Shit I've bought from eBay... #3 - HP ProLiant DL380 G5 Rack Mount Server

[Novaer]

Have you ever seen a server before?
Well if not; you’re about to get a really good look at one.

I was just browsing around on eBay when this caught my eye… A HP ProLiant DL380 G5 Server; £21.00 with just 3 hours left until the end of the bidding. My first thought was laughter; “Alright? What’s wrong with it?” I asked myself. So I looked into the listings where it said what the specs were plus it had a full complement of hard drives and RAM (Kind of), it was all working and had Windows Server 2016 installed on it (Unlicenced). It was advertised as having light cosmetic scratches along the top but otherwise it was working perfectly and good as new. Now; I had a spare licence for Windows Server 2016 kicking around so activating it wasn’t going to be a problem for me. I am passionate about computers, and tech, and this is something I’ve never had my hands on before, and I wanted to have my hands on it, to open it up and see what it looked like on the inside, and that passion spurred me to share my experience publically… Even if no one reads or cares about this. But there was also another reason for this...

I was in need for a server to host a private Intranet site on, and/or a NAS (Network Attached Storage) and I figured, twenty odd quid for all of that? Fuck yes. The price of a NAS is usually hundreds of pounds or more, and a server can be in the thousands of pounds (GBP, £. Not weight.) And so I bidded, to test the waters I put in a bid for £22, only to see that the person I bidded against was still the highest bidder - because he actually put a max bid amount in. So I figured ‘Okay… I’ll come back to this.’ And come back to it I did. 

At 17:00, about 15 minutes before the bidding ended, it was up to £25, and I figured I’d get it near enough to the last minute, or at least make the guy I was trying to outbid pay more. The maximum I was willing to pay was £30; that plus the £15 would have made it £45 all told and on the 17 Oct 2018 at 17:00, I put in the bid. 

At £26.00 I was the highest bidder, and now I just had to wait for 15 minutes to see if the guy was going to try and out bid me… But it never came. I expected him to bid snipe me at the very last minute; which is the act of putting in a bid at literally the last second of an auction. It was exciting. In fact I watched the count down, and when it reached zero, I refreshed to be sure and got the great green banner “You won this auction.” 

I grinned from ear to ear, although that was short lived as I figured what if the seller didn’t want to let it go for that price, and expected more money for it? Well… That never happened, I got a message through the eBay messenger system saying that (Long story short) there wasn’t a tracking number, nor would there be a courier delivery because of their corporation’s members of staff lives (Literally) across the road from me, and he was going to bring it here on Monday evening on his way home. 

I am not going to lie; as I said to @Aura in the Discord it was “so fucking sketchy you could make an art book and start a Patreon on it.” But I decided to trust them, because this wasn’t just some random no body. This was a serious corporation, selling their old equipment on eBay.

Item: HP ProLiant DL380 G5 2U Server
Price I paid: £26.00
Shipping: £15.00 Delivered by one of the Corporation’s Member of Staff
Total Cost: £41.00
Came With: Rack Mount Brackets

When this server was handed to me at my front door I was taken back by how god damn heavy it was. My computer that I have featured in this topic weighs a tonne but it’s because of the bulky case it’s in. This server didn’t look like much but my goodness it weighs more than my computer, at least, and the weight is quite unbalanced. But once I got used to it, I carried it back into the living room and saw the guy who delivered it off, I think I can carry this thing quite comfortably now I know where the balances are. I’ve since had it on a set of scales and it weighs around 24.2KG ± 0.5KG (~53.35lbs ± ~1.1lbs)

Did I plug it in and start playing around with it first? Hell fucking no. In true David Jones (EEV Blogs) style of “Don’t just turn it on, take it apart!” I opened that bad boy up and had a look inside.

This picture will most certainly be relievent.

I was instantly impressed (Considering the age of server) by how plug and play all of the components were. The fans have a red tab that you can literally pull out of the the sockets, in fact you can see three sockets behind the four fans in the upper left corner of the image, with another socket at the bottom of the three that are opened up, covered with a cap, and you can see two capped sockets behind the fans at the bottom. I’d assume that this server can have 12 intake fans, max. But these fan’s are ridiculously loud; as I will show in a video later. When the server first turns on; it is between 70dB and 80dB, according to the sound meter app on my phone (So take it with a grain of salt). You can hear for yourself by subjectiveness how loud they are with this video I recorded on the 4th of November: 

 

Once the system quite calms down, assuming that nothing goes under such insane load that it starts heating up to the point that the fans need to go on full whack again, it’s around 55dB which is something I didn’t think about when I bidded for this. This means that my plans on using it as an internal private internet server, or NAS wasn’t feasible; because there’s no where, where we can stick it with reasonable access to the router that wouldn’t be a pain in the arse. 

Specs:

