Stable Diffusion - Mac vs RTX4090 vs RTX3060 vs Google Colab - how they perform.

Render Realm
29 Aug 202309:25

TLDRIn this video, the creator compares the performance of Stable Diffusion across different systems, including a MacBook Pro M1 Max, a mid-range PC with an RTX 3060, a high-end PC with an RTX 4090, and Google Colab. The benchmarks include text-to-image and image-to-image tasks at various resolutions. The RTX 4090 outperforms all other systems, especially with high-resolution tasks, while the RTX 3060 offers a good balance between cost and performance. Google Colab, even with its subscription plans, lags behind the dedicated GPUs. The M1 Max MacBook Pro shows significant performance issues with Stable Diffusion, suggesting it's not yet optimized for Apple's silicon. The video concludes that for those requiring high computing power and willing to spend, the RTX 4090 is the top choice, while the RTX 3060 is recommended for a mid-range system. For budget-conscious users or those new to the platform, Google Colab is suggested as an alternative.

Takeaways

  • 💻 The comparison is between a MacBook Pro M1 Max, a mid-range PC with an RTX 3060, a high-end PC with an RTX 4090, and Google Colab for running Stable Diffusion.
  • 📈 The RTX 4090 outperforms the RTX 3060 and the Mac in benchmarks, taking only 2.1 seconds compared to 3.6 seconds for the RTX 3060.
  • 🚀 The RTX 4090 is the clear winner in performance, nearly four times better than the RTX 3060 and significantly faster than the Mac and Google Colab.
  • 🔍 Google Colab's performance was expectedly lower, using an older Tesla T4 GPU.
  • 🍎 The Mac, despite having a powerful M1 Max chip, showed performance issues with Stable Diffusion and wasn't optimized for it.
  • 📦 The high-end PC with RTX 4090 also has 24 GB of VRAM and 64 GB of RAM, which contributes to its superior performance.
  • 📉 The performance gap widens at higher resolutions, with the RTX 4090 remaining the top performer.
  • 💔 The Mac struggled with high-resolution tasks and threw an error when using the automatic 1111 version of Stable Diffusion.
  • 💰 The RTX 4090, while offering the best performance, also has the highest power consumption and cost.
  • 💭 For those on a tighter budget, the RTX 3060 or similar mid-range GPUs are recommended, or considering Google Colab which offers a free basic version.
  • ⚙️ If you already own an Apple Silicon Mac, it can be used for Stable Diffusion, but it's not recommended to buy one solely for this purpose due to its current performance limitations.

Q & A

  • What is the main topic of the video?

    -The main topic of the video is a comparison of how stable diffusion performs on different systems, including a Mac, mid-range PC with an RTX 3060, a high-end PC with an RTX 4090, and Google Colab.

  • What are the specifications of the Mac used in the comparison?

    -The Mac used is a MacBook Pro M1 Max with 10 CPU cores, 32 GPU cores, and 32GB of memory.

  • Which two versions of stable diffusion were tested on the Mac?

    -The two versions of stable diffusion tested on the Mac are the usual automatic 1111 and the Vlad stable diffusion, which is optimized for the Apple silicon GPU.

  • What are the approximate costs of the mid-range and high-end PCs used in the comparison?

    -The mid-range PC with an AMD Ryzen 5 and Nvidia RTX 3060 costs about a thousand Euro or less, while the high-end PC with a Ryzen 9 and RTX 4090 costs over 3,000 Euro.

  • What GPU does Google Colab provide in its free version?

    -In the free version, Google Colab provides an Nvidia Tesla T4 GPU.

  • How many benchmarks were conducted and what was the method to determine the results?

    -Nine benchmarks were conducted, each with five iterations. The mean value of the last four iterations was used as the result for each benchmark.

  • What was the surprise result regarding the Mac's performance with stable diffusion?

    -The surprise result was that the Mac's performance with stable diffusion was significantly lower than expected, indicating that stable diffusion is not yet optimized for the Mac.

  • Which system had the best performance in the benchmarks?

