HPE ML350 Gen10 Review // Tower Server for a HomeLab?

This is my new home lab server, the HPE ProLiant ML350 Generation 10. And it's a real beast, an enterprise tower server with hot swap drives, remote management, lots of expansion, and enough room to build something much bigger than a typical home server setup. And some of you might wonder, why the heck did I even buy this? Didn't I say in previous videos you should probably just use mini PCs or desktop PCs for your home lab? Well, the features and experience gained from such enterprise gear, especially with the value of refurbished hardware, may outweigh the extra noise, power, and size concerns that sometimes come with it. But I want to be clear, both sides have a point here, and it really depends on what matters more to you. So, that's why I wanted to test both setups from my own perspective, and I want to show you the advantages you get with such a device, but also the problems you might run into and the tradeoffs, so that you can better decide for yourself if professional server hardware really makes sense for your home lab or if you should just keep running consumer hardware instead. Now, to explain why I ended up with that specific server, I need to go back a little because my own home lab hardware journey started almost completely opposite from where it is today. When I worked as an IT service technician, professional server hardware was just normal for me. I worked with

rack servers almost every day. I swapped redundant components, used integrated management systems, and I installed server operating systems, mostly Windows servers, of course. And back then, the hardware felt exciting. It was something completely different from the hardware that I used at home. And this is true, but when you bring this kind of hardware into your own space, you will quickly notice the other side of it because these systems can be huge, loud, heavy, and power hungry. And if your office is right next to you, which currently is the case for me because I moved to a new apartment when I started my YouTube channel, hearing the fans constantly spin up like a tiny jet engine, that just becomes annoying very quickly. So, for a while, I moved completely away from running enterprise gear at all, and I started building my home server rack and used rack mount cases, but with just normal consumer components. I used normal MATX ITX boards, power efficient CPUs like mobile chips, quiet and fans.

And for most of my use cases, these systems had enough flexibility for what I needed. And honestly, for many home lab setups, I think this is still a great middle ground. But then, when I started to build my Proxmox cluster, even these desktop style systems started to feel a bit too large for the job. So, then I tried out mini PCs because they were just getting better and better, and suddenly they made a lot more sense for my setup. So, even though I originally built my two-node Proxmox cluster using desktop PC components in rack server cases, I finally migrated to use smaller mini PCs here as well because they are very small, quiet, power efficient, and surprisingly powerful. But this is also where the expectation matters. If your goal is to just run a couple of services at home, maybe run Docker, a few VMs, or even run multiple Proxmox nodes for just self-hosting, mini PCs are really great.

And honestly, for many, many home labs, they're probably the perfect choice. But if your goal is to learn how infrastructure hardware actually works in companies, then at some point you miss a lot of the experience you only get with real server platforms. Because the truth is companies usually don't run their critical infrastructure on tiny mini PCs or on consumer hardware. Because they need to use systems with redundant power supplies, with hot swap drive back planes, remote management, more memory capacity, more PCIe expansion, error correction or protection, hardware monitoring, and all of these important things that make servers feel like servers. And there was a point where I started thinking about, well, maybe I should get at least one professional server again for my home lab. Not because everyone needs one, but because I wanted to keep experimenting with this hardware again, use it in my own lab, and make better tutorials for you guys around the things that you only learn when you actually touch this stuff. But I still did not want to go back to a professional rack server that I used in my days of being an IT service technician. Because from my experience, especially these low-profile one unit or two unit systems can be brutally loud.

So, for me, it needed to be some kind of compromise here. And that's why a tower server from HPE felt like the more realistic middle ground here. Of course, there are many other choices and options for professional enterprise gear available from Dell, Supermicro, Lenovo, and other vendors. But if you go down this path, and before we talking about the different hardware vendors and specifications, there's one more important question we need to talk about. Where do you actually buy professional server hardware for your home lab without just gambling on some random eBay listing? So, this is where Bargain Hardware comes in, who sponsored this video and sent me over the ML350 Generation 10 for testing. Bargain Hardware is a UK-based company that specializes in refurbished enterprise IT hardware. So, we're not talking about a normal PC shop here. We're talking about servers, workstations, storage hardware, networking gear, CPUs, memories, controllers, caddies, and all of the other parts that you usually need when you build with real enterprise platforms. The hardware they refurbish is carefully removed from real working environments. Then it goes through diagnostics and testing before and after it arrives at their facility. And that is super important because with used enterprise gear, the question is not only does it turn on, the real question is what condition is it in? Is the

