
# Maximizing Value with NVIDIA A40 & RTX A6000

For many people and organizations, the cost of high-end hardware is prohibitive when it comes to tasks like fine-tuning large language models (LLMs) and other AI workloads. It may not make sense to purchase a super-powered machine like the NVIDIA A100 or H100 when a more affordable option exists and can get the job done just about the same.

For instance, the NVIDIA A40 and RTX A6000 GPUs are incredibly attractive options for the more budget-conscious user – at least when compared to such expensive higher-end machines! Not only do they balance performance and cost, but they're also far more readily available than the A100 and H100 and can scale AI projects quickly.

### NVIDIA A40 & RTX A6000: Similarities and Differences

The A40 and A6000 are both professional-grade GPUs, well suited for high-performance computing. The A40 is intended for server environments and data centers while the A6000 is designed for desktop workstations, but otherwise they're quite similar with just a few minor differences.

Both GPUs are Ampere architecture-based with a PCIe Gen 4.0 interface and have 48GB of GDDR6 RAM including error-correction code (ECC). However, the A40 offers 696 GB/s of peak memory bandwidth while the A6000 provides a touch more at 768 GB/s of bandwidth – as well as a slightly higher clock speed.

The two GPUs are tailor-made to handle demanding, large-scale AI workloads, with each featuring 10,752 CUDA cores (shading units), 84 second-gen RT cores, and 336 third-gen Tensor cores. Both include hardware support for a fine-grained structured sparsity feature, which can be used to accelerate inference and other deep learning workloads.

The A40 is passively cooled and employs bidirectional airflow, allowing air to move in either direction through the heatsink. This makes it better suited for use in servers. The A6000, on the other hand, has active cooling. Both GPUs use up quite a bit of power with a maximum power consumption of 300 watts.

The A40 has three display outputs, although they are disabled by default since the GPU is configured out of the box to support virtual graphics and compute workloads in virtualized environments. This makes it highly suitable for cloud-based applications and services, as it can easily deliver high-performance graphics and compute capabilities to remote users. The A6000 has four display ports that are enabled by default but are not active when using virtual GPU software.

Unlike the NVIDIA A100 and H100, the A40 and A6000 do not support Multi-Instance GPU (MIG), so it's not possible to run separate and fault-isolated workloads in parallel on the same physical GPU. However, their memory can be expanded by integrating a second GPU using NVLink technology, which allows the two to pool resources and operate as one unit with reduced latency and a combined memory of 96GB – plenty for most uses!

### Benefits and Uses

Because the A40 and A6000 GPUs are so well suited to cloud environments and are more readily available than higher-end hardware, they allow organizations to scale their AI initiatives in a cost-effective and operationally efficient manner. And now, with the introduction of 10x GPU servers, the A40 and A6000 can be deployed in extremely powerful configurations to undertake AI projects that are even more ambitious and require significant computational resources.

According to some benchmarks, the A6000 can run about 10% faster than the A40 overall, due to its higher clock speed and memory bandwidth. But this advantage must be evaluated against the features of the A40, such as its secure and measured boot with hardware root of trust; NEBS Level 3 compliance (making it ideal for use in a wide array of network and telecom applications where stability and reliability are critical); and superior suitability to server environments.

Based on specs and performance, some examples of appropriate uses for the A40 and A6000 GPUs include:

- **AI and Deep Learning Workflows** – Training sophisticated neural networks, fine-tuning LLMs, running AI inference at scale, and deploying AI applications across various sectors like healthcare and finance.
- **Scientific Research and Engineering Simulations** – Running detailed simulations, modeling, data analysis, and computer-aided engineering (CAE) tasks in areas like climate study, bioinformatics, and the automotive, aerospace, and manufacturing industries.
- **Advanced Visualization** – Performing tasks where rapid rendering and visual fidelity are paramount, such as professional content creation and graphic design, virtual production, broadcast-grade streaming, real-time visual effects, and animation for film and game studios.