• CPU: 2x Intel Xeon E5335 
  • LGA 771, 65nm Chip
  • X86-64 Architecture
  • Quad Core @ 2.00GHz (4 Threads)
  • 4x 32KB L1 Cache
    • Single Bit ECC
  • 2MB L2 Cache per Core (8MB total L2 Cache)
    • Single Bit ECC
  • 65W Typical Power Consumption (80W TDP)
  • 64GB Max Supportable RAM*
• Chipset: Intel 5000P rev. B1 (North Bridge) + Intel 6321ESB rev. 09 (South Bridge)
• RAM: 8GB DDR2 (8 x 1GB)
  • 2x Elpida (1024MB each) 
    • ECC RAM
    • Bandwidth: PC2-5300 (333 MHz)
    • Timing: 5.0-5-5-15-20 @ 333 MHz (CL-tRCD-tRP-tRAS-tRC @ frequency)
    • Model: EBE11FD8AJFT-6E-E
  • 2x Micron Technology (1024MB each)
    • ECC RAM
    • Bandwidth: PC2-5300 (333 MHz)
    • Timing: 5.0-5-5-15-19 @ 333 MHz (CL-tRCD-tRP-tRAS-tRC @ frequency)
    • Model: MT18HTF-12872FDY-667D6D4
  • 2x Hyundai Electronics (1024MB each)
    • ECC RAM
    • Bandwidth: PC2-5300 (333 MHz)
    • Timing: 5.0-5-5-15-20 @ 333 MHz (CL-tRCD-tRP-tRAS-tRC @ frequency)
    • Model: HYMP512F72CP8N3-Y5 AC-C
  • 2x Samsung (1024MB each)
    • ECC RAM
    • Bandwidth: PC2-5300 (333 MHz)
    • Timing: 5.0-5-5-15-20 @ 333 MHz (CL-tRCD-tRP-tRAS-tRC @ frequency)
    • Model: M395T2953EZ4-CE65
• Optical Media: HL-DT-ST RW/DVD GCC-C10N
  • Firmware: 2.0
  • Can read:
    • CD-R
    • DVD-R
    • DVD-RW
    • DVD+R
    • DVD+RW
    • DVD-RAM
    • DVD+R DL
  • Can Write:
    • CD-R
• Network Cards: 2x QLogic BCM5708C Gigabit Ethernet
• Graphics: ATI ES1000 (Integrated)
  • VGA Output only.
  • 32MB SDRAM Video Memory
• PSU: 2x 800w Power Supplies (Redundancy) 
  • Model DPS-800GB A (Rev: 06M)
  • Model ATSN 7001044-Y000 (Rev: C)
• HDD: 8x 72GB 2.5” SAS Hard Drives in RAID5 Configuration (~504GB Usable Space)
  • HP Invent: 72GB, 10,000 RPM (DG072A9BB7)
  • HP Invent: 72GB, 10,000 RPM (DG072A8B54)
  • HP Invent: 72GB, 10,000 RPM (DG072A9BB7)
  • HP Invent: 72GB, 10,000 RPM (DG072A9BB7)
  • HP Invent: 72GB, 10,000 RPM (DG072A3515)
  • HP Invent: 72GB, 15,000 RPM (DG072BB978)
  • HP Invent: 72GB, 15,000 RPM (EH0072FAWJA)
  • HP Invent: 72GB, 15,000 RPM (DN072ABAA6)

* There is some conflicting information about this. Whilst the CPUs claims to be able to support a maximum physical address size of 64GB, HP claims that the server can only support 32GB. This could indeed be the case via a BIOS limitation or something, I don’t know. 

The next thing I did was took a look at the hard drives to see if there was remnants of data left over on them from the previous owners - there wasn’t, it was clean, which means that they did full and proper wipe of the hard drives. So, thumbs up for that. 

I can’t tell you the amount of times I’ve purchased second hand hard drives online, that people haven’t properly formatted the hard drives of. I’ve had access to various kinds of sensitive materials that have either been too hot to handle or have been wiped properly by myself. But the stuff I will talk about openly is that I’ve also had access to a company’s customer database which included names, addresses, phone numbers and credit card information of their customers. But most commonly I’ve had access to people's private and personal files, and in one case I got access to someone’s private porn library. The hard drive was used to store their porn on it… I don’t personally think they had good tastes, but it was rather tame and vanilla by my standards… In one case I got a hard drive from a friend and it had lots of pictures on it that they lost; so I recovered them for her, and sent the pictures back to her on a flash drive so that she could actually have them back. She was extremely thankful and didn’t know how I did it, I tried explaining it to her, but I don’t think she understood. She just understood that I knew what I was doing and she got her pictures back.

So yes… To see someone actually take data protection seriously and to wipe their drives before selling them is a good thing to me. I actually like that they do it properly. But I always format again; just to be sure. That and you don’t know who could tamper with the OS you’re working on. If a computer comes installed with an OS from eBay; I never keep that OS on, I always uninstall, format, and then install before I start actually using it. It does seem more and more every day that people are actually starting to do this. Gone are the days when I always got hard drives that weren’t formatted properly and had private information on them or something left over from the previous owner. The next two subjects of things I’ve bought from eBay were also cleaned properly. But I will get to them in their own topics.

One of the first things I did after I took a look inside, took some photos and checked the hard drives is that I, wanted to test the internal temperatures of the server to see if I could work out as to why it was so loud; thinking that it was loud because it was overheating, which turns out that I was quite correct. Although something popped up that I never seen before; the RAM was doing far worse than the CPUs were. 

At idle the CPUs was around: 59-65oC depending on core and CPU
- At idle, the RAM on the other hand was around: 50oC at the coolest and 78oC at the hottest.

Typically when I see RAM that is far beyond a normal temperature that I expect; I expect the RAM to operating under overclocking, or some kind of overvoltage than the spec allows. It is something I will investigate later in this post. I was also kind of concerned about the high CPU temperature under no load other than Windows Server 2016 running, all updates were complete and the hard drives weren’t really doing much so I tested it under synthetic load (Although Prime95 didn’t work at all) I used a different program to simulate a heavy load (Although I don’t remember what it was called). Whilst the CPU usage skyrocketed to 100% across both CPU’s the Temperatures didn’t really move all that much, about 5 degrees centigrade, that’s pretty much it. 

Now for all I know this could all indeed be absolutely normal for a Server and its RAM/CPU. But I shut down the server as I happened to have some Arctic Silver 5 compound laying around from when I did my CPU in my computer and some solvents designed to remove old thermal compound. So I removed the cover and heatsink block from the server:

As a side note; I’ve never liked the application or use of Liquid Metal cooling in systems, where you have to take the heat spreader off just to apply the liquid metal. There are a tonne of risks and very few rewards from what I heard. Don’t get me wrong; I’ve heard the kind of results you can get with using a little bit of Liquid Metal on top of your CPU under it’s heatspreader, but in my experience Liquid Metal only comes in two forms for cooling: Mercury and Gallium. Mercury is extremely toxic, and Gallium eats and deteriorates the structural strength of Aluminium and Steel; both mercury and gallium also conduct electricity and if either of the two liquids were to leak onto the electronics, it could cause a short circuit which is bad, and that’s not even going into the whole “Mercury and Gallium” expands under heat; like the heat of a CPU. Now this is fine for some people, as I said I’m not knocking it. But I don’t take that kind of risk, apparently whilst it’s amazing in heat transference, it’s not recommended for absolutely everyone and I certainly wouldn’t use it myself.

