD, TPUs from Google, and Trainium from Amazon. We constantly run this benchmark using the newest inference engines, drivers, PyTorch versions, and other software. It runs every day through automated CI across the latest Chinese models from GLM, Zhipu, Moonshot, Kimi, Alibaba, and others. Initially, when we were benchmarking the differences between these chips, inference engines, and parallelism schemes, we used fixed context lengths. But with Agent X, we have now analyzed more than $5 million worth of Claude Code traces. This is real production traffic that users have donated to us, combined with internally generated data, so we now understand what an actual agent workload looks like. When we implement those workloads and run the benchmarks, it turns out that the chip you are using is very important, but how you handle memory offload can be even more important. An Nvidia GPU is faster than an AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads. Similarly, you can use a less powerful GPU with a much better storage solution and outperform the best GPU when it lacks those solutions. Simply buying the newest GPU does not necessarily give you the best inference economics. You need to layer in other innovations, including storage and memory.” Interviewer: “Who is the top player on your chart? Can you tell us?” Dylan Patel: “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly.”

Daniel Romero

38,220 görüntüleme • 1 ay önce

"As of right now, it's a little emotional in there and there's a lot of try hard." Tony Vitello blames himself for the lack of offensive production over two games, citing a need for the locker room to loosen up.

"As of right now, it's a little emotional in there and there's a lot of try hard." Tony Vitello blames himself for the lack of offensive production over two games, citing a need for the locker room to loosen up.

KNBR

1,188,648 görüntüleme • 5 ay önce

UC Berkeley just open-sourced FreeToken. (2–4x faster local LLM inference than Ollama) the results are wild: - Qwen3.6-35B on an 8GB GPU at 39.3 tokens/s - DeepSeek-V4-Flash 284B on a 32GB GPU at 22 tokens/s - GLM-5.2 753B on a 96GB GPU at 14.9 tokens/s a 35B model at 16-bit precision needs about 70GB just for its weights. even at 4 bits it is close to 18GB, and FreeToken serves it on an 8GB GPU. let me explain how: all three models mentioned above are Mixture-of-Experts, and that is what FreeToken takes advantage of. each layer holds hundreds of separate experts plus a small router that picks a few of them per token. Qwen3.6-35B activates roughly 3B of its 35B parameters per token. DeepSeek-V4-Flash picks 6 of 256 experts per layer, so 13B of its 284B run at a time. so compute was never the bottleneck. the weights a single step touches fit comfortably on a consumer GPU. every expert the router might pick still has to exist somewhere. they sit in system RAM, and the GPU keeps a cache of the ones the model has been using recently. so everything comes down to what happens when the router picks an expert that is not on the GPU. there are two ways to serve that miss: 1. copy it over PCIe and run it on the GPU 2. run it on the CPU, where it already lives both read from the same system memory, so they compete for one pool of bandwidth instead of adding to each other. existing engines pick one option and freeze it when the model loads. but routing changes on every token, so a fixed choice misses most of what the model asks for. FreeToken measures both bandwidths on your machine and splits each step's misses between the two paths in proportion. the GPU and CPU results then merge exactly, with no approximation. two machines with the same GPU can end up wanting opposite strategies, which I did not expect. a 5090 in a gaming desktop should push nearly everything over PCIe, while an 8GB laptop is better off computing most misses on the CPU. none of that is readable off a spec sheet, so the engine profiles it once per machine. the second half of the design is about agents. coding agents constantly rewrite their own history, and every edit normally forces thousands of tokens back through prefill. FreeToken saves its checkpoints at the exact boundaries agent frameworks cut on, so it only reprocesses the new part. its slowest first token stays under 44 seconds, while llama.cpp peaks at 232 and KTransformers at 946. it serves the OpenAI and Anthropic APIs under Apache 2.0, so Claude Code and Codex can point at it directly. releasing weights publicly decides who can download a model, not who can afford to run one. frontier open models keep shipping, and running them still assumes a rented cluster. meanwhile there are over a hundred million consumer machines with discrete GPUs sitting mostly idle. closing that gap was never a hardware problem, and work like this is what turns open weights into something you can actually use. paper: repo: almost every idea in this post, from why memory bandwidth decides the outcome to why moving weights costs more than computing on them, comes straight out of how a GPU is built. I wrote a detailed primer on that. the article is quoted below.

