Lightnews — Scholar-powered news

datacrunch.io @datacrunch.io · Aug 6

Up to 26% off: The most affordable Serverless Inference on the market.

Easy container deployment with request-based auto-scaling, scale-to-zero, and pay-per-use:

→ Interruptible spot pricing (50% off)
→ B200, H200, and more
→ Multi-GPU support

Learn more: datacrunch.io/serverless-c...

datacrunch.io @datacrunch.io · Jul 31

Text-to-image generation without the "AI look"? 📸

Built by Black Forest Labs and Krea, FLUX.1 Krea brings exceptional realism with a distinct aesthetic and flexibility.

Available on DataCrunch for cost-efficient inference at scale for $0.020 / image ⬇️

datacrunch.io/managed-endp...

datacrunch.io @datacrunch.io · Jul 23

✅ LIMITED OFFER: Get a 20% bonus on all your top-ups until the end of this week.

We're offering cloud credits to thank you for building with the DataCrunch Cloud Platform.

Sign in to top up: cloud.datacrunch.io/signin?utm_s...

Or learn how from our docs: docs.datacrunch.io/welcome-to-d...

1

datacrunch.io @datacrunch.io · Jul 15

Additional 8x NVIDIA B200 SXM6 – now available on the DataCrunch Cloud Platform.

Self-service access without approvals – peak flexibility with unmatched prices:

→ Fixed pay-as-you-go: $4.49/h
→ Dynamic: $2.80/h
→ Spot: $1.40/h

Deploy now: cloud.datacrunch.io/signin?utm_s...

datacrunch.io @datacrunch.io · Jul 11

10% off H200 SXM5 141GB – from $2.90/h per GPU down to $2.61/h ✅

This pricing applies to:
→ NVLink instances (1x, 2x, 4x, and 8x)
→ InfiniBand clusters (16x–64x) with 1-day contracts

Deploy now: cloud.datacrunch.io/signin?utm_s...

datacrunch.io @datacrunch.io · Jul 7

🇫🇷 TOMORROW: Our side event for Raise Summit with Hugging Face and SemiAnalysis on #SovereignAI.

Join us alongside other AI engineers and founders from 18:00 to 21:00 at Station F ⬇️

Sign up on Luma: lu.ma/qx7ydhe6?utm...

datacrunch.io @datacrunch.io · Jul 4

Instant Clusters – now available at the same price per GPU as VMs: $2.90/h ✅

→ 16x-64x H200 SXM5 141GB with 3.2 Tb/s InfiniBand™ interconnect
→ Pre-installed Slurm for easy job scheduling
→ Self-service access without approvals
→ 1-day contracts

cloud.datacrunch.io/signin?utm_s...

1

datacrunch.io @datacrunch.io · Jul 4

🇫🇷 Join our exclusive event in Paris on July 8 at 18:00-22:00 with Hugging Face and SemiAnalysis.

We'll explore Sovereign AI and the software-hardware stack making it a reality in regulated industries such as defense and healthcare.

Save your spot ⬇️

lu.ma/qx7ydhe6?utm...

Towards Sovereign AI with Hugging Face, SemiAnalysis, and DataCrunch · Luma

Join us for an exclusive event dedicated to Sovereign AI and the software-hardware stack required to make it a reality. Get first-hand insights into AI…

lu.ma

datacrunch.io @datacrunch.io · Jul 2

We tested the NVIDIA #GH200 system, where a GPU and a CPU act under a unified memory.

The NVLink-C2C connection offers a total bandwidth of 900 GB/s (450 GB/s per direction).

That is 7 times higher than a conventional PCIe connection.

Read more ⬇️

datacrunch.io/blog/data-mo...

datacrunch.io @datacrunch.io · Jul 1

Higher capacity = lower prices ✅

→ B200 SXM6 at $4.49/h
→ H200 SXM5 at $2.90/h

Both platforms are available on DataCrunch with self-service access and without approvals.

Deploy now: cloud.datacrunch.io/signin?utm_s...

datacrunch.io @datacrunch.io · Jun 26

🆕Inference API for the open-weight FLUX.1 Kontext [dev] by Black Forest Labs

The new frontier of image editing, running on the DataCrunch GPU infrastructure and inference services with the additional efficiency boost from WaveSpeedAI.

$0.025 per image: datacrunch.io/managed-endp...

datacrunch.io @datacrunch.io · Jun 25

❗️ We just expanded our capacity of B200 SXM6 180GB servers – available in the DataCrunch Cloud Platform.

The best thing is…

You can deploy the Blackwell platform without approvals.

Just sign in, select the instance type, and start your deployment:

cloud.datacrunch.io?utm_source=b...

1

datacrunch.io @datacrunch.io · Jun 24

Our step-by-step guide for integrating Pyxis and Enroot into distributed workloads using TorchTitan, ensuring scalability and reproducibility.