configuration complete? And can I actually trust it enough to build my lab around it? On the Bargain Hardware's website, you can start with a server platform like HPE, Dell, Supermicro, Lenovo, depending on what they have in stock, and then configure CPU, memory, drives. So, instead of buying one random pre owned box where you later find out that half of the drive caddies are missing, you can configure the machine around the job you actually want it to do. I can just tell you from my experience, the whole shipping process was very fast and satisfying. After the agreement was done, the order was released for assembly. They told me they would fully stress test the machine before shipping, and then it was sent over on a small pallet because, well, this thing is huge. By the way, if you're buying from UK and Europe, the delivery is really fast, and they even have an EU Ireland fulfillment angle to avoid some of the usual post-Brexit VAT pain, which honestly is quite relevant if you're buying from Europe, but Bargain Hardware also has worldwide shipping options and sitewide a 3-year return to base warranty. So, if you're looking for refurbished enterprise hardware for your own home lab, I think Bargain Hardware is definitely a much more sensible way to buy than gambling on the cheapest eBay listing you can find. Of course, I will leave you a link to the website in the description box down below. And now, let's take a closer look at the exact unit they sent me and which choices actually matter for a home lab build like this. All right, guys.

So, that's the exact unit that I got for my home lab, the HPE ProLiant ML350 Generation 10 tower server. I've made good experience with HPE servers from my early IT service technician days. They were pretty common in our environments and we have sold and set up a couple of them for enterprise customers, but I've also made a few good experience with Dell servers. So, there are many different options you can choose from and my practical advice is to just first of all decide what form factor you want to get. Do you want a bigger tower server where you have more expansion options or do you want a tinier rack server case? And then just compare the different vendors, compare the different models bargain hardware has in stock, and then decide based on your preference and use case what model you want to get. Now, this ML350 Generation 10 is particularly interesting to me because it is a proper enterprise platform, but in a tower form factor that I can still put somewhere outside a server rack. And that was important to me because I'm moving to a new house somewhere next year and I wanted to be a little more flexible with this unit so that I could also spontaneously decide to put this somewhere in the basement without having to have a separate server rack or something like this. But yeah, we will see. I will keep you updated about this project. Now, one thing is also important, do not only buy one specific

model because of the model name. Two ML350 Generation 10 can have completely different setups. You can have different drive bays, backplanes, controllers, rising cards, power supplies, memory, uh configuration, and networking. So, always check what is available in this generation for that specific vendor and model. Now, here you can see all the different uh CPUs and configurations you can put into that specific model. Uh so, depending on how much power you need, I have chosen the Intel Xeon Gold 6,240. That is an 18-core CPU with 36 threads, 2.6 GHz clock speed up to 3.9 GHz boost. So, this particular CPU should have enough cores and threads for my virtualization and lab workloads without going straight to the highest and or most expensive CPU option. Note that many enterprise servers can be configured with one or two CPUs at the same time. So, you might think you could easily just double the CPU power of these devices. However, I want to warn you, two CPUs are not automatically better for a home lab because a second CPU adds more idle power, more heat, more cooling noise, and sometimes some hardware virtualization complexity. And because many home lab workloads are actually more limited by RAM, storage,

or network before they actually start getting limited by CPU cores, I would really be careful choosing this. I know it can be a little seductive to choose two CPU cores because they're so cheap, but at the end of the day, it might be smarter to just choose one CPU. You can later on still always add a second CPU that you can buy from bargain hardware without having to buy an entire new server. But, you have to keep in mind, if you want to upgrade to a second CPU, those two CPUs have to match. So, if you add a second one, you'll also need a proper heat sink, cooling parts, and you should check whether the system needs an additional fan. And another warning. Each CPU has its own memory channels. So, if you need to add a second CPU to an enterprise server, you should also populate memory for that second socket. I would plan at least four memory modules in total, so two per CPU, because then the memory is most efficiently balanced across both CPUs instead of having one CPU starved or creating a weird memory layout. By the way, speaking about memory for this server, I got two memory modules with each 32 GB of memory. So, I got two Hynix memory modules. They are DDR4 with ECC memory support. So, ECC means error correction code and this is specifically

existing in professional server environments, because this type of memory can detect and correct certain memory errors, which is valuable for long-running services, especially in storage workloads, backups, databases, or basically anything where data integrity matters a lot. By the way, on many professional server mainboards, you find so many more RAM slots that you're used to when looking at consumer hardware. The ML350 Generation 10 platform supports up to 24 memory modules and up to 3 TB of DDR4 memory in a full two CPU configuration. With only one CPU installed, only the memory slots for that CPU are usable. Of course, this can get really expensive if you're trying to max it out, but it is honestly an insane upgrade potential and I love to have that freedom to easily add more memory if I need. On smaller desktop or mini PCs, you're always limited to two or four memory slots. So, if you have a bigger home lab, many virtual machines, many workloads you're testing frequently, then it can make a lot of sense to buy a real professional enterprise server. Now, let's also talk about storage and drive base. And this is also very, very flexible. On this ML350 Generation 10 setup, I have an