UC Berkeley just open-sourced FreeToken. (2–4x faster local LLM inference than Ollama) the results are wild: - Qwen3.6-35B on an 8GB GPU at 39.3 tokens/s - DeepSeek-V4-Flash 284B on a 32GB GPU at 22 tokens/s - GLM-5.2 753B on a 96GB GPU at 14.9 tokens/s a 35B model at 16-bit precision needs about 70GB just for its weights. even at 4 bits it is close to 18GB, and FreeToken serves it on an 8GB GPU. let me explain how: all three models mentioned above are Mixture-of-Experts, and that is what FreeToken takes advantage of. each layer holds hundreds of separate experts plus a small router that picks a few of them per token. Qwen3.6-35B activates roughly 3B of its 35B parameters per token. DeepSeek-V4-Flash picks 6 of 256 experts per layer, so 13B of its 284B run at a time. so compute was never the bottleneck. the weights a single step touches fit comfortably on a consumer GPU. every expert the router might pick still has to exist somewhere. they sit in system RAM, and the GPU keeps a cache of the ones the model has been using recently. so everything comes down to what happens when the router picks an expert that is not on the GPU. there are two ways to serve that miss: 1. copy it over PCIe and run it on the GPU 2. run it on the CPU, where it already lives both read from the same system memory, so they compete for one pool of bandwidth instead of adding to each other. existing engines pick one option and freeze it when the model loads. but routing changes on every token, so a fixed choice misses most of what the model asks for. FreeToken measures both bandwidths on your machine and splits each step's misses between the two paths in proportion. the GPU and CPU results then merge exactly, with no approximation. two machines with the same GPU can end up wanting opposite strategies, which I did not expect. a 5090 in a gaming desktop should push nearly everything over PCIe, while an 8GB laptop is better off computing most misses on the CPU. none of that is readable off a spec sheet, so the engine profiles it once per machine. the second half of the design is about agents. coding agents constantly rewrite their own history, and every edit normally forces thousands of tokens back through prefill. FreeToken saves its checkpoints at the exact boundaries agent frameworks cut on, so it only reprocesses the new part. its slowest first token stays under 44 seconds, while llama.cpp peaks at 232 and KTransformers at 946. it serves the OpenAI and Anthropic APIs under Apache 2.0, so Claude Code and Codex can point at it directly. releasing weights publicly decides who can download a model, not who can afford to run one. frontier open models keep shipping, and running them still assumes a rented cluster. meanwhile there are over a hundred million consumer machines with discrete GPUs sitting mostly idle. closing that gap was never a hardware problem, and work like this is what turns open weights into something you can actually use. paper: repo: almost every idea in this post, from why memory bandwidth decides the outcome to why moving weights costs more than computing on them, comes straight out of how a GPU is built. I wrote a detailed primer on that. the article is quoted below.

Akshay 🚀

318,151 görüntüleme • 4 gün önce

Every time I think the ESP32-S3 has hit its limit, someone finds another gear. This open-source Jet renderer is now pushing the S3 even harder. Real-time screen-space reflections, water ripple distortion, lens flare and 8 ships on screen, all running on a tiny MCU. Performance dropped from around 60 FPS to 45–50 FPS, but that's still wild for an ESP32-S3. And the developer thinks there is more performance left to squeeze out. No GPU. No desktop CPU. Just an ESP32-S3 doing real-time 3D graphics.

Every time I think the ESP32-S3 has hit its limit, someone finds another gear. This open-source Jet renderer is now pushing the S3 even harder. Real-time screen-space reflections, water ripple distortion, lens flare and 8 ships on screen, all running on a tiny MCU. Performance dropped from around 60 FPS to 45–50 FPS, but that's still wild for an ESP32-S3. And the developer thinks there is more performance left to squeeze out. No GPU. No desktop CPU. Just an ESP32-S3 doing real-time 3D graphics.