So I used Arctic Silver 5; which is a high density micronise Silver and ‘Enhanced Thermally Conductive Ceramic Particles’ (What I call Enhanced TCCP), which is typically around 88% thermally conductive materials (by weight) with ‘Three Unique Shapes and Sizes’ of 99.9% pure silver to maximise the particle contact area and thermal transfer, suspended in Polysynthetic Oils. It is not electrically conductive either. You can find more information here (Non-Affiliate Link). This is not an advert; and I am not sponsored by Arctic Silver, but it is in my experience, the best stuff to use for thermal paste solutions, at the very least it is almost usually much better than stock solutions. It is not cheap though (~£1,800 per KG. You buy it usually in 3.5g syringes that cost around £6-7 and it does lots of applications). 

When I took the heatsinks off; I noticed that the told thermal paste solution - most probably stock, was dry and there was way too damn much of it. So I took the chips out of their Sockets and placed them on my anti-static mat before getting to work on cleaning them up, alongside their respective heatsinks, which was absolutely caked in thermal compound. 

LGA771 (Socket J) CPU sockets

I then put the CPU’s back into their sockets carefully, and the result? Nice clean CPU’s ready for new thermal compounds:

After this operation took place; at idle the CPUs was around: 31-48oC depending on CPU and core. One CPU was running slightly hotter than the other, but the first and primary CPU was running pretty much cold. Which meant that I was pretty much successful in changing the thermal compound, although I hadn’t quite expected such drastic results. These results carried through to under load; where the temperature of the first CPU didn’t even hit 50oC and the second just went over 66oC but it made absolutely no difference to the noise of the server. It was just as obnoxiously loud as ever.

Pleased with the results; I considered that job done. My next job was to remove the OS that the corporation I purchased this from put on there, and completely format/wipe the hard drives. Which I did; and surprisingly with this RAID5 setup; it took quite a bit of time to fully format the drives.

I benchmarked the system; including the hard drives but before I get into the results there's something I kind of want to talk about that’s been bugging me for a while. It’s a small rant so it’s hidden in the spoiler below. If you want to read it, click on it to read it. Otherwise we can move right on ahead.

Spoiler

 

There has been some revisionism going on in how size of objects is defined; compared to what it was when I grew up with it, what I was taught, and what I know to be true. Nowadays people will try to tell me that:

• 8 bits = 1 byte, then magically:
• 1000 bytes = 1KB (Kilobyte)
• 1000KB = 1MB (Megabyte)
• 1000MB = 1GB (Gigabyte)
• 1000GB = 1TB (Terabyte)

And so on; whereas:

• 8 bits = 1 byte
• 1024 bytes = 1KiB (Kibibyte)
• 1024KiB = 1MiB (Mebibyte)
• 1024MiB = 1GiB (Gibibyte)
• 1024GiB = 1TiB (Tebibyte)

I absolutely reject this fervidly, to me it feels like dumbing something down for adults that haven’t quite finished being children yet. They make the argument that in normal scientific notation a Kilo, Mega, Giga, and so on are 1000’s, which; okay I understand that. BUT computers are not base ten and have never been Base 10 - except it fucking rare circumstances. They say that Hard Drives use this formula; okay, but RAM doesn’t, nor to bits in a processor architecture, nor does bits in a colour pallet, nor does ROM; my Basic Amstrad CPC464 had 64K (65,536 Bytes) and my RAM in my main computer is 16GB (16,384MB - 16,777,216KB or 17,179,869,184 Bytes) not 16,000,000,000 bytes. 

Computers do not work in Base 10; they work in Base 2. 

Hence: 

• 8-bit
• 16-bit
• 24-bit (In the case of colour and sound) 
• 32-bit
• 64-bit
• 128-bit
• 256-bit 

… and so on. 

Hell, even the NTFS file system nowerdays writes sectors as 4096 byte sectors. Why? Because it’s Base-2 not Base-10. 

In Binary the is literally only 1’s and 0’s. Two Values. Not 10 values. 

In Hex (Base-16, which Base-2 is compatible with, where as Base-10 isn’t) there are 16 values, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, A, B, C, D, E and F, not 20 values, where each Hex Value equals 4 bits, or when grouped together 1 byte (Example: 0F = 0000 1111).

Encryption as a standard use base-2 as well such as 128-bit, 256-bit, 512-bit, 768-bit, 1024-bit, 2048-bit, 4096-bit, 8192-bit and so on.

Processor Instruction Sets/ALUs uses Base-2 as well, such as 8-bit, 16-bit, 32-bit, (Some I think use 48-bit) 64-bit, and in some rare circumstances/theoretical but not implemented, 128-bit. 

Therefore to me: 

• 8 bits = 1 byte
• 1024 bytes = 1KB
• 1024KB = 1MB
• 1024MB = 1GB
• 1024GB = 1TB 

and so on… Stop telling me that your shit cake is a chocolate fudge brownie. It’s a shit cake.

That is what I grew up with and that’s what I know. I really dislike those that have formed an elitist tribe over this, where if you don’t agree with it, you will be shamed and chastised into submission; just like the definition of Sexism, Racism and Fascism has been morphed nowerdays. It feels post-modernistic. If you disagree with me; then fine, I disagree with you. We can talk about it like adults if you like, providing you don’t turn into a belittling, condescending little shit about it. I will spare you the same courtesy. I am not above having my mind changed on the subject; I just want some solid and damn good explanations, and so far everyone I’ve spoken to has not given me one, and everything I’ve read has not given me one either. To make matters worse people use MB and MiB interchangeably, so both actually mean both to some people, which is not good either.

