.svg){kind=small}
.svg){kind=small}
September saw the team at QWERKY AI expand in a very exciting way: with our new server, stocked with 8 NVIDIA RTX Pro 6000 Blackwell Max-Q Workstation Edition cards! While it may look like just another rack of hardware, for our team it represents a leap forward in possibilities. We’re especially excited that this system runs on NVIDIA’s new Blackwell architecture. Stay tuned for our next blog post, where we share more on our updated Blackwell-powered tooling.
Each unit boasts 96GB of GDDR7 memory, fifth-gen Tensor Cores and fourth-gen RT Cores. Until now, optimizing large AI models meant carefully rationing compute resources. With our new server, our engineers have direct, in-house access to GPU power at a scale we didn’t have before.
That means:
- Lower latency in experimentation
- Developing and testing at a pace consistent with our product roadmap
- No more waiting in line for cloud resources
- Reducing our dependence on expensive cloud capacity
Qwerky’s mission is to develop AI technology that reduces effort, lowers costs and helps people and businesses get more done with less. This server enables us to optimize larger and more complex models and run more experiments in parallel — making AI models more efficient, more accessible and more sustainable.
Installing the server wasn’t just plug-and-play. Our team had to carefully manage power, cooling and logistics to get it running smoothly. Fun fact: it took three people just to get the server in position — a reminder that breakthroughs sometimes begin with good old-fashioned heavy lifting.
Here are some notable specs from our server build:
- Dual AMD EPYC Turin 9475F processors with a combined 96 cores and 192 threads: the kind of compute muscle designed for massive parallel workloads
- 1.1 TB of DDR5 ECC memory: more than enough to handle some of the largest AI models and data pipelines without breaking a sweat
- 8x NVIDIA RTX Pro 6000 Blackwell Max-Q Workstation Edition GPUs, each with 96GB of ultra-fast GDDR7 memory, totaling nearly 800GB of GPU memory and unlocking the ability to optimize huge models in-house
- Over 18TB of enterprise NVMe SSD storage: ensuring lightning-fast data throughput for large-scale training and inference