### The Bottom Line

Ultimately, the NVIDIA A40 and RTX A6000 are an excellent choice for organizations and professionals who may not want to pay a supremely high price for the NVIDIA A100 or H100 – or who are happy to make the trade-off between faster processing times and lower cost plus availability – while still enjoying the ability to tackle the largest workloads in AI, visual computing, and data science.

All of that being said, the A40 and A6000 do still cost a pretty penny! Fortunately, purchasing your own hardware isn't a requirement to get started with these powerful machines. Here at Vast.ai, with low-cost cloud GPU rental through our platform, you can access a wide range of compute power across our network of hosts around the world, anytime, anywhere.

We offer the best prices for GPU rental, with on-demand pricing as low as $0.12/hr for the A40 and $0.50/hr for the RTX A6000 – and interruptible instances available via spot auction-based pricing for even more savings. We're proud of our mission to help democratize AI and ensure that its benefits are available to all!

---

**[The table below compares various features and specs between the NVIDIA A40 and RTX A6000, with differences highlighted.]**

| Feature                                              | A40                                                        | RTX A6000                                                  |
| ---------------------------------------------------- | ---------------------------------------------------------- | ---------------------------------------------------------- |
| Architecture                                         | Ampere                                                     | Ampere                                                     |
| Memory Size                                          | 48 GB                                                      | 48 GB                                                      |
| Memory Type                                          | GDDR6                                                      | GDDR6                                                      |
| Error-Correcting Code (ECC)                          | Yes                                                        | Yes                                                        |
| CUDA Cores                                           | 10,752                                                     | 10,752                                                     |
| 3rd-gen Tensor Cores                                 | 336                                                        | 336                                                        |
| 2nd-gen RT Cores                                     | 84                                                         | 84                                                         |
| Render Output Units (ROPs)                           | 112                                                        | 112                                                        |
| FP16 (Half) Performance                              | 37.42 TFLOPS                                               | **38.71 TFLOPS**                                           |
| FP32 (Float) Performance                             | 37.42 TFLOPS                                               | **38.71 TFLOPS**                                           |
| FP64 (Double) Performance                            | 584.6 GFLOPS                                               | **604.8 GFLOPS**                                           |
| Pixel Rate                                           | 194.9 GPixel/s                                             | **201.6 GPixel/s**                                         |
| Texture Rate                                         | 584.6 GTexel/s                                             | **604.8 GTexel/s**                                         |
| Memory Interface                                     | 384-bit                                                    | 384-bit                                                    |
| Memory Bandwidth                                     | 696 GB/s                                                   | **768 GB/s**                                               |
| Clock Speed                                          | 1305 MHz                                                   | **1410 MHz**                                               |
| Boost Speed                                          | 1740 MHz                                                   | **1800 MHz**                                               |
| Memory Clock                                         | 1812 MHz (14.5 Gbps effective)                             | **2000 MHz (16 Gbps effective)**                           |
| Slot Width                                           | Dual-slot                                                  | Dual-slot                                                  |
| Thermal Solution                                     | Passive                                                    | Active                                                     |
| Power Consumption (Total Board Power)                | 300 W                                                      | 300 W                                                      |
| Suggested Power Supply                               | 700 W                                                      | 700 W                                                      |
| System Interface                                     | PCIe 4.0 x16                                               | PCIe 4.0 x16                                               |
| Display Ports                                        | 3x DisplayPort 1.4                                         | **4x DisplayPort 1.4**                                     |
| vGPU Support                                         | Yes (default)                                              | Yes                                                        |
| NVLink                                               | Yes                                                        | Yes                                                        |
| NEBS Ready                                           | **Yes (Level 3)**                                          | No                                                         |
| Secure and Measured Boot with Hardware Root of Trust | **Yes (optional)**                                         | No                                                         |
| Graphics APIs                                        | DirectX 12.07, Shader Model 5.17, OpenGL 4.68, Vulkan 1.18 | DirectX 12.07, Shader Model 5.17, OpenGL 4.68, Vulkan 1.18 |
| Compute APIs                                         | CUDA, DirectCompute, OpenCL, **OpenACC**                   | CUDA, DirectCompute, OpenCL                                |
| Release Date                                         | Oct. 5th, 2020                                             | Oct. 5th, 2020                                             |


Maximizing Value with NVIDIA A40 & RTX A6000...