This trend in society of dumbing everything down to the lowest common denominator and the fetish for re-defining words and meanings for things that have already been defined perfectly for decades really pisses me off. I know things change all the time, but if you want to make it base 10; then you can go and invent your own words for it. I’ve spent a lot of time and energy training my mind to be able to calculate in Base-2, Base-16 and to me it comes naturally, so to have the definitions changed like this doesn’t sit well with me at all.

That said… 

 

Unfortunately the benchmark program CrystalDiskMark 6.0.2 defines the following benchmarks as the stupid 1KB = 1000 bytes therefore 1000KB = 1MB, and if you want it to mean 1024KB = 1MB you’re actually talking about MiB… Bullshit. 

Whilst the test performed (in numbers) better than my next ‘Shit I’ve bought from eBay’ topic, the creation of test file was ridiculously slow. I was waiting just over an hour for it to create the 32’GiB’ file. But the results of the benchmark were extremely interesting. Now bear in mind I’m running RAID 5 (You can find more information on RAID 0, RAID 1 and RAID 10 here and for more information on RAID 5 and RAID 6 you can click here.) across 8 drives; idealistically that gives me 7 drives and one drive’s worth of parity, which means I can withstand one of the hard drives failing completely, and only have the data capacity of 7 out of 8 of the drives. 

So here is the raw data of the test using Crystal Diskmark (Freeware):

I also used SiSoftware (Sandra) to get an overall benchmark of the system as I hold a valid licence for the software (2016 version). Which I have compared to my desktop computer (Because as unfair comparaisons go this was the best baseline I have; other than old Pentium 2 Processor machine I have in my airing cupboard). 

I think to understand the benchmarks we should probably see how the benchmark software sees the server because the one thing I noticed when running various things is that the various tools saw things differently. For example CPUID saw there being two E5335 (Clovertown) processors where as HWiNFO64 v5.90 saw one of them as being an L5335 (Clovertown LV); which is not the case - because I’ve had both processors out and I know what both of them said.

So to keep this brief; this is how SiSoftware Sandra saw the server:

Computer
Model : HP ProLiant DL380 G5
Serial Number : (REDACTED)
Chassis : HP Rack Mount Chassis
BIOS : HP P56 08/16/2015
Total Memory : 8GB DDR2 FB-DIMM

Processors
Processor : Intel(R) Xeon(R) CPU E5335  @ 2.00GHz (4C 2GHz, 2x 4MB L2)
Processor : Intel(R) Xeon(R) CPU E5335  @ 2.00GHz (4C 2GHz, 2x 4MB L2)
Socket/Slot : FC-LGA6

Chipset
Memory Controller : HP 5000P Chipset Memory Controller Hub 4x 333MHz (1.33GHz), 4x 2GB DDR2 FB-DIMM 666MHz 256-bit

Memory Module(s)
Memory Module : Elpida EBE11FD8AJFT-6E-E 1GB ECC DDR2 FB-DIMM PC2-5300FB DDR2-666 (5-5-5-15 3-20-5-3)
Memory Module : Elpida EBE11FD8AJFT-6E-E 1GB ECC DDR2 FB-DIMM PC2-5300FB DDR2-666 (5-5-5-15 3-20-5-3)
Memory Module : Micron 18HF12872FD667D6D4 1GB ECC DDR2 FB-DIMM PC2-5300FB DDR2-666 (5-5-5-15 3-18-5-3)
Memory Module : Micron 18HF12872FD667D6D4 1GB ECC DDR2 FB-DIMM PC2-5300FB DDR2-666 (5-5-5-15 3-18-5-3)
Memory Module : Hynix (Hyundai) HYMP512F72CP8N3-Y5 1GB ECC DDR2 FB-DIMM PC2-5300FB DDR2-666 (5-5-5-15 3-20-5-3)
Memory Module : Hynix (Hyundai) HYMP512F72CP8N3-Y5 1GB ECC DDR2 FB-DIMM PC2-5300FB DDR2-666 (5-5-5-15 3-20-5-3)
Memory Module : Samsung M395T2953EZ4-CE65 1GB ECC DDR2 FB-DIMM PC2-5300FB DDR2-666 (5-5-5-15 3-20-5-3)
Memory Module : Samsung M395T2953EZ4-CE66 1GB ECC DDR2 FB-DIMM PC2-5300FB DDR2-666 (5-5-5-15 3-20-5-3)

Video System
Monitor/Panel : (Standard monitor types) Generic Non-PnP Monitor (1600x1200)
Video Adapter : Microsoft Basic Render Driver (SM5.2, 4GB)

Graphics Processor

Storage Devices
HP LOGICAL VOLUME (513.6GB, RAID, SCSI-5, 15000rpm) : 478GB (C:)
HL-DT-STCD-RW/DVD DRIVE GCC-C10N (ATA33, DVD+-R, CD-RW, 2MB Cache) : N/A (D:)

Logical Storage Devices
System Reserved : 500MB (NTFS, 4kB) @ HP LOGICAL VOLUME (513.6GB, RAID, SCSI-5, 15000rpm)
Hard Disk (C:) : 478GB (NTFS, 4kB) @ HP LOGICAL VOLUME (513.6GB, RAID, SCSI-5, 15000rpm)
Optical Drive (D:) : N/A @ HL-DT-STCD-RW/DVD DRIVE GCC-C10N (ATA33, DVD+-R, CD-RW, 2MB Cache)

