What is our primary use case?
We've only been using it for about a month so far. This is a system that's on loan to us from HPE. It's a Gen10 version with eight NVIDIA V100 GPUs and four nodes. We have already purchased the unit. This is on loan to us until we receive the Apollo 6500 that we ordered.
For storage we're using a Seagate SSD Array, all-flash array, as well as EL4000. The Apollo 6500 is for machine-learning, specifically for wafer generation, wafer analysis, for one of our operations sites in Minnesota.
What needs improvement?
I would want to see the flexibility of being able to run various network protocols including InfiniBand, Fibre Channel, as well as iSCSI, with iSCSI going up to 100 gigabytes per second -that would be outstanding. That in conjunction with what Mellanox offers would provide us with a very high-speed networking interface.
The other thing is we may could, perhaps, use more GPUs in the future, go from eight to 16 GPUs per instance. That could run head-to-head against the DGX-1, the DGX-2 that NVIDIA has developed in their own chassis. That would be interesting to see.
For how long have I used the solution?
Less than one year.
What do I think about the stability of the solution?
Excellent product. It's extremely reliable so far. The loan-er model we have is excellent. We have had no problems with it.
What do I think about the scalability of the solution?
In some ways, we think it may go beyond what we need moving forward. We don't know yet. We're going to buy another Apollo 6500. We may configure it with half the number of GPUs because that may be all we need. In a sense, we can see the Apollo 6500 being so powerful that we only need half the GPU capability that we have now. But that's what we think we're going to end up seeing as we continue to go through this process of machine-learning.
How is customer service and technical support?
Tech support has been outstanding. In fact, what HPE is doing is helping us develop the software stack for us to be able to move forward with this whole approach. Our intent is to develop a machine-learning and inference capability within all of Seagate operations, which include eight sites around the world.
My expectation is that this is going to be a rather huge improvement in our operations process. It takes about six months for us to build a single hard drive, and we sell millions of them per year. So you can imagine how important it is for us to develop an analytics capability that HPE is offering us. So it isn't just the Apollo 6500, it's also the software stack that runs on top of it.