Maximizing Value with NVIDIA A40 & RTX A6000 | June 2024


Welcome to another update from the team at Vast.ai! We remain dedicated to continually refining our GPU rental platform, ensuring it not only meets but exceeds the expectations of our valued users.

To that end, we're currently still beta testing AMD support. If you'd like to test an AMD machine, simply select a ROCm template from the recommended templates. If you see no machines, select 'unverified' and you'll find some AMD servers online. Enjoy!

Back to regularly scheduled maintenance… As with any technology-driven service, occasional bugs come with the territory. Addressing these promptly remains a top priority. Here are some of the latest bug fixes we’ve implemented since our last update:

- Fixed our cluster rental form resizing issue on mobile
- Safeguards are now in place to prevent users from using instance copy with no SSH keys
- Expired instances now show as expired and not scheduling
- Fixed issue with instance price detail showing wrong pre-discount price
- Reserved instances now do not stop when primary balance goes negative as long as instance balance is positive
- Template editor now filters out CUDA tags that are known not to work

### Cloud GPU Rental Platform Updates

Here at Vast.ai, we also continuously strive to enhance the quality and functionality of our platform. This past month, we've introduced several key improvements that boost the flexibility and performance of our GPU rental marketplace. Below are some highlights of the latest changes:

- [Machine Hosting Setup guide](https://cloud.vast.ai/host/setup/) has been updated to include AMD support
- Added AMD ROCm templates for base ROCm and ROCm Pytorch
- AMD GPUs that are listed on Vast have been added to search filters
- New automated port testing for verified machines to check port mappings/settings
- Updated Docs with [Windows Powershell SSH](https://vast.ai/docs/use-cases/windows-ssh-scp-guide) directions
- Added sort options on host machine page along with sort by hostname
- Added display if instance is static/dynamic

Check out the full change log for our April 2024 product updates at the bottom of this page. We're confident these updates will enhance your experience with Vast!

If you have any questions about our services or need assistance maximizing your use of our platform, don't hesitate to get in touch. Whether you're a longtime user or just starting out, we're here to support all your GPU rental and hosting needs. Feel free to chat with us directly through the chatbox at the bottom right, email us at [support@vast.ai](mailto:support@vast.ai), or join our vibrant Discord community for immediate advice and resources.

### Change Log

- We are still beta testing AMD support. If you want to test an AMD machine, select a ROCm template from the recommended templates. If you see no machines, include unverified. You will see some AMD servers online.
- Updated host setup guide to include AMD support - [https://cloud.vast.ai/host/setup/](https://cloud.vast.ai/host/setup/)
- Added AMD ROCm templates for base ROCm and ROCm Pytorch
- Added a pop-up after clicking the instance-to-instance copy button for clarity
- Bugfix: resized SSH key fields on instance card
- Bugfix: fixed our cluster rental form resizing issue on mobile
- New automated port testing for verified machines to check port mappings/settings
- Negative credits are being shown to users now
- Cloud dialog can be properly scrolled
- AMD GPUs that are listed on Vast have been added to search filters
- Updated Docs with Windows Powershell SSH directions (easier than Putty) - [https://vast.ai/docs/use-cases/windows-ssh-scp-guide](https://vast.ai/docs/use-cases/windows-ssh-scp-guide)
- Bugfix: safeguards in place to prevent users from using instance copy with no SSH keys
- Bugfix: clarified cloud copy 'directory not found' and related errors
- Bugfix: fixed bug with ./vast.py create subaccount --explain
- Schedule maint command now supports human-readable time
- Added sort options on host machine page along with sort by hostname
- Added notification re: why credit card can't be removed for negative balance accounts
- Moved maintenance info icon under report icon on instance card
- Added amortized bandwidth cost estimate in dlp/$ and auto search metrics based on duration and disk size
- Improved Rclone cloud copy with better options. API & CLI now support most of the Rclone options
- Added display if instance is static/dynamic
- Bugfix: expired instances show as expired and not scheduling
- Bugfix: fixed scroll jump after closing and canceling cloud copy dialogue
- Bugfix: fixed issue with instance price detail showing wrong pre-discount price
- Bugfix: fixed issue with listing results becoming out of sync/delayed
- Bugfix: reserved instances now do not stop when primary balance goes negative as long as instance balance is positive
- Bugfix: fixed wrapping issue in template UI
- Bugfix: template editor now filters out CUDA tags that are known not to work


