[sameen]

if u have used AMD Phenom and Intel then what would u preferr AMD or INTEL ?

This post has been edited by sameen: 17 February 2008 - 02:14 AM

[jcarle]

There's no reason to buy ANY AMD today. Intel is where it's at.

[puntoMX]

sameen, on Feb 17 2008, 02:11 AM, said:

If u have used AMD Phenom and Intel then what would u preferr AMD or INTEL ?

The Phenom performs well for it price, however, you would want to wait till the new "stepping" comes out. The Phenoms you find now have a big bug inside that could slow down the system by 20 to 30% in some cases. If you don’t overclock and you don´t want a Quad cored CPU, AMD would be the best choice.

If you overclock, then go with Intel, if you want a Quad cored CPU then its Intel again.

Don’t pay attention to people who have a short vision like jcarle; he’s an Intel freak for the last years and doesn’t know more then that.

jcarle, on Feb 17 2008, 02:06 PM, said:

There's no reason to buy ANY AMD today. Intel is where it's at.

Err… Right…

[cluberti]

As always, as a HEAVY virtual machine user, I have to respectfully disagree with jcarle on this. Intel is NOT where it's at .

[jcarle]

I feel Noah. Yes, the one with the Ark.

I've always stuck by Intel for a reason. Performance has fluctuated over the years, granted, but Intel CPUs have ALWAYS been stable. Sometimes they've fallen behind but they always come fighting back and stronger too.

Yes, they've had their share of problems (floating point anyone? deep pipeline combined with low speed first generation pentium fours?) but the problems associated with Intel have always been few and far between. AMD's had some pretty serious issues of their own. The performance for Intel has always been rock solid. AMD's been all over the place.

The Intel / AMD war will rage on as it always has for the rest of time. Some are hardcore AMD fanboys, some are hardcore Intel fanboys (such as myself). I've seen so many dead CPUs go through my hands which were AMD (Especially in the Athlon XP series) that I've been skeptical ever since. No to mention that I ran into really bizarre problems that I could only be replicated while running on an AMD system.

People have often quoted as price being the main reason to go for AMD. You can buy an Intel motherboard and an Intel CPU (Celeron, for example) for the same price or less then you would from AMD. Perhaps AMD would have faired better IMHO if it had improved it's compatibility between motherboards when new CPUs were released or spent a little more time trying to bring down their heat generation.

I guess I just have to try to tone down my Intel fanboyism and try to accept that for some people, AMD suits them better even if I can't see how.

[cluberti]

jcarle, on Feb 18 2008, 10:22 AM, said:

I guess I just have to try to tone down my Intel fanboyism and try to accept that for some people, AMD suits them better even if I can't see how.

I think that's the true tenet of fanboi'ism of any kind (Intel/AMD, Apple/Microsoft, Linux/rest of the world's OSes) - you're too close to the chalkboard to see any position other than the one you've sidled up to and is right in front of your nose. Technically, there really isn't anything wrong with that either - what you have served you well, there's no compelling reason to do anything different, IMO.

I know we'll never agree, but judging AMD on anything prior to the later K7 models (when they finally were big enough and had the R&D budget to "get it right", so to speak) really does limit your ability to judge favorably. I'll stick with my Opterons and Tyan motherboards for virtualization and database performance, as I have tried Intel in this role, and the Xeons honestly just can't keep up.

[jcarle]

Have you looked at the Xeon 7300 series with the new APIC Task Programmable Register?

[cluberti]

jcarle, on Feb 18 2008, 11:36 AM, said:

Have you looked at the Xeon 7300 series with the new APIC Task Programmable Register?

Yes. Intel still hasn't fixed their I/O bottleneck from the CPU to system memory, just moved it - the DHSI, which was supposed to fix the bottleneck, simply moves it from the CPU to the northbridge. All memory and I/O from CPU to memory subsystem has to go through the northbridge, which under heavy load becomes more like a bike path than a superhighway (single point of convergence). The AMD chips and chipset have no such limitation, as each processor can address directly through the hypertransport any memory bank (and two processors can access each bank simultaneously, although at a slight performance hit if the traffic gets hot and heavy).

Intel did introduce VT-d to do a direct bypass of the CPU for hardware I/O in virtualization, which AMD won't release until the next chipset release later this year, and that was pretty cool. Not enough to sway, but definitely a cool addition.

If you don't put a heavy load on your box, you're good to go (better even) with the Xeons. If you run lots of VMs doing a decent amount of memory I/O, the Xeons quickly fall behind the Opterons.

[jcarle]

cluberti, on Feb 18 2008, 12:53 PM, said:

Intel still hasn't fixed their I/O bottleneck from the CPU to system memory, just moved it - the DHSI, which was supposed to fix the bottleneck, simply moves it from the CPU to the northbridge. All memory and I/O from CPU to memory subsystem has to go through the northbridge, which under heavy load becomes more like a bike path than a superhighway (single point of convergence).

I've read that Intel's upcoming Nehalem and Bloomfield (due in Q1 2009) architectures are supposed to resolve this. On-die memory controllers with a 45nm process.

[cluberti]

That would be good, but I need servers now that work. I can't wait until 2009 for them to fix a problem they've had since the P4 released!