Peripherals
LPC Hub Controller 1 : Intel 631xESB/6321ESB/3100 LPC Interface Controller
LPC Legacy Controller 1 : SMSC SCH4307
Serial Port(s) : 1
Disk Controller : HP ProLiant DL140 G3
Disk Controller : HP Smart Array P400 Controller
USB Controller 1 : HP ProLiant DL140 G3
USB Controller 2 : HP ProLiant DL140 G3
USB Controller 3 : HP ProLiant DL140 G3
USB Controller 4 : HP 631xESB/6321ESB/3100 Chipset USB Universal Host Controller #4
USB Controller 5 : HP ProLiant DL140 G3
USB Controller 6 : HP iLO2
SMBus/i2c Controller 1 : Intel 5X00/7X00/8X00 XMB Ch 21-0 SMBus
SMBus/i2c Controller 2 : Intel 5X00/7X00/8X00 XMB Ch 21-1 SMBus
SMBus/i2c Controller 3 : Intel 5X00/7X00/8X00 XMB Ch 22-0 SMBus
SMBus/i2c Controller 4 : Intel 5X00/7X00/8X00 XMB Ch 22-1 SMBus
SMBus/i2c Controller 5 : Intel ICH SMBus
SMBus/i2c Controller 6 : IPMI T1 Controller

Printers and Faxes
Printer : Microsoft XPS Document Writer v4 (600x600, Colour)
Printer : Microsoft Shared Fax Driver (200x200)
Printer : Microsoft Print To PDF (600x600, Colour)
Printer : Microsoft XPS Document Writer v4 (600x600, Colour)
Printer : Microsoft Print To PDF (600x600, Colour)

Network Services
Network Adapter : QLogic BCM5708C Gigabit Ethernet (NDIS VBD Client) #42 (Ethernet)
Network Adapter : QLogic BCM5708C Gigabit Ethernet (NDIS VBD Client) #41 (Ethernet, 1Gbps)

Operating System
Windows System : Microsoft Windows 2016 Server 10.00.14393
Platform Compliance : x64

It’s a good idea to understand how exactly the software see’s both machines; and in the next part I will compare this server to the device there too, which is a more realistic comparaison. 

And the raw data:

Processor Multi-Media
Analysing...
Aggregated Score : 99.76MPix/s
Result ID : 2x Intel(R) Xeon(R) CPU           E5335  @ 2.00GHz (4C 2GHz, 2x 4MB L2)
Speed : 2000MHz
Capacity : 8Unit(s)
Finished Successfully : Yes

Cryptography
Analysing...
Aggregated Score : 0.693GB/s
Result ID : 2x Intel(R) Xeon(R) CPU           E5335  @ 2.00GHz (4C 2GHz, 2x 4MB L2)
Speed : 2000MHz
Capacity : 8Unit(s)
Finished Successfully : Yes

Processor Financial Analysis
Analysing...
Aggregated Score : 7.91kOPT/s
Result ID : 2x Intel(R) Xeon(R) CPU           E5335  @ 2.00GHz (4C 2GHz, 2x 4MB L2)
Speed : 2000MHz
Capacity : 8Unit(s)
Finished Successfully : Yes

Processor Scientific Analysis
Analysing...
Aggregated Score : 7.19GFLOPS
Result ID : 2x Intel(R) Xeon(R) CPU           E5335  @ 2.00GHz (4C 2GHz, 2x 4MB L2)
Speed : 2000MHz
Capacity : 8Unit(s)
Finished Successfully : Yes

.NET Arithmetic
Analysing...
Aggregated Score : 15.88GOPS
Result ID : 2x Intel(R) Xeon(R) CPU           E5335  @ 2.00GHz (4C 2GHz, 2x 4MB L2)
Speed : 2000MHz
Capacity : 8Unit(s)
Finished Successfully : Yes

Memory Bandwidth
Analysing...
Aggregated Score : 5.350GB/s
Result ID : Intel 5000P Chipset Memory Controller Hub; 4x 2GB Elpida EBE11FD8AJFT-6E-E DDR2 FB-DIMM (666MHz 256-bit) PC2-5300 (5-5-5-15 0-20-5-9)
Speed : 666MHz
Capacity : 8192MB
Power : 108.08W
Finished Successfully : Yes

Cache & Memory Latency
Analysing...
Aggregated Score : 61.9ns
Result ID : Intel 5000P Chipset Memory Controller Hub; 4x 2GB Elpida EBE11FD8AJFT-6E-E DDR2 FB-DIMM (666MHz 256-bit) PC2-5300 (5-5-5-15 0-20-5-9)
Speed : 666MHz
Capacity : 8192MB
Power : 108.08W
Finished Successfully : Yes

File System Bandwidth
Analysing...
Aggregated Score : 244.494MB/s
Result ID : HP LOGICAL VOLUME (513.6GB, RAID, SCSI-5, 15000rpm) (NTFS, 4kB)
Speed : 15000rpm
Capacity : 513.62GB
Finished Successfully : Yes

File System I/O
Analysing...
Aggregated Score : 1417.0IOPS
Result ID : HP LOGICAL VOLUME (513.6GB, RAID, SCSI-5, 15000rpm) (NTFS, 4kB)
Speed : 2560Mbps
Capacity : 513.62GB
Finished Successfully : Yes

GP (GPU/CPU/APU) Processing
Analysing...
Error (339) : No devices found. : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Finished Successfully : No

Video Shader Compute
Analysing...
Aggregated Score : 25.03MPix/s
Result ID : Microsoft Basic Render Driver (SM5.2, 4GB) (D3D 11)
Finished Successfully : Yes

GP (GPU/CPU/APU) Cryptography
Analysing...
Error (339) : No devices found. : High Security (AES256 + SHA2-256) : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Standard Security (AES128 + SHA1) : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Higher Security (AES256 + SHA2-512) : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Finished Successfully : No

Cryptography
Analysing...
Aggregated Score : 0.478GB/s
Result ID : 2x Intel(R) Xeon(R) CPU           E5335  @ 2.00GHz (4C 2GHz, 2x 4MB L2)
Speed : 2000MHz
Capacity : 8Unit(s)
Finished Successfully : Yes

