DapuStor Haishen5 SSD H5100 30.72TB Basic Performance
For this, we are going to run through a number of workloads just to see how the drive performs. We would also like to provide some easy screenshots of the desktop tool so you can see the results quickly and easily compared to other drives you may have.
CrystalDiskMark 8.0.4 x64
CrystalDiskMark is used as a basic starting point for benchmarks as it is something commonly run by end-users as a sanity check. Here is the smaller 1GB test size:

Here is the larger 8GB test size:

In the event you want to see a side-by-side, here they are:

As we see with many or the larger and faster drives, the smaller transfer windows are a bigger challenge for the controller.
ATTO Disk Benchmark
The ATTO Disk Benchmark has been a staple of drive sequential performance testing for years. ATTO was tested at both 256MB and 8GB file sizes.

Here is the 8GB result:

For those who want to see the results compared side-by-side:

Here we are actually a bit below some of the other benchmarks. We can see the solid read speeds though.
AS SSD Benchmark
AS SSD Benchmark is another good benchmark for testing SSDs. We run all three tests for our series. Like other utilities, it was run with both the default 1GB as well as a larger 10GB test set.

Here is the 10GB test size:

Again, here is the side-by-side.

This is very impressive for such a large capacity drive.
Next, let us get into some of our Linux-based benchmarking.
DapuStor Haishen5 SSD H5100 30.72TB Four Corners Performance
Our first test was to see sequential transfer rates and 4K random IOPS performance for the DapuStor. Please excuse the smaller-than-normal comparison set. In the next section, you will see why we have a reduced set. The main reason is that we swapped to a multi-architectural test lab. We test these in more than 20 different processor architectures spanning PCIe Gen4 and Gen5. Still, we wanted to take a look at the performance of the drives.

Here is the 4K random read-and-write performance:

Overall, these are fairly good results. At some point, the 4K random read IOPS and the sequential results really perform well.
DapuStor Haishen5 SSD H5100 30.72TB Application Performance Comparison
For our application testing performance, we are still using AMD EPYC. We have all of these working on x86 but we do not have all working on Arm and POWER9 yet so this is still an x86 workload.

As you can see, there are a lot of variabilities here in terms of how much impact the DapuStor SSD has on application performance. Let us go through and discuss the performance drivers.
On the NVIDIA T4 MobileNet V1 script, we see very little performance impact on the AI workload, but we see some. The key here is that the performance of the NVIDIA T4 mostly limits us, and storage is not the bottleneck. We have a NVIDIA L4 that we are going to use with an updated model in the future. Here we can see a benefit to the newer drives in terms of performance, but it is not huge. That is part of the overall story. Most reviews of storage products are focused mostly on lines, and it may be exciting to see sequential throughput double in PCIe Gen3 to PCIe Gen4, and double again from Gen4 to Gen5, but in many real workloads, the stress of a system is not solely in the storage.
Likewise, our Adobe Media Encoder script is timing copy to the drive, then the transcoding of the video file, followed by the transfer off of the drive. Here, we have a bigger impact because we have some larger sequential reads/ writes involved, the primary performance driver is the encoding speed. The key takeaway from these tests is that if you are mostly compute-limited but still need to go to storage for some parts of a workflow, the SSD can make a difference in the end-to-end workflow.
On the KVM virtualization testing, we see heavier reliance upon storage. The first KVM virtualization, Workload 1, is more CPU-limited than Workload 2 or the VM Boot Storm workload, so we see strong performance, albeit not as much as the other two. These are KVM virtualization-based workloads where our client is testing how many VMs it can have online at a given time while completing work under the target SLA. Each VM is a self-contained worker. We know, based on our performance profiling, that Workload 2, due to the databases being used, actually scales better with fast storage and Optane PMem. At the same time, if the dataset is larger, PMem does not have the capacity to scale, and it is being discontinued as a technology. This profiling is also why we use Workload 1 in our CPU reviews. DapuStor’s random IOPS performance is really helping here. On Workload 2, and the VM Boot Storm, we see the performance of the drives is very good.
Moving to the file server and nginx CDN, we see much solid QoS and throughput from the DapuStor SSD. The drive pulls ahead on the file server due to its faster sequential speeds. On the nginx CDN test, we are using an old snapshot and access patterns from the STH website, with caching disabled, to show what the performance looks like in that case. Here is a quick look at the distribution:

The storage class memory drives are better here, but the PCIe Gen5 drives are performing very well.
Now, for the big project: we tested these drives using every PCIe Gen4 architecture and all the new PCIe Gen5 architectures we could find, and not just x86, nor even just servers that are available in the US.
Found another of Patrick’s Easter Eggs! “storage-class memeory drives”
As a data hoarder I’m so happy we’re finally getting some decent capacity drives trickling down to affordable second-hand prices. Still a way to go before I’m truly happy but maybe a few of those upcoming 128TB drives might just do the trick!
We really need a Synology/QNAP offering that offers 4-8 bays of PCIe gen5 in a standard small desktop format enclosure.
We are getting to the point where large scale SSDs make more sense from a capacity standpoint than hdds do.
@James: so a budget solution like Synology or Qnap to host enterprise grade drives? Doesn’t make sense to me.
No inside view? So disappoint.
> “There are also smaller storage-class memory drives that are often in the 800GB to 1.6TB range.”.
The KIOXIA FL6 SCM Series goes up to 3.2 TB.
So how can I get an idea on the prices of these drives mentioned here? It’s not like they’re available at AMZ.
How about prices? Yes, exact values may depend on factors such as quantity, but a figure should be provided. It’s not a national secret.
@James: even if we did have a NAS with 4 bays of gen5 U.2 bays, it would need a 400gb ethernet interface to saturate those drives. You couldn’t tell the difference between four gen3/4/5 drives with a more commonplace 10/25/100gb NIC.