eight SFF chassis where I bought six 1.2 TB SAS drives with 10,000 RPM and 12 GB per second interface speed. The drive base are hot-swap caddies, and this is one of the main reasons why you typically want a real server for storage. Hot-swap means the drives are accessible from the front and can be replaced while the server is running, assuming the storage setup supports it, of course. But this is a huge advantage because typical enterprise gear or servers are business-critical, and just to replace one drive, you cannot just reboot a server where many people are working on. So, that's why this is really important. SAS, by the way, means serial attached SCSI or scuzzy, and this is the enterprise server alternative to typical consumer SATA drives built for server backplanes, controllers, and continuous workloads. Enterprise drives can also come with hardware protection and other features that are not present in typical consumer drives. The same is true for SSDs, by the way. You will notice that these drives usually have a weird sizing, and also the RPMs are different than in consumer setups. This is also important because the higher RPM means the disk is spinning faster, so it can reach data faster, which results in

lower latency. That's why most enterprise drives are often 10K or even 15K RPM. However, there's of course also a trade-off. That is noise, heat, power usage, and capacity. So, even though the SAS drives are faster than consumer drives, that's why they have sometimes weird sizes like 1.2 terabytes. And this is completely normal in enterprise drives because vendors often optimize them around performance class, platter count, reliability, and server life cycle instead of the consumer capacity steps that we are used to. One interesting thought here, you might want to compare SAS to SSDs or NVMEs. So, SAS drives are still magnetic hard drives and they are of course much slower in latency. So, I would probably not choose SAS drives as the my fastest virtual machine database storage, but they can still make a lot of sense for bulk storage, backups, lab storage, and redundant storage pools where capacity, hot swap bays, and cost matters more.

This is also where the controller matters a lot. So, there are a few different options here that supports different features and RAID levels. I've chosen the HPE 208 that can support both SAS and SATA drives and it supports RAID levels 0, 1, 5, and 10. Note, if you might want to compare this to something like ZFS that we've covered in previous videos, ZFS is considered a software RAID. This controller as a real hardware RAID. So, it is different from software. With hardware RAID, the controller presents one logical disk on the operating system. So, the operating system would really just see one large drive. With ZFS, the operating system sees all the individual disks and ZFS handles all the redundancy checksumming scrubbing and self-healing at the file system storage layer. ZFS might be the better option, yeah? Because ZFS has a couple of features that hardware rates do not have. So, I really just wanted that for experimenting with hardware rate. Now, with this controller, you have both options and you can experiment with the one or another setup. However, it doesn't have a true HBA mode. There is a slight performance penalty when you're using that in software rate, but honestly, for my home lab setup, that was not really a problem. So, with this controller, I think I have good options

to test both setups. That particular unit that I got also has an HP NS204i NVMe controller with two 256 GB NVMe drives in a RAID 1 for the main operating system and most of the critical data. So, that means it is mirrored between those two NVMe's. You can only use the capacity of one NVMe drive, but in case that drive got damaged, the other drive can still run the main operating systems. All right. So, next, let's also talk about network interfaces. This unit has four regular network ports with 1 Gbit/s plus a separate dedicated iLO management port. We will cover iLO in a bit more detail later in the next section because I think this is also one of the most important advantages when using enterprise gear. You typically have multiple onboard ports and a lot of PCI Express expansion slot. You can see it supports up to eight full-height PCI 3.0 expansion slot. So, even though the four network interfaces are just 1 gigabit per second, you can easily expand this machine to add 10 or 25 or even more gigabit networking. This also leads into another big difference compared to most consumer platforms, by the way, because

in the normal desktop PC build, you always have to check how many real CPU lanes are available and how much is shared through the chipset. This is particularly a limiting factor for PCI Express expansion. If you want to use and many expansion cards, maybe for graphics cards, for different network cards, and maybe additional storage controllers, and so on. On desktop consumer hardware, you sometimes are limited. And this, of course, is also highly dependent on the CPU. Now, the Xeon 6240 has 48 PCI Express 3.0 lanes per CPU. This is also one of the huge advantages with tower servers. You have so much space, you have many PCI Express slots. You can actually put so many expansion cards to test into this thing. So, this is really amazing. Last but not least, I also want to talk about the power supply. Here, you can also choose between different power supplies depending on how many much power you need because you want to add more GPUs in and so on. And real enterprise servers usually also support not only one power supply, but two power supplies. A second power supply is mainly there for redundancy. So, if one power supply fails, the server can still keep running. For my use case, I typically wanted to get that high

