Help with getting full performance from AMD Lisbon dual-proc and Hyper-V

1st time here!

Not sure if to post this in AMD's forum or Asus' forum but I see a healthy thriving community at Hardforum and so I'll try here first.

System:
Asus KCMA-D8 Dual-proc motherboard with:
2 x AMD Opteron 4122 (Lisbon) quad-core, 2.2GHz
4 x 4GB Kingston DDR3-1333 ECC (unbuffered)
Various SATA hard drives not in any RAID configuration, each as a single drive.
Windows Server 2008 R2 x64 running as a Microsoft Hyper-V host

Comment: this was meant to be a home test-lab for Hyper-V and a variety of Microsoft servers and virtual machines. This is my first AMD build in a while.

Problem 1: Memtest86+ ver 4.10 gives a very poor memory test speeds. I get a memory test speed of about 3,000 MB/second. I think this is too low. Not sure if the AMD Opti and the motherboard is fully recognized by Memtest86+ but it seems to know that it's an AMD K10. Contrast this with my primary PC, an Intel Core i7-960, 3 x 2GB Kingston DDR3-1333 on a Gigabyte X58A-UD3R, which gets 14,000 MB/sec(!) And it's not overclocked.

Problem 2: The Hyper-V VMs are running quite slowly on this new server. This home test-lab server is going augment my previous home test-lab server, an Intel Q6600 (2.4GHz quad-core, not overclocked) with 8GB DDR2-800 ECC. I created a VM running Win 7 x64-ENT, using just the defaults of 1 "cpu" and 512MB of RAM. It ran sluggishly. I ran the trial copy of the Passmark suite of benchmarks to get a baseline for the AMD-based VM. I then copied the VHD file from the AMD to the Intel box--and thus the VMs were "identical"--and re-ran the Passmark benchmarks for CPU and Memory. The "older technology" Q6600-based system runs the VM nearly twice as fast as the AMD one.

I can't see why there is this poor level of performance from a current technology AMD solution. I bought the lowest-end Lisbon knowing that there would be some sacrifice to performance compared to the 6-core monsters. It was an intentional sacrifice because it's a home test lab and I liked both the low-power aspect of the Lisbon and it was great that the CPUs were so cheap (not more than $100 per CPU). But I'm seeing only 50% of the horsepower of a non-Nehalem Intel proc for a VM under Hyper-V and that's a bad result / investment.

Additional Info. Here are the Passmark scores:

Intel Q6600 Hyper-V host, Passmark scores when run on the host:
CPU: 2731
Memory : 606

Asus KCMA-D8 with 2 x Lisbon 4122, Passmark scores on the host:
CPU: 3568
Memory: 191.9

VM under Q6600, Passmark run in the VM (Win 7x64):
CPU: 603
Memory: 314

VM under AMD Lisbon, Passmark run in the VM (Win 7x64):
CPU: 328.4
Memory: 77 [I did another run and the 2nd run for memory showed "119"; still a bit low]

Questions:

1) Is it Hyper-V being a poor match for or does it have compatibility problems with the Lisbon 4122? Supposedly, I have 8 real cores at my disposal so in theory this was ideal for a home lab.
a) the CPU mark for the AMD is much higher than that of the Q6600. Memory performance is poor however and is in-line with what Memtest86+ is reporting.

2) Is it some sort of power or performance throttling by AMD in pursuit of energy saving (e.g. PowerNow)? I see that the Hyper-V host on the AMD system is reporting a number of cores being "parked". I put this out there because the memory speed fluctuates quite a bit in the AMD VM. If it's some sort of badly implemented power-management, is there a way of forcing full performance? I see a way of "disabling" PowerNow in the BIOS (but I'm a work now and can't try it yet).

Some motherboards disable the CPU Virtualization stuff by default , I would look in the BIOS and see if there is a setting for it.

Also are you using the same ESX Client to connect to both systems? try setting up remote desktop on that VM and see if that is faster?

yeah the problem is definitely the power saving. on the opteron's it's far to aggressive and is why it does so badly in benchmarks. it will wait til the very last second before it has to increase the cpu clock(usually 5-10 seconds) and then down clock the cores well before the benchmarks finish. same goes with the IMC.

try running some benchmarks that will actually utilize all 8 cores and see what kind of numbers you get.

^^^ All the AMD desktop boards Ive dealt with had Virtualization off by default, but I would think Hyper-V would warn of that, I know Virtual Box does.

I think you are right in your thinking to disable PowerNow, I would also try disabling C1E and EIST and any other power states to see if it is regulating power.

I regularly use the version of Memtest built into the latest Ubuntu distro (not sure on version without looking) to test workstations, and all K10s are not properly recognized by Memtest in my experience.

huh interesting all the AMD boards i've used had the virtualization enabled by default. guess it just depends on the manufacture's default bios settings.

Just a followup to my problem. I turned of PowerNow and things are definitely faster. The VMs are now running about 30% to 50% faster.

But the Memtest 4.10 (latest) result still shows an extremely poor memory speed (3000 MB/sec). And this is badly affecting my the performance of the VMs. I've got a support request into Asus about this.

Just to complete the picture,.the VMs are now running close to what my OLD virtual server was able to do on now aging Q6600. :-( It's quite sad actually, I had great hopes of having decent performance out of 8 "real" cores as well as the benefit of ECC memory that this board can support. The price was great.

Just a note that for fun, I temporarily converted my new primary computer into a Hyper-V server. Core i7-960 not overlclocked, 12 GB of DDR3-1333. The VMs under this quick test server are ROCKING! They are very fast and they are doing exactly what I was hoping to do with the dual-proc AMD. So nothing is wrong with my VMs and my Hyper-V config; it's just that the dual-Lisbon has something wrong with it that hopefully Asus can guide me past.