Praveen Kumar Verma

15,924 görüntüleme • 6 gün önce

First I optimized CPU, then memory, now I'm onto the renderers (Metal first). Scrolling Neovim now sustains ~175fps easily (vsync off of course). More room for improvement, but this is already best-in-class (I can't find a terminal at the moment that does better on macOS). The road to get here was long. To quote Thorsten Ball I've been "moving the chess pieces into place over the past few months." I'll write about it in a devlog, eventually. And, people will inevitably say "high fps doesn't matter for a terminal." And yeah, maybe, but what's the fun in that. My answer to "why" is "because I can!" (Disclaimer: I use the Japanese bible as a sort of stress test for certain IO tests. I'm not religious and there isn't meant to be any sort of hidden message in this.)

First I optimized CPU, then memory, now I'm onto the renderers (Metal first). Scrolling Neovim now sustains ~175fps easily (vsync off of course). More room for improvement, but this is already best-in-class (I can't find a terminal at the moment that does better on macOS). The road to get here was long. To quote Thorsten Ball I've been "moving the chess pieces into place over the past few months." I'll write about it in a devlog, eventually. And, people will inevitably say "high fps doesn't matter for a terminal." And yeah, maybe, but what's the fun in that. My answer to "why" is "because I can!" (Disclaimer: I use the Japanese bible as a sort of stress test for certain IO tests. I'm not religious and there isn't meant to be any sort of hidden message in this.)

Mitchell Hashimoto

374,114 görüntüleme • 2 yıl önce

HBM creator Kim Jung-ho reveals his lab's tHBM flipped stack publicly for the first time saying Samsung's zHBM with memory stacked on a hot GPU simply melts and is likely not the right fundamental direction "So the idea is, let's just put the HBM on top of the GPU. Then you never go outside the building, you only ride the elevator, and the speed jumps." "That's the structure Samsung is now proposing as zHBM. The z is, from the x, y, z axes we learned back in school, the vertical direction, so it means stacked in the z direction." "But unfortunately, I don't see that as the fundamental direction, because if you stack HBM on top of the GPU, the GPU gets hot and the memory melts." "So I suspect it's for very simple computation, the kind of use people call an LPU or an NPU. They showed it in passing as a fun technology, and the really important technology doesn't get shown at places like that." "So my lab researches tHBM. I believe today is the first time we're saying this publicly. We flip it, stack the HBM, and put the GPU on top of that."

HBM creator Kim Jung-ho reveals his lab's tHBM flipped stack publicly for the first time saying Samsung's zHBM with memory stacked on a hot GPU simply melts and is likely not the right fundamental direction "So the idea is, let's just put the HBM on top of the GPU. Then you never go outside the building, you only ride the elevator, and the speed jumps." "That's the structure Samsung is now proposing as zHBM. The z is, from the x, y, z axes we learned back in school, the vertical direction, so it means stacked in the z direction." "But unfortunately, I don't see that as the fundamental direction, because if you stack HBM on top of the GPU, the GPU gets hot and the memory melts." "So I suspect it's for very simple computation, the kind of use people call an LPU or an NPU. They showed it in passing as a fun technology, and the really important technology doesn't get shown at places like that." "So my lab researches tHBM. I believe today is the first time we're saying this publicly. We flip it, stack the HBM, and put the GPU on top of that."

Fireside Alpha

110,837 görüntüleme • 8 gün önce

I've made this animation a few months ago. The city and the cars assets come from Epic Ganes Market Place. All the rest is my stuff. It runs in realtime with an rtx 2070...Lots of room for improvement but i had a lot of fun making it..

I've made this animation a few months ago. The city and the cars assets come from Epic Ganes Market Place. All the rest is my stuff. It runs in realtime with an rtx 2070...Lots of room for improvement but i had a lot of fun making it..

Alvizello

33,408 görüntüleme • 1 yıl önce