I'm currently playing with Amazon Web Services and EC2, Amazon's highly flexible VPS hosting service. Gandi Hosting is providing a similar Xen based hosting service, and last day I saw that they benchmarked their basic "one share" server with Unixbench and got a score of 40.

I was curious to see the score I could get with a small Amazon EC2 server, so I quickly started an instance and the same Unixbench release. The score I got for a small EC2 instance was 34.5 :

BYTE UNIX Benchmarks (Version 4.1-wht.2)
System -- Linux ip- 2.6.21.7-2.fc8xen #1 SMP Fri Feb 15 12:39:36 EST 2008 i686 GNU/Linux
/dev/sda1             10321208    578372   9218548   6% /

                     INDEX VALUES            
TEST                                        BASELINE     RESULT      INDEX

Dhrystone 2 using register variables        376783.7  2016468.0       53.5
Double-Precision Whetstone                      83.1     1267.8      152.6
Execl Throughput                               188.3      835.7       44.4
File Copy 1024 bufsize 2000 maxblocks         2672.0    11164.0       41.8
File Copy 256 bufsize 500 maxblocks           1077.0     3059.0       28.4
File Read 4096 bufsize 8000 maxblocks        15382.0    74927.0       48.7
Pipe-based Context Switching                 15448.6    21771.7       14.1
Pipe Throughput                             111814.6    58972.3        5.3
Process Creation                               569.3     1361.2       23.9
Shell Scripts (8 concurrent)                    44.8      171.8       38.3
System Call Overhead                        114433.5   670336.1       58.6
                                                                 =========
     FINAL SCORE                                                      34.5

So if we believe Gandi.net, basic Amazon EC2 and Gandi instances have roughly the same power. But later I found the following Gandi benchmark, which shows much lower results that Amazon's:

...
     FINAL SCORE                                                      20.4

But I didn't want to test myself Gandi Hosting because I would have needed to create an account and pay an initial fee of 14 Euros, just for playing 30 minutes with one server. With Amazon Web Services, running this benchmark cost me only $0.10 ! That's what I really find interesting with AWS: everything you use (CPU, Storage, ...) is billed on an hourly basis so it's very cheap for quick tests. And if tomorrow I ever want to run this benchmark on a "Medium High CPU" (5 times more CPU than a "small" instance) it will only cost me $0.20. But I admit that the big plus for Gandi Hosting is that bandwidth is free whereas you'll be billed up to $0.17 for each GB transferred on AWS.

Update !

I asked a friend to run the benchmark on its Gandi VM and here are his results:

==============================================================
BYTE UNIX Benchmarks (Version 4.1-wht.2)
System -- Linux 2.6.18-xenU #4 SMP Mon Sep 22 17:59:36 CEST 2008 i686 GNU/Linux
/dev/xvda1             2466700   1212380   1129016  52% /

Start Benchmark Run: mardi 9 mars 2010, 19:59:57 (UTC+0100)
 19:59:57 up 201 days, 18:57,  2 users,  load average: 1.17, 0.31, 0.10

End Benchmark Run: mardi 9 mars 2010, 20:16:47 (UTC+0100)
 20:16:47 up 201 days, 19:14,  2 users,  load average: 17.42, 7.55, 3.83


                     INDEX VALUES            
TEST                                        BASELINE     RESULT      INDEX

Dhrystone 2 using register variables        376783.7   920333.0       24.4
Double-Precision Whetstone                      83.1     1240.4      149.3
Execl Throughput                               188.3      536.1       28.5
File Copy 1024 bufsize 2000 maxblocks         2672.0    20431.0       76.5
File Copy 256 bufsize 500 maxblocks           1077.0     7771.0       72.2
File Read 4096 bufsize 8000 maxblocks        15382.0    79094.0       51.4
Pipe-based Context Switching                 15448.6    29433.3       19.1
Pipe Throughput                             111814.6   210581.0       18.8
Process Creation                               569.3      869.3       15.3
Shell Scripts (8 concurrent)                    44.8       62.9       14.0
System Call Overhead                        114433.5   355747.7       31.1
                                                                 =========
     FINAL SCORE                                                      34.0

We see that, Gandi VMs performance is: - 50% lower for CPU integer, process creation, shell scripts, system calls - the same for FPU performance - higher for disk related operations - much higher for pipe throughput (caused by a different kernel?).

So that's hard to see a clear winner here, and you should try with your application to determine which one is better. Running an apache or database benchmark would be something interesting !