AMD Vi vs Intel VT-d

[rockafella]

I summarized my issue on another forum (http://www.tomshardware.com/forum/359329-28-intel-running#t2730285), and I believe I have come to the conclusion that the reason my VM does not run as snappy on my Phenom II 965 desktop as on my new MBP Retina is that the i7 in the latter supports VT-d, but my desktop mobo with the 890GX chipset does not support AMD's Vi (IOMMU).

Can anyone offer an opinion on whether I will realize most of the performance gain by simply replacing my mobo with one that has an 890FX or 990FX chipset? Or will the Intel with vt-d always be superior?

On the Mac I use Parallels Desktop (VMWare Fusion ran faster there too). On Windows, Parallels Workstation is more responsive than VMWare Player - but both are much slower feeling than on the Mac. I know some people think these consumer products don't even use VT-d, but from the Parallels forums it appears that Workstation DOES use it.

 

[faugusztin]

It is unlikely due VT-d, because VT-d is just a hardware device (and respective direct memory access) virtualization. In other words, it could help you if you assign a PCI-E device to specific VM, but it doesn't help the general performance of the virtual machine alone. What you are seeing is the performance difference between Phenom II 965 and Ivy Bridge Mobile Core i7.

VT-d is for use cases like you got a extra NIC you want to assign directly to the VM as it is connected to it, and not virtualized. Or a graphics card, tv card, sound card etc.

 

[rockafella]

It is unlikely due VT-d, because VT-d is just a hardware device (and respective direct memory access) virtualization. In other words, it could help you if you assign a PCI-E device to specific VM, but it doesn't help the general performance of the virtual machine alone. What you are seeing is the performance difference between Phenom II 965 and Ivy Bridge Mobile Core i7.

VT-d is for use cases like you got a extra NIC you want to assign directly to the VM as it is connected to it, and not virtualized. Or a graphics card, tv card, sound card etc.

Okay, I get it. I knew it was for the devices, but I thought perhaps somehow it was also helping with memory/disk/video access (but video is already pass through even without VT-d from what I understand).

So it would seem even upgrading to Piledriver wouldn't help (not that it would be any simpler - I'd still need to change my mobo, at which point I'd just get an i7 anyway)? I've found other comments in various forums that Windows VMs will always be more responsive on Intel setups, although it's all anecdotal (some mention the fact that Microsoft uses the Intel compiler as a strong reason though to support their conclusions).

 

[faugusztin]

It is more likely down to stuff like power saving feaures or storage bottlenecks. It could be that the "performance" difference you are seeing is the difference in system drawing of the UI itself.

 

[dandragonrage]

I use an OCed Sandy Bridge and I/O performance in VMWare Workstation in a Windows VM is absolutely terrible for me too... They have 2-3 controllers you can try. I haven't really screwed around with it much at home because I don't use those VMs that much anyway.

 

[rockafella]

Hmm, you both have given me some new things to try. First, I do have Cool n Quiet enabled (to throttle down the CPU to 800 Mhz when not in use), and I see it's always doing that with my CPU monitor widget I always have open. I'll try turning that off!

The storage controller could be something too. I noticed my VMs of course ran faster once I moved them to my desktop's SSD array, but considering I always felt my SSD RAID 0 array were not running at full potential, perhaps there is room for a fix or upgrade there. The Mac's SSD is much newer tech and very high performing (OS X boots in less than seven seconds, including reopening my browser, and any of my VMs boot in less than 8-10 seconds, whereas they seem to be taking 15-25 seconds on my desktop).

I agree it's definitely something with the system drawing. For example, just moving a window around the desktop I notice tearing of the window contents as I rapidly move it around, or if I click to drag a selection on the desktop, I can see that the boundary line is not straight, like it's being displayed halfway through drawing the new boundary of my selection. On the Mac, I notice none of this, Windows truly feels like it's bare metal.

It just sucks that there's no real way to know what upgrade might make the big difference I am looking for. If it comes down to upgrading the processor to i7, I will just stick with my Mac laptop for my VM needs for now.

 

[rockafella]

I use an OCed Sandy Bridge and I/O performance in VMWare Workstation in a Windows VM is absolutely terrible for me too... They have 2-3 controllers you can try. I haven't really screwed around with it much at home because I don't use those VMs that much anyway.

Does yours have the VT-d support, or is the K model with the unlocked multiplier? (or is that only relevant for Ivy Bridge - I am not really in the loop on processors this last year!)

 

[dandragonrage]

Does yours have the VT-d support, or is the K model with the unlocked multiplier? (or is that only relevant for Ivy Bridge - I am not really in the loop on processors this last year!)

No VT-d support. As mentioned before, it would make no difference anyway unless using it to pass through a NIC or HD controller or whatever. Even if it were enabled on the machine, VMWare Workstation would have no way to use it as Windows is already using it. The best I/O performance you can get from VMWare Workstation is when using a paravirtualized HD controller with a guest OS that supports it. But otherwise, the performance between the emulated controllers CAN vary, especially depending on the guest. I bet I could improve things in my WinXP VM if I were to play around with which controller is used.

I do have a VMWare ESXi server that uses VT-d. I pass through a NIC for one VM and a RAID card for another VM.

