I have run a casual image-processing (Imagic-V based) test of the computers used since 1997. A rough indicator for when any given batch job will end, to aid day-to-day, or even for monthly scientific planning.
The longest continuous computing jobs lasted up to 6 weeks, but most jobs were completed within 8-12 hours.
The below data are designed only to reveal the general trend in performance.
[Note normalised values, given a change in test in 2002]

Summary:
A single Pentium 200 machine (1997) may be compared to 3x fan-less (55 Watt TDP) i3770 Intel processors (2013) = approx. 3,500x increase in performance for the same cost and slightly less power.
A computing batch job on my desk (2013) that took 30 days in 1997 now takes 12.5 minutes, and in silence with no air-conditioning. Recent work is conducted modestly on a single i7-8700K with two GTX 1070 GPUs, 64GB DDR4-2666 RAM, 2TB SSD + 2x 12TB Seagate HDDs, before submitting more lengthy jobs elsewhere. Offline backups of key data are made to 4TB HDDs and stored at two different geographical locations.

Core System Specifications / RAM / Storage
Time taken to process script (secs), date, ~system cost (in GBP)

Compaq iPaq 3630 Pocket PC (!) running a custom Linux distro / 32MB / 256 CFII
205, 3/02, 200

Pentium 200 / 128MB / 2GB EIDE
170, 1/97, 2500

Celeron 300 Laptop / 192MB / 3.2GB EIDE
78, 2/00, 1000

DEC Alpha 255 / 164MB / 16GB SCSI
67.5, 1/97, 6000

Pentium III-450 / 256MB / 8GB EIDE
3, 3/99, 700

Celeron 433 single CPU / 512MB / 8GB EIDE
46.5, 4/00, 450

Celeron 433@541 single CPU / 512MB / 8GB EIDE
21, 4/00, 450

Athlon 500 / 128MB / EIDE
21, 11/99, 1000

DEC Alpha PWS 433 / 256MB / SCSI
17, 1/98, 4500

DEC Alpha PWS 500 / 320MB / SCSI
14, 2/99, 2000

DEC Alphastation 500/500 / 256MB / SCSI
13, 1/98, 10000

Athlon Tbird 850 / 512MB / EIDE
12.1, 7/01, 700

Celeron 433@541 dual CPU / 512MB / EIDE
18, 4/00, 450

Athlon Tbird 1000 / 512MB / EIDE
14.1, 9/00, 1200

Athlon Tbird 1000@1100 / 512MB / EIDE
9.8, 9/00, 1200

Sony FX405 Athlon4@1000 / 256MB / EIDE
9.4, 11/01, 1300

Athlon Tbird 1400 (PC133)
12, 9/01, 1000

Compaq XP1000@500MHz / 512MB / SCSI
10.5, 4/99, 6600

Compaq XP1000@667MHz / 512MB / SCSI
7.8, 9/00, 4500

Sony FX405 Athlon4@1000 / 256MB / EIDE
6.1, 11/01, 1300

P4 2.0 / 533 FSB/ 256MB / EIDE
3.72, 3/03, 430

Centrino 1.7 single CPU / 512MB / EIDE
3.23, 12/04, 1500

Athlon 1900+ MP single CPU / 1024MB / EIDE
3.2, 4/02, 900

Celeron 2.4 / 512MB / EIDE
3.1, 7/03, 330

P4 2.8 / 533 FSB / 1024MB / EIDE
2.9, 11/02, 1150

Athlon 2000+ XP/ 256MB / EIDE
2.85, 3/03, 380

Athlon 2400+ XP-M / 1024MB laptop
2.7, 9/03, 700

Athlon 2600+ XP/ 256MB / EIDE
2.36, 11/03, 400

Athlon 2400+ XP/ 512MB / EIDE
2.35, 1/03, 500

Athlon 2000+ MP dual CPU / 512MB / EIDE
1.76, 8/02, 700

AMD 3800+ dual 2ghz / 1024MB / 500GB
1.71, 10/05, 350

Intel Core2Duo E6600 4ghz / 1024MB / 250GB
1.47, 10/06, 500

Intel Core2Quad Q6600 9.6ghz / 2048MB / 250GB
0.59, 06/07, 700

Athlon 2000+ MP CPU (x10) cluster; mpi
0.454, 2/03, 3500

Athlon MP & shuttle CPU cluster / UPS
0.162, 3/04, 9000

Athlon MP & shuttle CPU cluster + long test
0.138, 12/04, 11500

Athlon MP & shuttle CPU cluster + short test
0.118, 12/04, 11500

AMD shuttle + Intel CPU cluster + short test
0.067, 06/07, 16500

2007: move to new institution (Imperial -> QMUL): loss of dedicated a/c facility.
8 x Q6600 servers; standalone Imagic-V licence Feb. 2008 / EMAN2 / GRIP, Chimera/PyMol, and others.
0.17, 09/07, 4800

2010; a/c reinstated, larger cluster refurbishment commences.
2012; 3 x i3770K intel servers / SSD raid
0.05, 12/12, 4000

2016; 2 x i7-4790K, 16TB, 32GB: short-term, pre-cluster, image alignment and classification.
0.164 (6 cores total) 06/16, 1680, no a/c required ->deprecated

2018; 1 x i7-8700K Coffee Lake hexacore, 24TB, 64GB, 2x GTX 1070 pre-cluster work
0.153 (5 cores, no GPU), 05/18, 2600, no a/c required -> to Viles group

2018 November: AMD 2950X, dual 1080 Ti, 14TB, 64GB, very quiet, no a/c required in large room. -> to Viles group