After expert Andrew Kutz chose Intel over AMD for the best new processor for virtualization, some readers had differing opinions -- and good reasons as to why. Read on for what Andrew Kutz said, and how other users responded.
Andrew Kutz:
Currently Intel's line of 64-bit chips that support Intel Virtualization Technology (VT) (Vanderpool/Silverdale) are the best choice for use with virtualization software. This includes their laptop and desktop chips, the Core 2 series and their Xeon serious server chips. AMD has their own chip-virtualization technology called AMD-V (Pacifica), but it is currently only offered (according to the AMD Website) in their Opteron series workstation/server chips.
Even though AMD offers chips with similar functionality, I recommend Intel because although both VT and V are relatively new, Intel-VT has been out longer and is likely to have more support from virtualization software vendors.
I do not get into many 32-bit vs. 64-bit debates because I usually do not think the extra width makes a lick of difference. However, it is important when designing a new virtualization infrastructure to go with 64-bit chips, because a 64-bit enabled host server is required to host 64-bit Virtual Machines (VMs). Since VMs are often used as test beds for new software or developing applications, you will want to be able to easily deploy a 64-bit version of an OS so you can test a 3rd party application, or your own application, on it.
Your responses...
GEdwards: The current Intel chip may indeed be better than the AMD Rev E Opteron but I would still prefer the AMD dual core Rev F chip. As intensive as memory handling is for smooth and successful virtualizations, AMD's ability to 'virtualize' the chips memory component directly on the die as opposed to no memory vt for intels bus limited architecture puts AMD way out front.
Also, consider the two chips's road maps. It does not look like Intel will abandon the memory bus (FSB) architecture and it's limitation. Woodcrest showed how Intel tried to clear the single memory bus architecture problem by using two buses. Will quad boxes have a four bus architecture? Either way that still leaves the enhanced features of on-chip virtualization for memory lacking for Intel chips. Currently VMware passes VM guest calls to hardware that should occur on ring 0 of a chip down through outer rings to ring 0. With both AMD and Intel virtualization technology, these can be localized at ring 0. The difference as mentioned will be that with memory processing for Intel out on a bus, their chips cannot virtualize memory processes efficiently where AMD has their memory processing on each chip. Given the impact that memory handling has on virtualization this is a deal maker for AMD chips.
AMD finally moving to DDR2 is also another benefit.
AMD may or may not have forseen the choke point as each CPU directly addresses its memory without a bus choke point. You won't hear about how fast (wide) the Front Side Bus is for AMD because that is not even an issue.
When Intel presents its quad core chips this will really be two dual core stuck together. AMDs quad, out a few months later, will, one - be a true quad core, and two - (perhaps more importantly) plug right into their latest sockets on their current chipset boards!
For virtualization build outs, each major revision Intel undertakes makes VMotion compatibility next to impossible. AMD touts their integrated road map as one you can live and grow with.
I like Intel a lot and am grateful both major chip firms are US firms. But truth be told Intel pulled the same mistake Microsoft (MS) did when Oracle got the jump on them as the default internet data repository. Like MS Intel is playing catch up and is throwing good products into the market as fast as it can but with little forthought as to an integrated roadmap.
And yes, 64-bit is the way to go in that it boosts 32-bit application performance and gives you the ability to have 64-bit guests.
I have Intel and AMD chipset VMware ESX implementations and prefer the AMD for their roadmaps integration efforts, power and heat reduction efforts, and their focus on making virtualization a central aspect of their forward looking strategy. But that is just today. Ask me again tomorrow..
Editor's Note: Andrew Kutz responded. Here's what he had to say in his own defense.
One IT pro: I can hardly believe [Andrew Kutz's] response. Xeons in any form have a serious crippling disability. It's called a front side bus, and they are mostly 400 MHz, though Intel claims 1.267 GHz. Try and find one. 667 MHz variants are a bit more common, but shockingly there is a lot of new 400 MHz stuff floating around. Now come on Mr. Kutz, 400 MHz versus 6.4 Gigs? Be serious! And Van Der Pool virtual extensions? Pacifica edges Intel on that one as well. Then there is the lab tests to ponder. Adding 22 MB to the cache only goes so far, great for two way, but look at your four way results again. In mine, the Xeon's lag the Opterons, by a very wide margin, and the AMD 65 nano stuff is not even mainstream or widely available yet. Intel needs to giddyup with fixing the FSB (front side bus), or carry on hoping the "Itanic" (Itanium) is going to come to their rescue and these chips will fill the gap in the meantime. Xeons suck. It's official. The "Itanic" is the greatest example in the commercial world of throwing good money after rotten bad piles of money. How much has the Itanic project cost so far? $26 billion plus? Ahem -- wakey wakey!
I think you have gushed on the fact that your new laptop running a Centrino Duo is so fast and have extrapolated the same results to the Xeons over the Opterons. Sadly, the truth is, there is a performance jump for some type of computing systems streaming video - heavy database paints a different picture. Again I have to say that Xeons just suck.
Sound off! Who's right? Andrew Kutz or his critics? Email me your response and I'll include it on this page.