 

[Nate7311]

Hmm, you both have given me some new things to try. First, I do have Cool n Quiet enabled (to throttle down the CPU to 800 Mhz when not in use), and I see it's always doing that with my CPU monitor widget I always have open. I'll try turning that off!

The storage controller could be something too. I noticed my VMs of course ran faster once I moved them to my desktop's SSD array, but considering I always felt my SSD RAID 0 array were not running at full potential, perhaps there is room for a fix or upgrade there. The Mac's SSD is much newer tech and very high performing (OS X boots in less than seven seconds, including reopening my browser, and any of my VMs boot in less than 8-10 seconds, whereas they seem to be taking 15-25 seconds on my desktop).

I agree it's definitely something with the system drawing. For example, just moving a window around the desktop I notice tearing of the window contents as I rapidly move it around, or if I click to drag a selection on the desktop, I can see that the boundary line is not straight, like it's being displayed halfway through drawing the new boundary of my selection. On the Mac, I notice none of this, Windows truly feels like it's bare metal.

It just sucks that there's no real way to know what upgrade might make the big difference I am looking for. If it comes down to upgrading the processor to i7, I will just stick with my Mac laptop for my VM needs for now.

RAID 0 or not, it sounds like your issue is more of a HDD throughput limitation than anything else. Obviously, there's a processing performance deficit for the Phenom II (I'm running the same chip myself), but given that it's basic Windows stuff, it sounds more like the updated SSD in the Mac is simply outperforming the array of older SSD's. SSD's have come a long way in the last 2 years. What is the older array made up of? SSD's and controller.

 

[rockafella]

RAID 0 or not, it sounds like your issue is more of a HDD throughput limitation than anything else. Obviously, there's a processing performance deficit for the Phenom II (I'm running the same chip myself), but given that it's basic Windows stuff, it sounds more like the updated SSD in the Mac is simply outperforming the array of older SSD's. SSD's have come a long way in the last 2 years. What is the older array made up of? SSD's and controller.

I think you're right. I was going to get one of the Corsair ones but I got these Patriot Pyros because they were on sale. At the time they seemed to be the latest generation (or at least not very old tech), but like I've said myself, I never felt like they were stellar:

http://www.frys.com/product/6728115#detailed

The RAID controller is the built-in one on my motherboard:

http://www.newegg.com/Product/Product.aspx?Item=N82E16813131633

Are you thinking upgrading my SSD and perhaps using a different controller via expansion slot would be beneficial? In my other posting, a user with a 3.4 Ghz clock on a six core said theirs ran very smooth (of course it's all subjective), and mine is 3.6 Ghz. I was going to try loading up my not so stable 4 Ghz overclock to see if that made any difference. But I think the drive/controller might be the bottleneck!

 

[Nate7311]

Looking at the reviews, that's more than a bit on the slow side for a modern SSD.
http://www.legitreviews.com/article/1662/6/

Even in RAID0, those aren't going to match the speed of a current-gen or a next generation SSD. A Crucial M4, Samsung 840 Pro (or 830 if you can still find them), or high-end Sandforce drive should easily rock better numbers w/o having to deal with firmware-raid arrays and the reliability potentials involved. You might also look at firmware updates for your Pyro's as well. Firmware updates can make a DRAMATIC difference in performance.

ALong with that, let's keep in mind the utterly different platforms that you are comparing the virtualization on:

Mac w/ i7 and faster SSD running Parallel's Desktop
PC w/ older Phenom II and RAID0 of slower SSDs running VMWare Player.

 

[rockafella]

Looking at the reviews, that's more than a bit on the slow side for a modern SSD.
http://www.legitreviews.com/article/1662/6/

Even in RAID0, those aren't going to match the speed of a current-gen or a next generation SSD. A Crucial M4, Samsung 840 Pro (or 830 if you can still find them), or high-end Sandforce drive should easily rock better numbers w/o having to deal with firmware-raid arrays and the reliability potentials involved. You might also look at firmware updates for your Pyro's as well. Firmware updates can make a DRAMATIC difference in performance.

ALong with that, let's keep in mind the utterly different platforms that you are comparing the virtualization on:

Mac w/ i7 and faster SSD running Parallel's Desktop
PC w/ older Phenom II and RAID0 of slower SSDs running VMWare Player.

Hmm, that is bad. That explains why I "only" get 330 sustained read speed from the array! I think I had been set on getting another one but then I saw the sale and perhaps was told they were just as good or something. (or maybe I knew going in they weren't the best available but didn't care, since I hadn't had any SSD before that)

I'll check out the firmware, and consider just upgrading or at least adding a newer SSD for the VM.

I realize they are different platforms (except I did also try Parallels Workstation on the PC, which did perform noticeably better than VMWare Player for me). I guess it just bugs me that my desktop works great still in the native/host OS, but I just can't get the VMs to feel the same. Ah well! At least upgrading the SSD is easy and is something I'd do anyway eventually assuming I plan on keeping my desktop operational overall and up to date.

Thanks for the feedback and finding that report!

 

[fantabulous]

One other food for thought:

I've found that raid chunk sizes had quite the effect for different VM workloads. When I used a 16k chunk size with a simple spinning platter RAID 10 (linux mdadm), I had adequate small file performance, but terrible loading times in games. I opted to simply split up the array rather than test with larger chunk sizes.