    -The RTX 4090 system had the best performance in the benchmarks, performing nearly four times better than the RTX 3060 and several times better than the Mac and Google Colab.

  • What was the conclusion for users with different budgets and needs regarding the use of stable diffusion?

    -For users who need great computing power and can afford it, the RTX 4090 is the best choice. For those with a tighter budget, the RTX 3060 or a similar mid-range system is recommended. For Apple silicon Mac users, the machine can be used for stable diffusion but is not optimal. For users with a very low budget, Google Colab is suggested as an alternative.

  • What was the issue encountered when using the automatic 1111 version of stable diffusion on the Mac?

    -The Mac threw an error when using the automatic 1111 version of stable diffusion, suggesting that it may not be compatible or there might be a configuration issue.

  • What is the advantage of using Google Colab for stable diffusion tasks?

    -Google Colab offers the advantage of not requiring a powerful personal GPU, as it provides access to Google's servers with powerful GPUs. It also has a free basic version and offers subscription plans with more powerful GPU options.

Outlines

00:00

🖥️ Mac vs PC Performance on Stable Diffusion

The speaker begins by introducing the topic of comparing the performance of Stable Diffusion across different systems, specifically highlighting their personal journey from using a Mac to a mid-range PC and then to a high-end PC. They discuss their initial use of Stable Diffusion on a MacBook Pro M1 Max and the subsequent acquisition of a PC with an AMD Ryzen 5 and Nvidia RTX 3060 for projects requiring Unreal Engine, which is not well-supported on Macs. The need for more computing power led to the purchase of another PC with an RTX 4090. The speaker also mentions using Google Colab for demanding tasks like training Stable Diffusion models. They conducted 9 benchmarks across different platforms to compare performance, using various Stable Diffusion versions and models, and rendering tasks. The results were surprising, particularly the underperformance of the Mac despite its powerful M1 Max chip, indicating a lack of optimization for Macs.

05:01

📊 Benchmark Results and System Performance Conclusions

The speaker presents the results of their benchmarks, noting the RTX 4090's superior performance, especially at higher resolutions, with the RTX 3060 and Google Colab also performing well at lower resolutions. However, the Mac struggled with high-resolution tasks. The RTX 4090 was the clear winner in terms of performance, outperforming the RTX 3060 by nearly four times, the Mac by five to six times, and Google Colab by three and a half times. Despite its high cost and power consumption, the RTX 4090 was deemed the best choice for those requiring significant computing power. For those on a tighter budget, the RTX 3060 or similar mid-range systems were recommended. The speaker advised against purchasing a Mac solely for Stable Diffusion due to its current performance issues with the software. Google Colab was suggested as a cost-effective alternative for those with low budgets or who wish to avoid high expenses on hardware.

Mindmap

Keywords

💡Stable Diffusion

Stable Diffusion is an AI model used for generating images from textual descriptions. It is a significant topic in the video as the host compares its performance across different hardware platforms. The host mentions using Stable Diffusion on various systems, including a Mac, a mid-range PC with an RTX 3060, a high-end PC with an RTX 4090, and Google Colab, to evaluate how each setup handles the AI model's demands.

💡MacBook Pro M1 Max

The MacBook Pro M1 Max is a high-end Apple laptop featuring Apple's own M1 Max chip, which includes 10 CPU and 32 GPU cores along with 32GB of memory. In the video, it is used as one of the systems to test the performance of Stable Diffusion, highlighting its capabilities and limitations when running the AI model.

💡RTX 3060

The RTX 3060 is a mid-range graphics card from Nvidia, equipped with 12GB of VRAM and often used in gaming and graphic-intensive tasks. The video compares its performance with other systems when running Stable Diffusion, noting its efficiency and value for money.

💡RTX 4090

The RTX 4090 is a high-end graphics card from Nvidia, featuring 24GB of VRAM and a powerful CPU, making it suitable for demanding tasks like running Stable Diffusion. The video presents it as the top-performing system in the benchmarks, capable of handling large projects with ease.

