A Speed & Power Consumption Benchmark and Comparision for Pegasos Efika, WinUAE and Amiga

Because of the repeated question how the performance of a Pegasos and/or an Efika running MorphOS is compared to maschines running WinUAE or Commodore/Escom 68k based Amigas is like, I did a brief benchmark regarding speed and electric power uptake with my maschines for a simple comparison.
Please note for the 68k JIT emulation under MOS 1.4.5 running on the 603e processor, that this processor is not ideally suited for JIT emulation since it's L1 cache is rather small and a L2 cache is absent at all.
The list is thought to be extended...

Performance of the 68k emulation

Standard Mandelbrot in 1024x768 pixel resolution

Mac mini - G4/1500: 0.13s (PowerUp) | 0.85 - 0.90s (68k JIT)
Pegasos1 - G3/600: 0.47s (PowerUp) | 2.50 - 2.62s (68k JIT)
WinUAE PIII/650: 4.20s

WinUAE Celeron 2.4:

4.09 (massive gfx artifacts!)
A1200: 68k(AOS): 48.70 (68040) || 109.22s68k Emu (MOS 0.4) || 1.87s (PowerUp - MOS 0.4)
A1200MOS: 1.87s (PowerUp) || 8.37s (68k JIT)
Efika 5200B: 0.93s (PowerUp) | 5,2s - 5,4s (68k JIT)

Figure 1: Factors for test Benoit. Reference value is the time required for the task by the A1200 68k (68040 @ 25MHz).

Maxon Cinema
Scene "Glassletter", 640x480 pixels resolution, Rendering without display.

Powerbook 5.6 - G4/1667: 17s (68k JIT) || 106s (ohne 68k JIT, static 68k)
Mac mini - G4/1500: 29s-31s (68k JIT) || 119s (w/o 68k JIT, static 68k)
Pegasos1 - G3/600: 54s-49s (68k JIT) || 218s (w/o 68k JIT, static 68k)
WinUAE PIII/650: 344s

WinUAE Celeron 2.4:

A1XE 1000: 43s (68k JIT) || 215s (w/o 68k JIT, static 68k)
A1200: 68k(AOS): ca. 1620s (68040) | 2220s (MOS 0.4 static 68k-Emulation)
A1200MOS: 677s (68k JIT)
Efika 5200B: 208s (68k JIT)

Figure 2: Factors for test Maxon Cinema. Reference value is the time required for the task by the A1200 68k (68040 @ 25MHz).

Data throughput

Harddisk performance
tested with MasterISO V2.4 (68k) by Asimware. This test uses mostly small files and is rated as some kind of 'worst case test' and is intended only for comparision between the different setups rather than for absolute hdd performance.

Mac mini - G4/1500: IDE Device 34643 KB/s w/ Hitachi HTS54108 80GB 5400 rpm 8MB Cache, ~ 9% CPU load
Pegasos1 - G3/600: VIA IDE Device 26208 KB/s w/ Western Digtal WD800JB 80GB 7200 rpm 8MB Cache, ~ 14% CPU load
10328 KB mit Maxtor Diamond Max 80GB 7200 rpm 2MB Cache, etwa 18% CPU-Last
A1200MOS: SCSI Blizzard Device 2156 KB/s w/ Quantum AtlasII 4.5 GB, 7200 rpm, ??? Cache, 15% CPU load
X-Surf IDE Device 1490 KB/s w/ IBM DNJA 13.5 GB, 7200 rpm, 512 kB Cache (?), ~ 30% CPU load
Efika 5200B: 5200B IDE Device 5258 KB/s w/ Samsung MP0402H 40GB 5400 rpm 8MB Cache, ~ 70% CPU-Last

System RAM speed
values obtained with memtest (MorphOS).

Mac mini - G4/1500 write: ~759 MB/s | read: ~368 MB/s
Powerbook 5,6 - G4/1667 write: ~714 MB/s | read: ~351 MB/s
Pegasos1-G3/600 write: ~71 MB/s | read: ~176 MB/s
Efika 5200B write: ~137 MB/s | read: ~127 MB/s

CPU load during operation

values obtained with Task Manager (MorphOS included). Layers3D (L3D) increases gfx performance, but adds additional cpu load to the system.

Mac mini - G4/1500 Idle ~0,8% (L3D off) || ~1,4% (L3D on) || Songplayer mp3-Replay (128 kbps) ~ 2,3-2,5%
Pegasos1-G3/600 Idle ~0.9% || Songplayer mp3-Replay (128 kbps) ~ 6.7%
A1200MOS Idle ~4% || Songplayer mp3-Replay (128 kbps) ~40%
Efika 5200B Idle ~1,9% (L3D off) || ~3,5% (L3D on) || Songplayer mp3-Replay (128 kbps) ~ 20%

Electrical power uptake

The power uptake is a netto measure at the electrical outlet, i.e. the efficieny factor of the PSU is included within the obtained values.
Notes regarding the Vaio laptop measurements: The measurements were done *without* inserted battery to avoid higher power consumption values due to some battery loading currents. Typical denotes the usesage with WinUAE, Win-typical denotes the typical power uptake running Windows at typical load (WinUAE is *not* running), the Minimum value denotes the absolute minimum power uptake of the ready maschine (WinUAE is *not* running).

