Motherboard

Lenovo 1036 (C422, LGA 2066)

Processor

Intel Xeon W-2140B (Skylake 8c16t, iMac Pro CPU, 120w PL1, 144w PL2)

Memory

4x32gb Kingston DDR4 2400 RDIMMs

Storage

1x Intel D3 S4510 (SSDDSC2KB480G8) (Windows boot drive)

2x Intel 670p 1tb, VROC array on onboard M.2 ports

Graphics

AMD RDNA configuration: AMD Radeon RX 5700 XT (Gigabyte GV-R57XTGAMING Rev 2)

AMD Polaris configuration: AMD RX 580X 8gb (Dell OEM card)

All configs are driving two 2560x1440 60hz monitors

Peripherals

FiiO K3 DAC/AMP (probably worth one or two watts, but included in all tests of all configs)

USB switch, bus powered, with a wireless (2.4ghz, 1000hz) mouse and wired USB keyboard (no backlight) attached

DVD drive I stole out of another computer. Not sure how much this adds.

OS

Windows 10 Enterprise IOT LTSC (21H2) with Hyper-V disabled, NOT a fresh install, but background tasks minimized and only about 5gb of memory “used”at idle.

Ultimate Performance power plan, because it’s my desktop and not having Ultimate Performance enabled doesn’t seem to change much at all (idle and “wave the mouse around” power draw remain the same, and Ultimate Performance did increase Geekbench scores when I tested it a while ago, which is enough persuasion for me.)

Power Supply

900w 80+ Platinum proprietary unit

Fans

3x Delta 92mm (?, recheck) definitely not low power or quiet, incl. one on the CPU cooler

Motherboard

Lenovo Sharkbay (H87, LGA 1150)

Processor

Intel Core i5 4670 (Haswell 4c4t)

Memory

2x4gb RAMAXEL DDR3 1600 UDIMMs

Storage

1x Intel D3-S4510 (SSDDSC2KB480G8)

Graphics

Intel Integrated HD 4600

All configs are driving one 1024x768 monitor via VGA, unless indicated

Peripherals

USB 2 dongle with a wireless keyboard (? hz) and mouse (1000hz)

configurations with 10gb NIC only: SolarFlare SFN7122F

configurations with i350-t4 only: Intel i350-t4

OS

Windows 10 Enterprise 2021 LTSC (21H2) with Hyper-V disabled. Fresh install. Default everything.

Power Supply

240w Bronze FSP proprietary unit

Fans

2x AVC 80 or 92mm, eyeballed. Front model number is DAZH0925R2U (9cm) and #2 is on the CPU cooler. DAZH0925R2U is rated 12v .6a so maximum of around 7w.

Motherboard Waters Corporation 11361K3 (C602, LGA 2011-3 2S platform)

Processor

2x Intel Xeon E5-2680 v2 (Ivy Bridge 10c20t)

Memory

8x16gb Samsung/Cisco DDR3 1600 RDIMMs

Storage

1x Intel DC S3500 (SSDSC2BB480G4)

Graphics

AMD FirePro W5100

All configs are driving a 1024x768 monitor via a DisplayPort to VGA adapter unless indicated

Peripherals

Same wireless mouse and keyboard as the E32

OS

Windows 10 Enterprise LTSC 2021 (21H2)

RHEL 9

VMware ESXi 8

Motherboard

Dell 0HHV7N (C612, LGA 2011-3 1S)

Processor

E5-2690 v4 (Broadwell 14c28t 3.2ghz)

Memory

2x32gb Micron DDR4 2133 RDIMMs

Storage

1x Intel DC S3500 (SSDSC2BB480G4)

Graphics

Nvidia Quadro M2000

Peripherals

Wireless mouse and keyboard

OS

RHEL 9

Fedora 39

Windows 10 Enterprise LTSC 2021

Windows Server 2022

Motherboard

Lenovo 1030 (C612, LGA 2011-3 2S)

Processor(s)

2x E5-2660 v4 (2x Broadwell 14c28t 2.6ghz)

Memory

8x8gb assorted DDR4 RDIMMs

Storage

2x 256gb SATA SSDs

4x recent Seagate HDDs

Graphics

RX560 for display output (it was $20, a waste because it is unused 99.9% of the time)

Peripherals

LSI 2008 (Dell H310 flashed to a generic HBA, still argues with the BIOS in this workstation)

OS

Proxmox 8

Power Supply

650w 80+ Platinum proprietary unit

Fans

3 Delta 92mm (if I remember correctly, including 2 CPU fans) and 2 Delta 80mm

Power on system (cold boot)

Wait approximately five minutes. Close G Hub, Parsec, Discord, Spotify background tasks.

Clear hwinfo averages and timer. Start logging. Wait sixty seconds. Stop logging.

Clear hwinfo averages and timer. Start logging. Move mouse around for thirty seconds. Stop logging.

