diff --git a/HPC_core_losses.md b/HPC_core_losses.md
index b7add66..3c62297 100644
--- a/HPC_core_losses.md
+++ b/HPC_core_losses.md
@@ -1,7 +1,10 @@
 As requested, over the next 12 months we will (probably) be losing:
-  1/8th of Polaris (5312/8 = 664 cores)
-  Arc2: (3040 cores, plus private nodes)
-  
+  1/8th of Polaris (5312/8 = 664 cores, plus 192 private cores)
+  Arc2: (3040 cores, plus 2352 private cores)
+  Marc1: (1236 private cores, including a couple of 3TB RAM nodes)
+
+  => 7484 cores (3704 public, 3780 private)
+
 Arc4 will probably replace the 3040 we're losing from A2, but until we know how 
 much we can spend we can't really estimate that core count.
 
@@ -11,4 +14,12 @@ GPU is useful but we might want to consider spending in a similar way to A3, ie:
   2nd tranche we take a greater look at novel architectures, should the community 
   wish.
 
-Anything else?
+That's assuming we alter the way we schedule cluster purchases. We currently
+buy a completely new cluster every 2 years and retire after 4 (or so)
+years. This has the advantage of always having two machines so that WHEN one
+breaks, the other can be used.
+
+The existing model probably remains the best practical option with biennial
+procurements after all, as a new -> upgrade -> new -> upgrade cadence implies
+running compute for much longer than 4 years to maintain this resilience.
+
diff --git a/README.md b/README.md
index f42387e..842dfbe 100644
--- a/README.md
+++ b/README.md
@@ -12,11 +12,37 @@
 * We now have about 2-3 times the number of active users of the ARC service than we did 5 years ago. Since investment in both people and equipment has remained static, the amount of support and resource available per researcher has dropped.
 * We are probably heading for a contraction in the number of cores next year as quite of kit is getting old and we have insufficient investment to replace at a similar level.
 
+### Machine details
+
+#### Current machines
+
+* ARC3 (EOL 2021?)
+..* 6336 cores, 26 GPUs and 2 Xeon Phi "Knights Landing" for general use
+..* 888 cores and 18 GPUs (2017 vintage) owned by research groups in FBS, ENV, MAPS
+* ARC2 (EOL 2019?)
+..* 3040 cores for general use
+..* 2352 cores (various 2010-2015 vintages) owned by research groups in ENG, ENV, MAPS
+* MARC1 (EOL 2019?)
+..* 1236 cores owned by LIDA research groups in MEDH, ENV
+
+#### Previous major procurements
+
+* 2009 (ARC1, Sun) - 445 gbp/core  [240 nodes / 1888+64 cores / 869565.21 gbp]
+* 2010 (MHD, HP) - 362 gbp/core  [160 nodes / 1920 cores / 695175 gbp]
+* 2012 (N8HPC, SGI) - 463 gbp/core  [332 nodes / 5312 cores / 2459060.40 gbp]
+* 2013 (ARC2, HP) - 329 gbp/core  [190 nodes / 3040 cores / 999600 gbp]
+* 2014 (HERC1, SGI) - ? gbp/core [1 node / 256 cores / ? gbp]
+* 2015 (MARC1, HP) - 614 gbp/core  [98 nodes / 1236 cores / 758936.16 gbp]
+* 2016 (ARC3, Dell) - 302 gbp/core  [167 nodes (4008 cores) + 2 GPU nodes(4 GPUs) => cost equiv to "96" std cores / 1239348.0 gbp]
+* 2017 (ARC3.5, Dell) - 295 gbp/core  [121 nodes (2904 cores) + 6 GPU nodes(24 GPUs) => cost equiv to "576" std cores / 1027606.42 gbp ]
+
+
 ## Regional High Performance Computing
 
-* We have also been running an additional 5,500 cores  as a regional machine on behalf of the N8 Research Partnership (soon to be 
+* We have also been running an additional 5,312 cores  as a regional machine on behalf of the N8 Research Partnership (soon to be 
 decommissioned)
 * 500  external users of the N8 machine.
+* Also has 192 cores owned by a research group in MAPS
 
 ## Integrated Research Campus