Mac mini G4/1500 typical: 30-32 Watt
Max value: ~50 Watt (CPU load 100%, heavy DVD/HDD activity)
Min value: 23-25 Watt (idle)
ATX standby: ~5 Watt
Pegasos1-G3/600 typical: 36-38 Watt
Max value: ~55 Watt (CPU load 100% and max. CDRW/HDD activity)
Min value: ~33 Watt (idle)
ATX standby: 9 Watt
Efika 5200B typical: ~14 Watt (w/o display) || ~29 Watt (display on)
Max value: 34 Watt (3D accelerated game, Wipeout 2097, display on) || 19W (display off)
Min value: 12 Watt (idle, display off) || 27 Watt (idle, display on)
ATX standby: 6 Watt
Sony Vaio w/o battery!!
typical: 39 Watt
Max value: 41 Watt
Min value: 18 Watt (idle, UAE inactive, display off)
Win typical: 35 Watt
Win typical (Display off): 22 Watt
Windows sleep: 7 Watt
AT standby: <5 Watt
Eee 900A
typical: 18-20 Watt (235 V AC operated *w/o* attached battery) || <10 Watt (battery operated)
Max value: 23 Watt (235 V AC operated) || 40 Watt (235V AC + batteryattached (charging power))
Min value: 14 Watt (idle, display off, 235V AC w/o battery)
Linux sleep: 6 Watt (235V AC)
ATX standby: <3 Watt (235V AC)


Powerbook - G4/1667 Apple Powerbook 5.6. Processor: Freescale MPC7447A (G4) @ 1667MHz. 1536 MB RAM. Gfx: ATI 9700M, 64MB VRAM.
OS: MorphOS 3.9

Mac mini - G4/1500 Apple Mac Mini G4. Processor: Freescale MPC7447B (G4) @ 1500MHz. 1024 MB RAM. Gfx: ATI 9200, 64MB VRAM. harddisk: Hitachi HTS54108 2,5". DVD/CDRW Combo-Laufwerk, usb-Keyboard and Mouse.
OS: MorphOS 2.4

Pegasos1 Pegasos 1 April 1. Processor: IBM 750FX (G3) @ 600MHz. 512 MB RAM. Gfx: Club3D ATI 9250SE, 128MB VRAM. harddisk: Maxtor Diamont Max 7200rpm. Samsung 52xCDRW, standard 300W ATX power supply.
OS: Morphos 2.4 and MorphOS 1.4.5

Efika 5200B Efika 5200B (V2.3). processor: Freescale MPC 5200B (e300) @ 400MHz. gfx: Club3D ATI 9250SE gfx card, 128MB VRAM. Integrated 10,4 inch SVGA-TFT display. harddisk: 40G Samsung 2,5" (MP0402H). Pico PSU & Morex 60W 12V AC/DC converter
OS: MorphOS 2.4

WinUAE PIII/650 Sony Vaio PCG QR10. processor: Intel PentiumIIImobile (Celeron) Coppermine 650 Mhz. 320 MB RAM. Gfx: Intel 815EM, shared VRAM
OS: Windows Me + WinUAE 0.9.91 + AOS3.9 BB2, 64 MB Z-3 Ram, 4 MB Chipram, 8 MB gfx RAM, ECS, Picasso96, no sound, 68040 with FPU, JIT max Cache.

Eee 900A Asus Eee 900A. processor: Intel N270, 1024 MB RAM. gfx: Intel 945GSE Chipsatz, shared VRAM. harddisk: 8 GB SSD. Display: 8,9" mit 1024x600 px Auflösung.
OS: Xandros Linux

WinUAE Celeron 2.4 Processor: Intel Celeron 2.4 Ghz, 1024 MB RAM. gfx: ATI Radeon 9250, 128 MB VRAM.
OS: Windows 2000 SP4 + WinUAE 8.17 + AOS3.9 BB2, 64 MB Z-3 Ram, 4 MB Chipram, 8 MB gfx RAM, ECS, Picasso96, no sound, 68040 with FPU, JIT max Cache.