Open Discord, Spotify, and a Firefox Developer Edition tab with uBlock, Unhook extensions, play LG OLED The Black on YouTube at 1440p 60fps. Clear averages and start logging. Wait sixty seconds. Stop logging.

Start Prime95 Blend test. Wait for Prime95 to load up system memory. Clear averages and timer. Start logging. Wait sixty seconds. Stop logging. Close Prime95 (stop test and exit.)

Start Furmark burn-in test. Clear averages and timer. Start logging. Wait sixty seconds. Stop logging.

Leave Furmark running. Start Prime95 Blend, wait for Prime95 to load up system memory. Clear averages and timer. Start logging. Wait sixty seconds. Stop logging.

Run 1: 5700xt

Suspended:

6.3w from the wall

Running a Fedora 39 live system and doing my very scientific “sit on the desktop” and “sit on the desktop and move the mouse around” stress tests, I see about 90w and 100w from the wall, respectively, but W10 is running the Ultimate Performance power plan and I didn’t touch intel_pstate or even the GNOME controls for scaling governor (or performance preference, I don’t remember what that toggle in the GUI does.) I don’t think 5-20w is worth crying over in this scenario, and I don’t care enough to try and get better Linux data right now because then I’d feel compelled to try a fresh Windows install, too. For my needs, that level of accuracy is not required.

Run 2: RX 580

Original config: with 5700xt, rougher results

140w power on

130w before POSTing (memory check)

95w typical near complete idle (Spotify, Discord, Notion open, after Windows has settled down (10 min post boot))

CPU package: 20w

total DRAM power: 20w

GPU: 38w

Rest: 17w

/tmp 120w typical idle ‘in use’ (moving mouse in near complete idle scenario - Spotify, Discord, Notion open, after Windows has settled down (10 min post boot))

CPU: 40w

DRAM: 25w

GPU: 38w

Rest: 17w

250w Snowrunner (no vsync)

150w booting two VMware Workstation VMs (Debian 4cpu 8gb 40gb on nvme raid 0) with 40 Firefox tabs loading

240w Prime95 Blend (8 workers w/HT) hwm reported split:

CPU package: 118-120w max 122w (120w tdp) spending about 95% of time in C0 and 5% in C3

DRAM 105-110w (reserved 125/128gb)

GPU: 38w

Rest: at least -21w (huh?)

330-350w pl2 Furmark only (160-190w reported GPU power) 300w settled Furmark hwinfo reported split:

• CPU package 50w
• dram 25w

405w settled P95 blend + furmark split:

• Furmark reports 130-190w fluctuating constantly
• hwinfo reports:
• 120w CPU
• 113w dram

130w YouTube 1080p60 Firefox Dev Edition 50w CPU 30w dram

150w browsing, Spotify, Steam download

These have a USB card reader built into the front of the case. I tested with it disabled, though it didn’t seem to increase idle power usage.

What does increase idle power usage is the front 9cm fan, which is rated for 12v .6a. Unplugging it nets about 1.2w of idle savings when the system’s using its noise-optimized thermal profile, but the resulting fan error prevents it from booting normally. Unfortunately, I use these E32s as headless servers, making this a significant point of frustration. Perhaps this could be solved by attacking the BIOS; Lenovo tends to use mostly standard AMT firmware - but that takes effort away from more interesting projects, so until I get really bored I’ll live with the extra watt.

Run 1: Empty

Idle: 20.5w from the wall

This thing is a little odd, but it’s working fine. ‘Production’ hypervisor running Proxmox 8 because this system didn’t like ESXi for some reason (don’t remember the specifics). The HBA isn’t compatible with ESXi, but I added that afterwards and would have passed it to a VM anway.

I like to leave well enough alone, so I won’t be pulling it out of its spot on a wire shelf tucked away from the world and putting Windows on it for benchmarking. I did slip my fake kill-a-watt on it overnight, though.

This bad boy only uses about 95w at idle, which is about 95w lower than I’d expected, considering it has four pieces of spinning rust in it and two CPUs. I suppose the newer architecture and lower capacity DIMMs are good on power and the processors spend 98% of their time in C6 at 1.2 GHz (yes, I checked.)

I could, and should, probably go at this one with powertop to try and cut a few more watts off by shutting down stuff like the card reader (Lenovo, card readers, and power consumption on Linux don’t go together nicely, as most ThinkPad T480s owners find out) but for now, just knowing how surprisingly good it’s being is enough for me. I would pull the second CPU, but then I’d lose 1/3 of my memory, and it’s already starved for memory as-is (a large portion is handed to ZFS for cache.) If I do go at it with powertop or swipe a processor out, I’ll probably update this page.

Regardless, this machine is going to be up for replacement shortly as it’s the only one on 24/7 that is not an i5 4670 (and I want it back in my lab because it’s beefy and has a ton of neat toggles in the UEFI to play with.)