Ongoing product updates are an important part of what we offer at Vast.ai...

Vast Product Update | April 2024


NVIDIA's highly anticipated next-generation GH200 Grace Hopper Superchip platform is finally available. Perfectly suited for the era of accelerated computing and generative AI, it's essentially the world's most powerful GPU paired with the most versatile CPU. 

  

With its scalable design, the NVIDIA GH200 can handle the most complex generative AI workloads, from large language models (LLMs) and recommender systems to vector databases, graph neural networks (GNNs), and beyond. 

  

"To meet surging demand for generative AI, data centers require accelerated computing platforms with specialized needs," [said](https://nvidianews.nvidia.com/news/gh200-grace-hopper-superchip-with-hbm3e-memory) NVIDIA founder and CEO Jensen Huang. This is precisely the gap that the GH200 aims to fill. 

  

Huang added, "[It] delivers this with exceptional memory technology and bandwidth to improve throughput, the ability to connect GPUs to aggregate performance without compromise, and a server design that can be easily deployed across the entire data center."

  

### Groundbreaking Memory and Bandwidth

  

The GH200's architecture redefines high-performance computing (HPC). It integrates the cutting-edge Hopper GPU and the versatile Grace CPU into a single superchip – bridged by the lightning-fast, memory-coherent NVIDIA NVLink Chip-2-Chip (C2C) interconnect. 

  

The NVLink-C2C interconnect is the heart of the GH200 Grace Hopper Superchip. The power of its coherent memory is pretty incredible; the interconnect delivers 900GB/s of bidirectional CPU-GPU bandwidth. That's seven times the performance compared to the PCIe Gen5 connections found in conventional accelerated systems! Plus, interconnect power consumption is slashed by more than 5X.

  

With NVLink-C2C, applications can oversubscribe the GPU's memory, harnessing the high-bandwidth memory of the Grace CPU directly. The GH200 offers up to 480GB of LPDDR5X CPU memory, meaning that – combined with 96GB HBM3 or 144GB HBM3e – the Hopper GPU can directly access up to 624GB of fast memory. 




![](/uploads/wxoZpDKUOFRqBvNJkCLco6gy3-gBofvjPwirVSYuHDdHVBbFVT-JDjSOQHIYWTSV6VTnVj4U5kFJ9lhfUeHalOA5N0QRdDAV0TxByVawrhTyJID0yActvuTghOYzziRHm9xDCI0xIwY3ubuD5nRSEas.png)


(Image credit: NVIDIA)

  

Key features of the NVIDIA Grace CPU include:

  

- 2X the performance per watt of conventional x86-64 platforms
    
- 72 Neoverse V2 Armv9 cores with up to 480GB of server-class LPDDR5X memory with error-correction code (ECC)
    
- Up to 53% more bandwidth at one-eighth the power per GB per second, compared to an eight-channel DDR5 design
    

  

Then there's the H100 Tensor Core GPU, based on the new Hopper GPU architecture, which showcases several innovative highlights:

  

- Unprecedentedly fast matrix computations by the new fourth-generation Tensor Cores, enabling a wider range of AI and HPC tasks
    
- Up to 9X faster AI training and up to 30X faster AI inference with its new Transformer Engine compared to the prior-generation NVIDIA A100
    
- Maximized quality of service for smaller workloads thanks to secure Multi-Instance GPU (MIG) partitioning the GPU into isolated, right-size instances
    

  

In a nutshell, if you need raw power, you've got it with the GH200. Unfortunately, it's so new that there hasn't really been much comprehensive benchmarking. But let's take a look anyway. 