A1200 A1200 mit BlizzardPPC. Processors: Motorola 68040 @ 25MHz and IBM 603e @ 200MHz. 2 MB Chipram, 64 MB 32 Bit Fast RAM, BVision/8MB VRAM.
OS: AOS3.9 BB2 and MorphOS 0.4

A1200MOS Identical to A1200 but with MorphOS V 1.4.5

Maschines used by other contributors:
A1 XE 1000 AmigaONE XE G4. Processor Feescale MPC 7457 @ 1000Mhz 1024MB PC133 RAM. Gfx: ATI 9200 256MB. Hardisk: Seagate ST325062 7200rpm. OS: Amiga OS4.1 update 2.
Measurements contributed by user "Rob" of amigaworld.net

Remarks and brief conclusion:

Mac mini G4/1500:
The Mac mini with MorphOS outperforms all other systems by far. In regard to the subjective (perceived) speed and OS reaction times I have never experienced something comparable. But also on heavy computing tasks the Mac mini is still in a good position. In general it is a very different league than the 68k based systems. The gap is obvious and huge. Also the emulated systems cannot compete in smothness and snapiness. Altivec accelerated software benfits even more. The system can compete with current low end x86 based systems (e.g. Intel Atom) pretty well. For example HD video replay works up to 720p w/o stressing the system too much, but reaches its limit at 1080.

Perceived the Pegasos1 feels much faster and smoother than WinUAE on the PIII or the 2.4 Ghz Celeron. The latter both feel surprisingly similar. MorphOS on the Pegasos seems much, much faster than OS3.9 under UAE - it pretty much benefits from being native and nothing has to be emulatd. With MorphOS 2.x the overall performance has been increased once more. With Layers3D the system gfx got faster and more powerful. In comparison to the Mac mini the Pegasos1 still feels conveniently fast and keeps a good system for many of the everyday tasks like webbrowsing, listening to music or editing texts. For current market prices it is a pretty good catch. Since the cpu needs no active cooling a 100% silent system can easily be build on basis of the Pegasos1.

MorphOS on the Blizzard is surprisingly fast! 68k applications are faster by factors than with the original 68040 processor and reachs the performance level of a 68060. Sincee the OS is PPC native the average overall perfromance is similar (or even a bit above) with teh overall performance of the WinUAE PIII setup.

Efika 5200B:
The Efika yields about two to four times the performance of the Blizzard setup. Subjective the Efika is well between the Pegasos and the Blizzard and feels faster than both WinUAE setups. The Efika RAM interface does at least 137MB/s writing and 127 MB/s reading, while the Pegasos1 (April1) does 176 MB/s reading and 71MB writing (memtest, MorphOS). The Efika is fast enough for many tasks: Surfing the web is usually convenient (e.g. google maps works fairly okay, for sites requireing lots of flash teh processor speed is not enough though), movies up to DVD resolution are fluently replayed (depends on the codec though), pdfs are rendered fast enough for a fluent work flow. It is good though not to use the most demanding MUI skins to avoid even slightest slowdowns of teh GUI and save available RAM. Layers3Dkeeps the system snappy even under heavy load. Since the MorphOS release V2.3 the speed got increased once again, usb works pretty stable now and the RAM usage got minimized (MorphOS 2.5 will even lower the RAM footprint once again, due to freeing gfx mem from the RAM and holding it exclusively in VRAM).
A downside of the 603e Core is the lack of a 2nd level cache. Especially teh 68k JIT emulation gets hit by this, hence ppc native code benefits on the 603e even more than on the G3/G4.

Electric power uptake:
the uptake of the Pegasos 1 and des Mac mini are similar and generally pretty low for desktop systems - while typical usage teh uptake is ranges at 30 W (Mac mini) or 35-40W (Pegasos 1). Thus, evenn long uptimes get neither desaterous for teh wallet nor the environment. Under maximum load the Mac mini peaks at 50W, the Pegasos1 at about 55W. Wile at maximum load the Mac mini delivers about three times the computing power. The usage of a low power 2.5" harddsik by the Mac mini adds to the positive effect of teh generally low uptake.
The Efika setup has an power uptake lesser than 15W w/o display. It can nicely be used as a low energy jukebox and/or small server. With the display in operartion the uptake stays usually below 30W and thus, at 3/4 of the typical uptake of the tested laptop but a bit above the netbook. The display of the Efika delivers teh best quality of these three devices. Withou the display the power uptake of the Efika is at about 2/3 of the uptake of the Vaio laptop w/o display and at about 40% of the uptake of the Pegasos 1. The Intel Atom based netbook delivers similar results. A board based on the Freescale MPC5121 could reduce the power uptake significantly to below 10W, while offering a increased system performance. A board based on the Freescale MPC 8610 could be a good basis for a very power saving system which would deliver performance above the Mac mini. At 1330 MHz an uptake below 20W at typical usage would be expected.
Currently the Mac mini offers the best ratio of computing power/Watt. For non demanding computing tasks the Efika with an uptake less than 15W offers a nice alternative.

©[ujb] 2010
last change: 12-12-2015

back to index page