Try it today with our Instant Clusters: 16x-64x H200 SXM5 141GB with InfiniBand interconnect:

datacrunch.io/blog/pyxis-a...

Pyxis and Enroot Integration for the DataCrunch Instant Clusters

Step-by-step configuration and testing of multi-node distributed workloads using TorchTitan and by integrating with Pyxis and Enroot.

datacrunch.io

datacrunch.io @datacrunch.io · Jun 23

📢 CUSTOMER STORY: How Freepik scaled FLUX media generation to over 60 million requests per month with DataCrunch and WaveSpeedAI.

Read the full story with ⬇️

- Our research into lossless optimizations
- Inference benchmarking
- Future predictions

datacrunch.io/blog/how-fre...

How Freepik scaled FLUX media generation to millions of requests per day with DataCrunch and WaveSpeed

Cost-efficient and low-latency image generation without compromising on model output with high GPU utilization, elastic scaling, and near-zero cold starts.

datacrunch.io

datacrunch.io @datacrunch.io · Jun 11

NVIDIA CEO, Jensen Huang, held his keynote today at NVIDIA GTC Paris 2025 and Viva Technology. As always, he gave an insightful presentation with numerous highlights!

One of ours was getting featured among the key European Cloud Service Providers!

What was yours?

1

datacrunch.io @datacrunch.io · Jun 10

We kicked off our summer at AaltoAI Hack 25.

It was amazing to see what 25 teams could build in 48 hours, with most deploying cutting-edge hardware on the DataCrunch Cloud Platform.

We thank AaltoAI for this opportunity to support the next generation of AI builders in Finland 🇫🇮

1

datacrunch.io @datacrunch.io · Jun 9

🇫🇷 DataCrunch is coming to NVIDIA GTC Paris 2025 and Viva Technology.

📨 If you've been looking to get in touch, feel free to connect with our CTO, Arturs Polis.

1

datacrunch.io @datacrunch.io · May 29

📢 Stay tuned for FLUX.1 Kontext [dev] – an open-weight version coming soon to the DataCrunch Cloud Platform with the WaveSpeedAI inference engine.

datacrunch.io @datacrunch.io · May 29

🆕 Inference API for FLUX.1 Kontext [max] & [pro] are now available on DataCrunch!

We are an infrastructure partner of Black Forest Labs for Kontext, a suite of generative flow matching models for text-to-image and image-to-image editing.

Learn more: datacrunch.io/managed-endp...

1 1

datacrunch.io @datacrunch.io · May 26

🚨 Summer Inference by Symposium AI is happening next Wednesday, June 4, at 16:00-22:00.

🇫🇮 This event will bring together 250 AI engineers, researchers, and founders under one roof in Helsinki.

🔗 You can still grab one of the last remaining seats: lu.ma/x5hhj79x

Symposium AI - Summer Inference · Luma

Join 250 leading AI builders for an epic night in Helsinki! Symposium AI events bring together top AI talent, researchers, and engineers who are actively…

lu.ma

1 1

datacrunch.io @datacrunch.io · May 22

📈 Due to high demand, we'll add more B200 SXM6 servers to our on-demand pool in early June.

⚡️ You'll have self-service access to more of this next-gen hardware without quotas, approvals, or sales calls.

🔗 Join the waitlist or reserve your capacity: datacrunch.io/b200#waitlist

datacrunch.io @datacrunch.io · May 20

We're in Taipei for Computex this week. Let's connect! You can reach out to Ruben, Jorge, and Anssi.

We also recommend you attend the after-hours meetup by SemiAnalysis on Wednesday.

lu.ma/b9bw7xxz

datacrunch.io @datacrunch.io · May 15

*Due to high demand, it can be hard to catch more than 4x nodes in the wild.

If you need larger capacities, drop us a line at:
- datacrunch.io/contact
- [email protected]

datacrunch.io @datacrunch.io · May 15

Great news about Instant Clusters:

1️⃣ We've lowered the minimum contract duration to 1 day
2️⃣ You can deploy 4x 8H200 nodes right away* for $121.44/h

Get instant, self-serve access: cloud.datacrunch.io

1

datacrunch.io @datacrunch.io · May 8

What's the secret sauce for efficient transformer inference? 🥫

At least one of the ingredients is Multi-Head Latent Attention.

Check out our comparison of theoretical and practical performance between GQA vs. MHA vs. MLA ⬇️

datacrunch.io/blog/multi-h...

Multi-Head Latent Attention: Benefits in Memory and Computation

Multi-Head Latent Attention (MLA) vs. Group Query Attention (GQA): Transformer inference optimization in DeepSeek V3 with lower KV cache and higher FLOPs/s.

datacrunch.io