### Benchmarks: Pitting the GH200 Against its Main Rivals

  

On the Linux benchmark site Phoronix, author Michael Larabel [published](https://www.phoronix.com/review/nvidia-gh200-gptshop-benchmark) some initial HPC benchmarks run on a GH200 workstation sourced from GPTshop.ai. Testing was done on the Grace CPU performance only.

  

The GH200 system in question was tested with 72 cores, a Quanta S74G motherboard, 480GB of RAM, and 960GB + 1920GB SAMSUNG SSD drives. (A complete description of specs and environment can be found on the Phoronix site at the above link.) Again, this benchmarking was merely preliminary, as it focused on CPU performance, and no power consumption numbers are available – but there were some interesting results nonetheless. 

  

According to the standard [HPCG](https://www.hpcg-benchmark.org/) memory bandwidth benchmark, the GH200 Grace CPU came in at a solid 41.7 GFLOPS. 

  

  ![](/uploads/OatdrZqcQK_Bp41PV_DBcUFPeSRnvYSJpF4O1uc1Q6D_dLyr4ESSPbeS5HzrYaK92ybZSAVi3ma0rKQ0FmlpXcgDBrhGdY06dP5C980mThA-hIlAexWMgn0yiXt3GXod1eAklyI8p1HdjCGyRpoEw-U.png)

Results for NVIDIA GH200 running HPCG benchmark. (Image credit: Phoronix)

5vFBmCRUcbfgb9PR7NpFoXpSeRSFqr47lr9V1YtOG-k2Aus1jPY1T217XD_EnBKKZKnn8MTFUs-Jx2lMYVR_8Ek4Q9eQcM-kml5l0C2B4wfzwdytunAxtEcGZNfi_pp-GQKy8jTPCrEpqRUtlrk8OXw

Another notable result came from the [NWChem](https://www.nwchem-sw.org/) benchmark, with the GH200 coming in second place at 1403.5 seconds.

  

![](/uploads/5vFBmCRUcbfgb9PR7NpFoXpSeRSFqr47lr9V1YtOG-k2Aus1jPY1T217XD_EnBKKZKnn8MTFUs-Jx2lMYVR_8Ek4Q9eQcM-kml5l0C2B4wfzwdytunAxtEcGZNfi_pp-GQKy8jTPCrEpqRUtlrk8OXw.png)

Results for NVIDIA GH200 running NWChem benchmark. (Image credit: Phoronix)

  

On a geometric mean basis across all the benchmarks, the GH200 Grace CPU performance was perfectly respectable overall. 

  
![](/uploads/O_VgErmTsszfyrnSV20nbY-BdZb4QvnVJe7LOp55l33JKKI3JFoQSRpWU2lia3g_QlqKTwi4_1VXjUQ2gE1iHKNYy_A4JkgtMsPnn5uzc7TBf9JjVzH1rifiNFA_n17d4TI_t7MB6bsjIMCMYGMcrXo.png)

(Image credit: Phoronix)

  

A set of GPU benchmarks run by researcher Simon Butcher [compared](https://blog.hpc.qmul.ac.uk/benchmarking-grace-hopper-nodes/) performance of the PyTorch ResNet50 [training recipes](https://github.com/NVIDIA/DeepLearningExamples/tree/master/PyTorch/Classification/ConvNets/resnet50v1.5) published by NVIDIA. The ImageNet 2012 dataset (150GB) was used, and training was run for 90 epochs, or about an hour. The GH200 showed excellent performance here.