efficiency titanium PSU with 800 W, I think that is completely enough. I don't really need a second one, but I could easily add it if I would want in the future. Just one more warning, if you think you can easily double the power of the PSU, so maybe you just want to get two 800 watts and you think you're going to get 1,600 watts, this is not true. It is just there for redundancies. All right, so yeah, these were all of the specs of the unit that I bought. Again, you can probably argue about the one or another decision, but yeah, bargain hardware supports many different configurations depending on what matters to you. One of the most important or most impressive features that professional server equipment usually comes with, and this is remote management. Because the HPE ProLiant series has something that is called iLO. iLO stands for Integrated Lights Out, and it's an HPE's BMC, so a baseboard management controller, which is basically a small separate management computer inside the server with its own network access independent from the installed operating system, and it is really cool. So, when you want to set up this box, you just connect a network cable to this iLO and let it get an IP address via DHCP from your router or firewall, and then open it in the browser. And this thing even works when the server is powered off, by the way.

So, as long as the server has standby power and iLO is connected to your network, you can access the server and even power it on from this uh management interface. And before you might say, "Oh well, for consumer PCs, I just use a remote KVM switch." iLO is more than just a KVM switch. You can see in here all of the details of your server, the CPU, the memory, the network settings. You can even get the IP addresses, MAC addresses, and the status. You have a device inventory of the RAID controller, the network and storage controller. You can inspect the temperature of all the different sensors in the system. You can add users, directories, groups. And when you take a look at this here, you can see you can launch a remote console, basically attaching a monitor, a keyboard, and a mouse to your server and manage it through the browser. Super simple and convenient. This is also the way how I have installed and set up this server, by the way. So, I just connected the iLO network interface and then opened the remote shell, and I've configured the storage controller before installing the operating system. Here you can choose all the different drive options. For production storage, you might think more carefully about the different RAID levels, rebuild risks, backups, and so on. Also, that's why ZFS

software RAID might be a bit more flexible. Again, I've already talked about this. Maybe I would switch that at some point in the future, but I really just wanted to experiment with this array controller here. Anyways, it is just a lab setup for now. And then I've used that to install Ubuntu Linux server. By the way, I did not want to run Proxmox again because I'm already running a free node Proxmox cluster, and I really wanted to have this server for more experimental tasks outside of my self-hosted stack that I yeah, already use in a productive way in my home lab. So, this part of my home lab is really just for playing around, experimenting, maybe doing a few videos about professional enterprise gear, and some of the options. By the way, if you're enjoying this video, don't forget to give it a thumbs up, and please leave me a comment what topics you're interested in. I've not really done so many videos about professional enterprise gear on my channel yet. So, that might be a good chance to just tell me what do you guys want to see and I can spend more time and effort experimenting with my new home lab server now and jump into specific topics in a little more detail.

But yeah, apart from that, it really feels like a normal server installation. I think I don't have to go through this in detail. However, there's one thing that I want to mention. As you can hear, I'm really excited about iLO, but one thing that is really annoying, if you're installing the server operating system, this is where licensing kicks in. This is one of the downsides of professional enterprise gear because post installation management and HPE seem to see that operating system installation part as post installation management as well. This is only half possible because the remote management console will kick you out of the session after some time without having the right iLO license. But honestly, this is mostly my fault because I didn't know at this point that iLO licenses are typically server specific. Bargain Hardware says on their homepage typically refurbished servers retain their original iLO license.

However, if a server's iLO license was an upgrade or add-on, it may not always transfer with the server when refurbished. So I probably was just unlucky. So in most cases, it shouldn't be a problem on refurbished server. If it is for you, then just reach out to the vendor. But if you want to manage your server from remote, you typically use SSH on a Linux server anyway. So you probably will not use iLO as a remote management to have a mouse, keyboard, and monitor on your server. Maybe just to see what's going wrong, press the reboot button, and something like this. For that, the unlicensed iLO is totally sufficient. One thing that I also probably want to mention at this point, one thing I noticed once I installed Ubuntu was that iLO could see the server hardware, but it did not show the full operating system level information. And iLO complained that it did not detect the agentless management service, the AMS, which is basically a small HPE service running inside Ubuntu that gives iLO the additional operating system level details like network interfaces, IP addresses, and so on. So, without that, iLO can still manage the server hardware, but it cannot see what operating system sees. And what I could easily done with my AI agent Pi, I could let it figure it out what was the problem. It confirmed, "Oh, it was AMS warning." And it was missing on Ubuntu Server LTS. So, Pi