GP (GPU/CPU/APU) Financial Analysis
Analysing...
Finished Successfully : No

Processor Financial Analysis
Analysing...
Aggregated Score : 7.88kOPT/s
Result ID : 2x Intel(R) Xeon(R) CPU           E5335  @ 2.00GHz (4C 2GHz, 2x 4MB L2)
Speed : 2000MHz
Capacity : 8Unit(s)
Finished Successfully : Yes

GP (GPU/CPU/APU) Scientific Analysis
Analysing...
Finished Successfully : No

Processor Scientific Analysis
Analysing...
Aggregated Score : 7.22GFLOPS
Result ID : 2x Intel(R) Xeon(R) CPU           E5335  @ 2.00GHz (4C 2GHz, 2x 4MB L2)
Speed : 2000MHz
Capacity : 8Unit(s)
Finished Successfully : Yes

Media (Audio/Video) Transcode
Analysing...
Aggregated Score : 1.587MB/s
Result ID : 2x Intel(R) Xeon(R) CPU           E5335  @ 2.00GHz (4C 2GHz, 2x 4MB L2) (Microsoft H264 Video Decoder MFT > H264 Encoder MFT; Microsoft AAC Audio Decoder MFT > Microsoft AAC Audio Encoder MFT)
Speed : 2000MHz
Capacity : 8Unit(s)
Finished Successfully : Yes

GP (GPU/CPU/APU) Bandwidth
Analysing...
Error (339) : No devices found. : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Finished Successfully : No

Video Memory Bandwidth
Analysing...
Aggregated Score : 2.485GB/s
Result ID : Microsoft Basic Render Driver (SM5.2, 4GB) (D3D 11)
Capacity : 4095MB
Finished Successfully : Yes

GP (GPU/CPU/APU) Memory Latency
Analysing...
Error (339) : No devices found. : Global Data Memory : In-Page Random Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Global Data Memory : Full Random Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Global Data Memory : Sequential Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Constant Data Memory : In-Page Random Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Constant Data Memory : Full Random Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Constant Data Memory : Sequential Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Shared Data Memory : In-Page Random Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Shared Data Memory : Full Random Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Shared Data Memory : Sequential Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Private Data Memory : In-Page Random Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Private Data Memory : Full Random Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Private Data Memory : Sequential Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Texture Memory : In-Page Random Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Texture Memory : Full Random Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Error (339) : No devices found. : Texture Memory : Sequential Access Pattern : GP(GPU) call failed. Try another interface (e.g. OpenCL/ComputeShader/CUDA/etc.) or update video drivers.
Finished Successfully : No

Cache & Memory Latency
Analysing...
Aggregated Score : 61.8ns
Result ID : Intel 5000P Chipset Memory Controller Hub; 4x 2GB Elpida EBE11FD8AJFT-6E-E DDR2 FB-DIMM (666MHz 256-bit) PC2-5300 (5-5-5-15 0-20-5-9)
Speed : 666MHz
Capacity : 8192MB
Power : 108.08W
Finished Successfully : Yes

Overall Score
Aggregated Score : 2.22kPT
Results Interpretation : Higher Scores mean Better Performance.
Decimal Numeral System (base 10) : 1GPT = 1000MPT, 1MPT = 1000kPT, 1kPT = 1000PT, etc.
Result ID : HP ProLiant DL380 G5 ProLiant (2x Intel Xeon CPU E5335 @ 2.00GHz; Intel 5000P Chipset Memory Controller Hub; 4x 2GB Elpida EBE11; HP LOGICAL VOLUME; Microsoft Basic Render Driver)
Finished Successfully : Yes

And now I will finish the topic with some comments I have about the machine as well as some neat pictures of the server itself.

First of all; I absolutely love how intermodular and hot swappable all of the parts are, and I think for at the time it may have cost around £3,000 - £4,000 on launch, which I think was actually in 2011. It has several redundant systems to keep the machine as reliable as possible, and it goes to show. Back in 2011 I was just 21 years old, and I seriously doubt I was the target market for the server, but that doesn’t stop me appreciating the build quality, or it’s fantastic engineering even here in 2018. I really can’t fault this machine at all, for what it was in 2011, I think almost everything in the system is modular, in that without screws you can take parts out and easily swap them which in critical deployments this is probably a very good idea. With two 800W PSU’s (Power Supply Units) the system can keep running even if one of them just dies, because it’s redundant. In front of these PSU’s are also two VRMs (Voltage Regulator Modules) that have massive heat sinks on them. I’ve not been inside many servers (Funnily enough) and this could very well be standard with other brand servers. 

This is what the VRM’s look like, right next to the PSU’s. The VRM’s are basically a buck converter that regulates the power for the CPUs. For more information on what VRMs are used for, here’s a good starting point.

On the front of the machine is eight bays of (Apparently) hot-swappable HDD bays that hold some of the fattest 2.5” HDD’s I’ve ever seen (Which means this might actually be upgradable to SSDs if the firmware allows). Each having their own mechanisms to both eject, and keep the hard drives securely into the unit. I believe they use 2.5” HDD’s to allow it to keep the 2U server rack form factor. You don’t get much wiggle room under RAID 5, and I was a little surprised that RAID 6 wasn’t even an option under the RAID Controller setup. Is this a problem persae? Probably not. But RAID 5 only allows 1 Failure, whereas RAID 6 will allow upto 2 Failures. But to be honest I think the hard drives seem to be extremely solidly built and there doesn’t seem to be any problems with any of them, and given the fact that the target audience would most likely have onsite or even offsite backup facilities to cope with catastrophic data loss. And RAID 5 would give the fastest read speeds as well as the most capacity, so maybe it makes sense from the potential target audiences point of view. Another issue I have with the hard drive array is a pill I just wasn’t able to quite swallow, it seems to have inconsistent read/write speeds… At the time I didn’t know if this is the consequence of having a RAID 5 set up, or if it’s an indicator that one of the drives is a little slower than other drives and it’s slowing the rest down or if there is something not quite set up properly. But despite the benchmarks and what they said; when it was writing test files for CrystalDiskMark it was writing at around 8MB/s according to Resource Monitor. Which how the benchmark was able to achieve 16-odd Megabytes per second in the write tests is beyond me. Despite it having better scores than the next item on my Shit I’ve bought from eBay series, the next item wrote this test file much faster than this server did. This bugs me, because by the numbers this HP server should have knocked it out of the park… But it didn’t. To make matters worse, is in the tests that I did by sending something over the ethernet port to my SSD in my desktop and receiving was quite abysmal speeds too that did not remain consistent in the 10GB test file I transferred. Then again this could also be down to the RAID Array Controller itself as well.