  

![](/uploads/qlwGIbeeRNXXquUTSoF6EDJmoss443fwqWUEQg9PtgAVPwIHTNTbi0U5BsEFx3Wx2dX10b0gfTdlgMfQxU_kHulwTU5QaOGlz7Oy9U8GBpdkvj4pCE7wjW7PJzfU_lTK-mdd3VUlbSjUaXEx2RVtloA.png)

(Image credit: Simon Butcher, Queen Mary University of London)

  

NVIDIA also [published](https://resources.nvidia.com/en-us-dgx-gh200/nvidia-grace-hopper-superchip-datasheet) some performance comparisons that may be of interest.

  

![](/uploads/dXjLeW3nCbOSpjMxABHSZ6LP84CLuzszoNLvi2m9CD1Dxf9GgTYk17wr2YlxWxqNM5xF3U6GXV-E6JFILwTqmj4zEmdd9Rv5MNAUNK9xaa6YeTfM9krmcLeMRaUQ4eF8qt_G0-xW81JoRiaEz4157xQ.png)

(Image credit: NVIDIA)

  

### Final Thoughts

  

The NVIDIA GH200 Grace Hopper Superchip delivers the performance required for truly giant-scale AI and HPC applications running terabytes of data. Whether you're a scientist, an engineer, or a massive data center, you'll find it up to the task. 

As for what the future holds? Nvidia [recently announced](https://www.tomshardware.com/pc-components/gpus/nvidias-next-gen-ai-gpu-revealed-blackwell-b200-gpu-delivers-up-to-20-petaflops-of-compute-and-massive-improvements-over-hopper-h100) the GH200's successor: Grace Blackwell B200, the next-generation data center and AI GPU.

  

Here at Vast.ai, we're proud to say that we recently added ARM64 and GH200 support to our lineup. On our GPU rental platform, just about anyone can get access to top-tier GPUs at affordable rates for a variety of computational needs. You can skip the cost-prohibitive investment in hardware and get right to work today – we'd love to help you discover the possibilities!

NVIDIA's highly anticipated next-generation GH200 Grace Hopper Superchip platform is finally available. Perfectly suited for the era of accelerated computing and generative AI, it's essentially the world's most powerful GPU paired with the most versatile CPU.

Examining the NVIDIA GH200


It's time for another product update! Here at Vast.ai, we strive to keep improving our product for the benefit of new and old customers alike, so we can continue to serve as your reliable gateway to affordable and powerful cloud GPU resources.

One of our big updates this month is that we've added beta hosting support for AMD GPUs! For AMD-specific directions, please see our updated hosting guide. If you have Radeon or Radeon Instinct GPUs, you can now run our Daemon software. (Please reach out via the chatbox at the bottom right if you experience any issues while setting up AMD GPUs.)

As always, we address bugs as soon as we're aware of them, in order to maintain the highest standards of performance and reliability on Vast. The following are some of our latest bug fixes:

- SHA256 Docker images now work
- Search results by geolocation issue fixed
- Fixed the VRAM cost-per-hour filter and sort setting in GUI
- Instance duration bug fixed so 3-day default is now accurate
- Fixed issue where logging in with OAuth + 2FA fails
- Issue with copy progress bar being lost on page refresh now fixed

### NVIDIA Cloud GPU Updates

With each update, we're committed to enhancing performance, expanding capabilities, and streamlining user experience. Our goal is to continually evolve to meet our customers' needs. Below are some highlights of the latest enhancements we've made to our website and platform:

- Reorganized our Docs and added a more detailed walkthrough of all pages inside the console
- Updated hosting documentation with requirements for datacenter status
- Added machine listing support for "geforce rtx 4080 super", "geforce rtx 4070 ti super", "quadro k620", "quadro k2200" to gpu list
- Added the ability to stop/start offline instances
- Autoscaler now supports templates
- Added cpu_ghz filter
- Cloud copy settings now remembered from prior use

For a full list of all improvements and bug fixes since our last update, check out the March 2024 change log at the bottom of this page.