💡Google Colab

Google Colab is a cloud-based platform that provides access to computing resources, including GPUs for running data science and machine learning tasks. The video discusses its use for running Stable Diffusion, noting that it offers an alternative for those without access to powerful hardware, although it may have limitations compared to local setups.

💡Benchmarks

Benchmarks are tests used to measure the performance of hardware or software. In the context of the video, the host conducts 9 benchmarks with five iterations each to assess how well each system runs Stable Diffusion. The benchmarks include text-to-image and image-to-image tasks at various resolutions, providing a comprehensive view of the systems' capabilities.

💡Reliberate Model

The Reliberate Model is one of the versions of Stable Diffusion used in the benchmarks. It is mentioned for its performance characteristics and how it utilizes the Apple silicon GPU more effectively than the other version tested, the automatic 1111.

💡SDXL 1.0 Base Model

The SDXL 1.0 Base Model is another version of Stable Diffusion tested in the video, used for image-to-image tasks at a resolution of 1024x1024 pixels. The host notes that this model does not support control nets, which is an important consideration for its performance evaluation.

💡Control Nets

Control Nets are additional neural networks used to guide the image generation process in Stable Diffusion, allowing for more precise control over the output. The video mentions their use with the Reliberate Model but notes that they are not supported by the SDXL model.

💡High-Res Fix

High-Res Fix refers to a modification or setting that allows for higher resolution image generation. The video discusses its impact on performance, noting that it significantly affects the Mac's ability to run Stable Diffusion smoothly.

💡Animation Rendering

Animation Rendering is the process of generating a sequence of images to create an animation. The video includes a benchmark that involves rendering a standard animation with 120 frames using both the Reliberate and SDXL models at different resolutions to test the systems' capabilities in handling complex tasks.

💡Performance Optimization

Performance Optimization refers to the process of enhancing the efficiency and speed of a system or software. The video highlights that Stable Diffusion is not yet optimized for the Mac, which is evident in the performance discrepancies when compared to the other systems tested.

Highlights

A comparison of Stable Diffusion performance on different systems: Mac, RTX 3060, RTX 4090, and Google Colab.

The user's experience with Stable Diffusion on a MacBook Pro M1 Max, highlighting its powerful capabilities but noting optimization issues.

Introduction of a mid-range PC with an AMD Ryzen 5 and Nvidia RTX 3060 for Unreal Engine projects and Stable Diffusion.

The acquisition of a high-end PC with a Ryzen 9 and RTX 4090 for more powerful computing needs.

Utilization of Google Colab for demanding tasks like Dream Booth trainings for custom Stable Diffusion models.

Benchmarking process involving 9 tests with five iterations each to assess system performance.

Surprising results from the benchmarks, indicating the RTX 4090's superior performance in comparison to other systems.

The RTX 4090 completed the first benchmark in just 2.1 seconds, significantly faster than the RTX 3060 at 3.6 seconds.

Google Colab's use of an older Tesla T4 GPU resulted in expectedly slower performance.

The Mac's performance was a significant letdown, indicating a lack of optimization for Stable Diffusion on Apple Silicon.

At 768x768 resolution, the performance gap between systems widened, with the RTX 4090 being the clear winner.

High-res fix tests showed the RTX 4090 maintaining performance while other systems struggled.

The RTX 3060 and Google Colab performed well at lower resolutions but faced challenges at higher resolutions.

The Mac encountered errors when using the automatic 1111 version of Stable Diffusion, suggesting compatibility issues.

For image-to-image tests with control nets, the RTX 4090 showed no issues even at high resolutions due to its 24GB of VRAM.

The Mac struggled with high-resolution images, particularly when using the automatic 1111, and threw errors.

In animation rendering tests, the RTX 4090 was unbeatable at 512x512 pixels, while other systems performed similarly.

The final conclusion: the RTX 4090 is the top choice for those needing great computing power and willing to spend, while the RTX 3060 is a solid mid-range option.

Google Colab is recommended for those with a low budget or who wish to try Stable Diffusion without a significant financial investment.

The Mac is not recommended for purchasing solely for Stable Diffusion purposes due to its current performance issues.