Top: Server HDD, Bottom: Normal Laptop HDD. Very THICC. 

The RAM is DDR2, which I think was about expected for the time, in addition it’s ECC RAM which you can find out more information about here from Luke Lafreniere over on Techquickie. The total in the system seems to be eight sticks of 1GB totalling 8GB, but it can be taken higher than that with official HP spec saying that the max supported is 32GB although the processors on the Intel page report that they can support upto 64GB. I don’t have eight - eight gigabyte sticks of ECC DDR2 RAM so I can’t check. To compound matters the Memory controller (Intel 5000P Memory Controller) also supports upto 64GB too, so my only guess is that it’s not supported in the BIOS, or perhaps it can support 64GB - If it’s not ECC, I don’t know, I don’t have eight sticks of non-ECC DDR2 RAM either. This won’t come as any surprise to @Aura who has been public in his criticism for HP Laptops for overheating (Something I completely agree with, all HP netbooks and laptops I’ve ever had have all died due to overheating) but even under idle loads the RAM is surprisingly hot, so hot in fact, that when I turned the server off, and discharged it after heavy benchmarking, I actually burnt my hand slightly when it grazed the RAM. Under idle the heat that the RAM can get to is around 70oC to 80oC. My first thought was that maybe it was being overclocked, and my second was, was that maybe it was way over voltage. Neither of these things were the case - apparently. Not that the BIOS allowed me to look into it. 

The two CPU’s however is the opposite story, since re-thermal pasting it, even under intensive load generated by the benchmarking software I couldn’t penetrate 60oC on the first processor and I couldn’t penetrate 70oC on the second one, which considering that these are both quad core Xeon’s running at full clock (2.00GHz) and aren’t being cooled by directly mounted fans to the heatsink, this is pretty good. And considering the RAM seems to get hot enough that I think I could fry and egg on it, yeah that’s good. 

There are two VGA ports, one on the back, one on the front that will allow the user to connect them up to a VGA compatible monitor which was a problem for me. My monitor has Display Port and HDMI only, but Neptune has a second monitor that has both VGA and HDMI on it, so I had to borrow her monitor - repeatedly. This is somewhat to be expected, I’m not sure if new servers today have HDMI (I’d think that they do) but servers priorities in most cases isn’t on giving you the best graphics for your money, and a VGA port more than sufficed for what I needed it for, because once the OS was set up I was able to control the system via Remote Desktop at full 4K glory. Although don’t count on it running a Minecraft Client, or Crysis. 

The graphics through the VGA ports is pretty much powered by this ATI ES1000 Graphics Chip, which is hidden under the PSU Cage. I think these have been used in old laptops, and you can actually buy these chips on eBay which apparently are sold as “Laptop GPUs”

It has three Ethernet Ports - one of them is for HP ILO2 (Integrated Lights-Out 2), a kind of web interface that allows you access the server when it’s offline (But there’s still power connected to it). 

It says that the system health is degraded because there wasn’t a power supply connected to the second bay. But you can reboot, turn on and turn off the server from here and use all kinds of interesting tools and insight into the server. Something I actually found to be really fascinating to say the very least. There was one major drawback to this however - no modern browser can use it. You see it uses TLS 1.0 which is obsolete and disabled in most modern browsers nowadays (For damn good reason too!), and there’s no way that I can tell to upgrade it to TLS 1.2. Google Chrome would not let me switch to TLS 1.0, and so I had to download and install Firefox, and manually enable it, whilst disabling the other TLS options. Even then I couldn’t use the Virtual Media, or Remote Consoles, as it needed to be used under Internet Explore with the appropriate Java Runtime Environment enabled… 

Yeah it’s more trouble than it’s worth. Asking me to compromise my security wasn’t a bridge too far enough; that I apparently needed to use an old, old JRE… On Internet Explorer. Nothing is worth that trouble. Fuck Internet Explorer.

Unfortunately there is no upgrading iLO2, because it’s on a chip. You can do firmware updates to it but you can’t updated it to iLO3, or iLO4 or anything like that.

The other two Ethernet Ports are both Gigabit Ethernet ports which is good, because that’s the maximum my router supports. Apparently it supports TCP off-loading (Which you can find out more about here) I think it could also make a good in between connection from my Router to my Modem, which could allow me to do various things, if memory serves me correctly. But again we get to the noise and power consumption issue. 

There are also five USB Ports, that are most likely USB 2. They’re certainly not USB 3, there are two on the back, two on the front and one inside the unit itself. The reason I say that they’re most likely USB 2.0 is because HWInfo64 says it is; but then again it also says that one of the processors is a L5335… So take it with a pinch of salt, I have not done any USB speed tests due to the aforementioned problems with the hard drives that I’ve mentioned about already, I don’t have a reliable baseline. There are also a couple of PS/2 ports for an old style Keyboard and Mouse. Now luckily, I actually have an old PS/2 connector Keyboard, I do not have a spare USB Keyboard, but I do have a spare USB Mouse so I was able to use this no problems. There is also a serial port on the back of the server although I have no serial devices to test it with, I don’t even know if it’s still in use anymore, or if it was in 2011. 