If you're new to Vast, we highly recommend joining our active Discord community – whether for helpful resources and support or just for fun. As of this month, we even have a new Discord bot named @Vasti that has been trained on our Docs and is always available for different support requests. Simply tag it in a public channel or send it a direct message!

For any other questions or assistance getting started with GPU rental or hosting on Vast, feel free to reach out at your convenience – we would be delighted to help you in any way we can. Drop us a line via the chatbox at the bottom right or email us at support@vast.ai any time!

### Change Log

- Added a reactive 2FA button
- Updated hosting documentation with requirements for datacenter status: https://vast.ai/docs/hosting/datacenter
- Added machine listing support for "geforce rtx 4080 super", "geforce rtx 4070 ti super", "quadro k620", "quadro k2200" to gpu list
- Added invoice info section to account and earnings page
- Added improvements to invoicing: refresh button, calendar range, re-trigger to get last invoice
- UI now shows recent failed billing attempts and explanation for when auto-billing is disabled
- Bugfix: fixed copy between two machines on the same network with the same IP
- Bugfix: fixed an issue in the templates with Extra_Filters not parsing non-floats properly
- Added the ability to stop/start offline instances
- Added CLI command 'create template' that can create a template, use --help for info
- Autoscaler now supports templates
- Added a 'machine_id' filter for hosts to find their own machine
- Added a button to show more results (so hosts can find their offers in relative rankings)
- Improved the add credit UI widget
- Improved max instance duration and price and other filters/sliders
- Added cpu_ghz filter
- Bugfix: SHA256 Docker images now work
- Bugfix: search results by geolocation issue fixed
- Bugfix: fixed issue with instance create scheduling
- Bugfix: fixed the VRAM cost-per-hour filter and sort setting in GUI
- Bugfix: added proper spacing support for GPU names in CLI searching
- Bugfix: fixed instance duration bug so 3-day default is accurate
- Fairly major reorganization of our docs, including a more detailed walkthrough of all pages inside the console - https://vast.ai/docs/console/introduction
- Added CLI function cancel_maint for hosts to be able to cancel a maintenance window
- Improved periodic auto cleanup of Docker storage for machines
- Bugfix: fixed 403 issue when creating new template
- Bugfix: host payout PDFs had total earnings and total payouts mislabeled, now fixed
- Bugfix: reserved pricing was displaying a NaN value, now it is showing correct values
- Bugfix: issues with instance→instance copy UI state
- Bugfix: fixed issue where logging in with OAuth + 2FA fails
- Bugfix: fixed billing PDF "include payments" issue where only auto billings were showing up
- Bugfix: crypto payments now properly exported in PDF
- Our new Discord bot named @Vasti has been trained on our Docs and can be handy for different support requests – give it a try by tagging it in a public channel or sending it a DM on Discord
- Added day to PDF invoices
- Bugfix: fixed issue with tag menu in template editor
- Bugfix: fixed invalid queries being sent from the webUI
- Added beta hosting support for AMD GPUs! (Our hosting guide is updated with AMD-specific directions. If you have Radeon or Radeon Instinct GPUs you can now run our Daemon software. Please reach out on our website chat if you experience any issues while setting up AMD GPUs.)
- Added an SSH key manager for each instance UI; click on the key icon to see what SSH keys are active for that instance
- Added a reminder to refresh the UI when there are deployed changes to the UI available
- Added UI help when using the instance copy command
- Cloud copy settings now remembered from prior use
- Bugfix: issue with copy progress bar being lost on page refresh now fixed


Vast Product Update | March 2024


We're pleased to reveal the latest round of updates here at Vast.ai! Our mission is to provide our customers with the best possible GPU rental platform, and we believe that every update is an opportunity to make our product shine that much more. For instance, we've recently added ARM64 and GH200 support – so we're now able to support ARM64 platforms including the Grace Hopper superchip!