checked the server over SSH on the IP address, confirmed it was that specific model running the Ubuntu latest version, checked whether AMSD was already installed and running, and then just installed everything the required app packages, tools, and then finally installed the utility. By the way, I will add these commands as a write-up on my website christianlemmer.de, so you can easily check it out if you're interested. And you might have an HP server where you have the same problem. But, yeah, apart from that, the experience was pretty well. I mean, there are a couple of things that are different in professional enterprise servers than there are in consumer hardware. And I think for that this project was already super valuable because you usually have no chance to touch on these enterprise gear. But, yeah, until now, I only talked about the good things of professional enterprise gear. Of course, there are also a couple of downsides or problems you might run into, especially when it comes to performance, noise, and power draw. By the way, one important thing, HPE servers have different BIOS workload performance profiles, and these can make a huge difference when it comes to throughput, fan behavior, and power consumption. And it's really a completely different world than in consumer grade hardware, where you can set, I don't know, the boost mode on CPUs. But these profiles, they really make a huge difference. I personally

didn't have much experience with these power profiles, also because I usually in my home lab don't have the right workloads to test them. I have just chosen virtualization maximum power because my main use case for this particular device is home lab virtualization and experimenting with all sorts of virtual machines, workloads, and I also seriously wanted to see what the server can do when it's tuned for more performance than for minimum idle power. So, keep that in mind what I'm saying about performance and power efficiency, that only makes sense together with these selected profile. This is the result that I got. So, I got a single core score of 1,506, multi-core score of 9,621. Honestly, it is not too impressive. I found similar consumer CPU benchmarks. For example, the Intel Core i9 uh 10,980XE benchmarks are somewhat in the same single core score and multi-core score range. That is also an 18-core, 36-thread CPU with a TDP of 165 W. So, this is pretty close. Then I've also checked the AMD Ryzen 9 5950X benchmarks. You can see that one is a little faster than my home lab server, and it's a consumer CPU with 16 CPU cores and 32 threads. So, again, this is

why refurbished enterprise gear can make a lot of sense. Even if it's not the most performant and most efficient system, it is pretty good deal. This CPU only cost 35 pounds, around like 40 euros or 46 dollars. But, what about the noise and the power consumption and the home reality? Can I actually live with this thing? Well, I'm going to say this. When the server is doing nothing, it is surprisingly quiet. But, when I'm starting virtual machines, booting the server, or if it's just in general under heavier load, you can definitely hear it's getting noisy. However, I also want to mention that this of course can vary depending on the specific model and the configuration of your server. So, this is of course just my experience as you can hear in the sample. However, when I'm moving to the new house next year, I hopefully have a separate room in a basement, which is much better for that. So, yeah, for smaller tests in the meanwhile, it's it's okay. But, yeah, that's just about the noise. So, for power draw, I also measured the power consumption directly in iLO. Unfortunately, the power meter or detailed power reporting is part of a licensed feature, but you still get the raw power consumption number in the power options. I also used an external wall power meter just to compare the numbers in the iLO to what the external

power meter reported, which by the way is kind of the same. And the idle power in efficiency mode was about 79 watts. I agree, this is not really a low number, especially compared to some of the desktop consumer hardware or the mini PCs that I've tested. When I was running Geekbench in the performance profile, it even went up to 150 watts. And this can be completely normal, especially when the CPU boosts and the server is using a more aggressive performance profile. Again, this comes down to the settings that you prefer. Especially with all the redundant enterprise components, the SAS drives, the storage controllers, the cards, the more fans, the more controllers, the more expansion you get, the higher the power draw will be. That is just normal. But again, if your goal is to have a home lab with a low power draw, professional enterprise gear probably might not be the best option. Still, I was expecting a much higher consumption, so it is certainly better than it was a few years ago when I still used enterprise gear with older generations. They were much more power hungry than this one. But yeah, it's always a compromise you have to make, and like I said in the beginning of this video, it really depends on what matters more to you. If you want to experiment with technology that is actually used in enterprise companies, this is exactly the kind of machine that I would want to get for my home lab. I can test iLO, I

can test the HPE components, redundant technology, and all of the great stuff you only see in enterprise servers. So now, I would love to hear your opinion. Please tell me, are you using enterprise gear in your home lab, or are you using mini PCs or desktop PCs? Just tell me, what do you think about it, and what is your experience? Let's have a chat on my Discord. And thank you so much for watching. A big thanks goes out to all of the supporters and members of my YouTube channel. You guys are really amazing, and of course, I'm going to catch you in the next video. Take care. Bye-bye.