As I said previously; the inside of the machine is amazing, and as you can see from the very first image repeated here:

It is very well designed; the system seems to take air in from the front, and in the case of the four fans towards the front of the machine it seems to direct as much airflow over the CPU and RAM as possible, with the bottom two pulling air through the hard drives and ejecting it through the PSU’s or out to the side vent holes. This is in an effort to keep them all cool, and on the most part - other than the RAM this is pretty successful… Although it’d beg the question of how hot the RAM would get if there wasn’t that kind of airflow. There are very few cables in this machine, and any cables that do exist are kept short and sweet, or are wrapped and kept as tidy and minimal as possible. There are two mesh wrapped cables containing lots of cables going from the PCI slot containing the RAID Controller to the Hard Drives. It’s clear a lot of thought and a lot of engineering went into the design of the system and the layout of components that would no doubt have to keep that in mind, but keep in mind also the distance between CPU, RAM and all the various components to keep it as fast and efficient as possible. 

This RAID Controller is probably the reason I was experiencing such inconsistencies between the benchmark and actual performance of the hard drives. Specifically the 256MB module you’re seeing (Which is an attachment to the RAID Controller) is a write cache, now it could be that there isn’t enough RAM to make it as efficient as possible - or… As my research has suggested, it’s possible that because this RAID controller isn’t hooked up to a battery (Which I will explain the purpose of in a moment) that it could be disabled entirely.

What is the purpose of the battery on a RAID Controller, and would not having the write cache enabled make the 256MB Module, redundant superfluous?

Well, the purpose of the battery on a RAID Controller is step for data protection in the event that something happens to the power of the server, you see the write cache gets stored on the RAM which is being written out to the hard drives to try and keep things running smoothly, especially if your server is reading from those drives at the same time. The problem is, is once it’s written to the Cache, the OS usually thinks it’s written to the hard drive.

From looking at the numbers on the chip (HYB18T512800AFL37) I can deduce that it’s DDR2 Volatile Memory rather than Flash so it can be rewritten to a number of times without degradation… As you know if you lose power, pretty much everything in the RAM is cleared out, where as with flash memory (USB Flash Drives, SSD’s, most BIOS’s, etc) don’t clear out on power down. There are tradeoffs for this, but generally speaking Volatile RAM is much faster, and can be overwritten to more times than Flash Memory. 

So, now that I’ve explained that, if something happened to the power of the server: Such as you’re only running it straight from the wall and not from a UPS (Uninterruptible Power Supply) when the power is cut or with a UPS or not, the power supply (or both of them) fail. Now if you have mission critical information stored on the write cache, and such a power outage happened, then it’s not written to the hard drive, and you’ll lose it all. So with all these redundancies and backups in place it’d make sense for it to be disabled if there is no battery attached to it. Does it make it superfluous, then? Well… Yes, kind of. In the research I looked up whilst writing this, I found that you can forcibly enable it, although where, I don’t know. I haven’t to my recollection seen that option in the BIOS or the RAID controller setup. 

Another thing that I thought was strange is when I was writing this and going through the pictures that I wanted to include, I noticed that there are five, 512 Megabit chips (64MB), which would equal 320MB of RAM on the cache module. The cache module is attached to the RAID Controller as a separate unit, so you can replace/upgrade the cache module without actually having to replace/upgrade the entire controller whilst you're at it. Take a look at the cache module on it’s own:

Although I hadn’t taken it off of the RAID card at this point, I could see 5 chips like I could before and three blanks, so it really got me thinking about the possibility that maybe the 256MB and 512MB cache modules are actually the same it’s just that one has more of the same chips on it. On a side note I also noticed the chip (PI74SSTUA32864NBE) at the bottom; which apparently is a registered buffer. Neat! So it looks like that even the Cache Module of the RAID controller is also ECC RAM. So why did mine have five and only show as 256MB? 

I figured that maybe it was simple; perhaps the extra chip is like… Well… RAID actually. One of the chips was a parity chip; or a redundant chip incase one of them fails, or a spillover in case there are bad cells in one of the chips (Like bad sectors on a hard drive) it’d spill over onto the spare chip. Great! I theorised that it was another redundancy thing. I knew that if all eight were filled then that would equal 512MB, but then where was the ninth chip?

I theorised again that it was on the underside of the module. So I opened up the server, took a look at the RAID Controller, disconnected it, took it out of the PCI slot and then took out the cache module. Turned it on it’s side, low and behold...

...There’s another blank spot for another RAM chip! That’s amazing. But it would seem you’d get less redundancy per chip if you had the 512MB cache module. 

Despite the problems with the hard drive being, quite frankly all over the place that might look into, but it can run Windows Server 2016 with absolutely no problems, the drivers aren’t hard to find and load, although you do have to go around the bush about it, not install it via the EXE but extract it and install the .ini file which installed absolutely fine.

So to close I’m going to show some more pictures of the server and title them up and close the topic.

I’m guessing from what’s printed on the board that this is some kind of test switches?

A shot without the PCI Cage installed.

Inside one of the PSU’s overhead shot.

Inside the PSU from the front.

The RAM. Yeah… I matched the colours… Probably not a good idea given that there is a set way that they’re laid out and two of them have slightly different timings.

Directly under the PCI Cage.

Under the PCI Cage and PSU Cage - Overhead Motherboard shot.

PSU Cage Removal and Installation Instructions.

PCI Riser Cage Removal and Installation Instructions

Inside of the top of the case. 

Also inside of the top of the case; and yes. It seems I did install the RAM incorrectly. I will have to fix that; because they really should (in my eyes) be type/brand matched in banks. I know one of the twin sticks of RAM actually has different timings than the rest too. 

So I couldn’t end up using this server, it’s too loud for what I wanted and it’s going to be too much of a drain on our electric. Ultimately I’m going to be giving the server over to someone who needs it and is happy to have it. My brother in law, who I consider to be a brother. I guess he’s getting a heavy christmas present this year.

Next on: "Shit I've bought from eBay" - IBM X3200 M3 Server

[Whoreo]

the next article isn't linked to this post. what did you end up doing with this server in the end? @Temaelrin

[Novaer]

Thanks for that. Fixed.

Oh and as for the the fate: My brother in law now owns it.