But before we delve into some of our bigger updates below, first up on the agenda are bug fixes. We strive to address every minor hiccup and bug as quickly as possible, so that you can enjoy a seamless, uninterrupted experience on Vast. Here are a few of our latest bug fixes:

- Linking to docs now aligns correctly
- Random logouts from the webUI have been fixed
- Users don't see the host setup command unless they are in a hosting account
- Fixed an off-by-1 error with the CLI command ssh-url
- Everywhere a template name is mentioned, it is now linkable with a hash ID
- An issue with number of referrals not displaying has been corrected
- Fixed issue with template editor in search interface

![](/uploads/feb24Image.png)

### NVIDIA Cloud GPU Updates

Over the past few weeks, we've made some important enhancements to our platform. It's our hope that these updates will help save you time and effort, as well as maximize your goals and results on Vast. The following are some highlights of the latest changes:

- As mentioned, we've added ARM64 and GH200 support; we now support ARM64 platforms including the Grace Hopper superchip
- Added a template for NVIDIA RAPIDS - [https://cloud.vast.ai/?template_id=ce0f0b38b72857851f17c71492870593](https://cloud.vast.ai/?template_id=ce0f0b38b72857851f17c71492870593)
- Changed Autoscaler CLI commands to work with template linking
- Added a search feature for templates and hover help (tooltip) to the template page
- Implemented flexible date format for CLI so it isn't just Unix epoch time
- Improved search API for performance
- Added a DLPerf filter

Please check out the full change log for our February 2024 product updates at the bottom of this page. Getting to know the latest capabilities will help you make the most of your time on Vast!

If you're looking for affordable, on-demand cloud GPU rental, you've come to the right place – and if you need some assistance getting started, we have the experience and know-how to guide you through whatever your project requires. Join our thriving Discord community for support or reach out to our customer service team to get all your questions answered! We would love to help you explore the possibilities that Vast holds for you today. Please feel free to contact us through the chatbox at the bottom right or drop us a line at [support@vast.ai](mailto:support@vast.ai) whenever you'd like!

### Change Log

- Added ARM64 and GH200 support; we now support ARM64 platforms including the Grace Hopper superchip
- Added a CLI parameter on create instance for a template hash; i.e. vastai create instance --template_hash HASH
- WebGUI uses cancel_unavail by default (instances are not created stopped)
- Added a template for NVIDIA RAPIDS - [https://cloud.vast.ai/?template_id=ce0f0b38b72857851f17c71492870593](https://cloud.vast.ai/?template_id=ce0f0b38b72857851f17c71492870593)
- Upgraded our SD guides with ai-dock template and info on setting a strong password
- Improved Autoscaler to use template info more completely
- Changed Autoscaler CLI commands to work with template linking
- Bugfix: args launch now doesn't accept an onstart script
- Bugfix: linking to docs now aligns correctly
- Bugfix: random logouts from the webUI fixed
- Bugfix: users don't see host setup command unless they are in a hosting account
- Bugfix: fixed an issue with number of referrals not displaying
- Bugfix: fixed host link to Stripe connect earnings dashboard
- Bugfix: on instance to instance copy, copy to /root is fixed
- Bugfix: fixed issue with template editor in search interface
- Added a search feature for templates
- Added a "Create new template" feature on templates
- Improved search API for performance
- Added the team name and member name for the API/team contexts on the account page
- Added hover help to the template page
- Added DLPerf filter
- Added flexible date format for CLI so it isn't just Unix epoch time
- Bugfix: everywhere a template name is mentioned, it is now linkable with a hash ID
- Bugfix: loading bar was showing before transfer commands, it is now fixed
- Bugfix: template tag dropdown selector is fixed, as it wasn't updating correctly
- Bugfix: fixed an off-by-1 error with the CLI command ssh-url


Vast

Maximizing Value with NVIDIA A40 & RTX A6000 | June 2024

Vast Product Update | April 2024

Examining the NVIDIA GH200

Vast Product Update | March 2024

February 2024 Product Update