[puntoMX]

jcarle, on Feb 18 2008, 10:22 AM, said:

but Intel CPUs have ALWAYS been stable... ...Yes, they've had their share of problems (floating point anyone? deep pipeline combined with low speed first generation pentium fours?) but the problems associated with Intel have always been few and far between. AMD's had some pretty serious issues of their own. The performance for Intel has always been rock solid. AMD's been all over the place.

The Core2 had a bug as well, not as fatal as the AMD bug but they had one: kb936357. Intel was so lucky that they could "fix" it with a BIOS update.

Just see things objective .

cluberti, on Feb 18 2008, 11:53 AM, said:

If you don't put a heavy load on your box, you're good to go (better even) with the Xeons. If you run lots of VMs doing a decent amount of memory I/O, the Xeons quickly fall behind the Opterons.

I forgot about that, VMware cursed the Intel chips for it too...

[nmX.Memnoch]

Intel's next architecture change is finally going to get rid of the shared FSB limitation. It's about time they get onboard with a point-to-point interface like hypertransport. I think they're doing it in a phased approach. The first change will be to include Common System Interconnect (CSI)--or QuickPath, or whatever they finally decide to call it--and then the next change will be to move the memory controller on the CPUs.

And if I'm not mistaken, the Itanium CPUs are actually going to get CSI first.

[cluberti]

Sweet - hopefully they'll have a good implementation, and drive competition up and costs down .

Too bad it took them 4 years to catch up.

[jcarle]

cluberti, on Feb 21 2008, 10:03 PM, said:

Sweet - hopefully they'll have a good implementation, and drive competition up and costs down .

Too bad it took them 4 years to catch up.

Sure, but it may take that long for AMD to catch up to the whole Core 2 show.

[cluberti]

jcarle, on Feb 22 2008, 04:10 PM, said:

Sure, but it may take that long for AMD to catch up to the whole Core 2 show.

Well, considering the current generation of Intel "quad-cores" are really dual-die dual-core processors that have to leave the CPU to speak to the other die taking on a massive performance penalty to do so, I don't think I want them emulating that . AMD's quad-core design is a true single-die quad-core, meaning I won't be spending hundreds of milliseconds every time the cores need to sync. And I'm not buying Core2 or Phenom (or any other desktop technology) - and if I'm paying the $$ for a Xeon, I'd like it to be at least a good design. Yes they were the first to release 4 cores in a package, and yes dual-die dual-core allows faster CPUs in the same package (less heat/power per core), but for actual real-world implementation the design is just not that good in my experience, especially as a heavy VM and database server guy. If I'm buying a server and I want a quad-core chip, I buy an AMD because it's an actual quad-core, not a dual-die dual-core that has latency built right into every operation that requires intra-cpu communication. Not to mention the hypertransport issues above.

[puntoMX]

Before Nehalem and Bloomfield will be released, we have the Dunnington. Sun leaked some info about the Dunnington and can be found here. Specs of the Dunnington were released back in October 2007 by the way...

[cluberti]

puntoMX, on Feb 25 2008, 01:01 PM, said:

Before Nehalem and Bloomfield will be released, we have the Dunnington. Sun leaked some info about the Dunnington and can be found here. Specs of the Dunnington were released back in October 2007 by the way...

So now it's three-die dual-core. I'll just wait until Nehalem releases with true single-die quad-core before I try another Intel Xeon.

[nmX.Memnoch]

cluberti, on Feb 25 2008, 01:31 PM, said:

So now it's three-die dual-core.

Yes, but they added an L3 cache that will be shared by all three die. They will use the L3 for communcation instead of the FSB. That should help with the FSB latencies.

I do agree though...they should just get their native quad-core design out the door.

I did some more research on CSI as well (which as it turns out, the official name is indeed QuickPath Interconnect). Both it and on-die memory controllers (OMC) will be available at the same time, as opposed to a staggered release like a originally thought.

I'm not sure how much I really like the idea of an OMC though. Unless they take a different course of action than AMD did, you'll have to match the motherboard and CPU with the RAM. Right now I can take a P35 DDR2 based board and through (pretty much) any LGA775 CPU from Pentium 4 up to Core 2 Xtreme on the board. AMD, of course, has the same problem which is one of the reasons they haven't implemented DDR3 yet. I'm going to wait and see how Intel handles it before I pass judgement though.

[cluberti]

nmX.Memnoch, on Feb 25 2008, 03:05 PM, said:

I'm not sure how much I really like the idea of an OMC though. Unless they take a different course of action than AMD did, you'll have to match the motherboard and CPU with the RAM. Right now I can take a P35 DDR2 based board and through (pretty much) any LGA775 CPU from Pentium 4 up to Core 2 Xtreme on the board.

True, but most newer memory controllers can support multiple types of memory. I would have liked to see the memory access controller be a bit more flexible, but I don't know the costs in that, so I can't speak to why this isn't the case with AMD processors. Again, when nehalem is a go, and the OMC is a feature, I'll definitely be checking it out. Hopefully the implementation is good, and gives us the performance back.

[Access Denied]

I have a question since you guys seem to know alot more about both than I. I have been thinking about getting an Intel E6600 Dual Core. Is there a better AMD dual core for the same price without an FSB bottleneck/slowdown? Many thanks as I learned more about cpu's reading this thread than ever knew about AMD/Intel, lol. My old P4 3Ghz with HT still runs nice, but its time for an upgrade to something faster.