AMD with 81.5% of marketshare

[chavvdarrr]

I view a server as something which will be running for 3 years, or 4 in some cases. In 3 years time the state of the art in terms of virtualization software will have moved on dramatically (e.g. minikernel hypervisors relying on next wave CPU virtualization extensions), so considering a certain amount of "future proofing" is no bad thing.

Sure. But I bet that (as usual) only 2nd incarnations of VT?pacifica will be "done right" , and right now VT lacks some features which Pacifica will have... so should we wait for pacifica? We wanted the job done now... and i do plan to percuade accounting department that we MUST get Xeons with VT no matter the price , these are our first x86 VM servers (there are several IBM "big irons") and we more or less do a testride session and in same time patching holes ...

Regarding host OS costs : Did you notice the announcement from Microsoft a month or so ago where they said that buying W2K3 Enterprise Edition gives one rights to 4 additional instances of that OS for free inside virtual machines on the same host? (not specifically virtual server either). Oh, and their other announcement that you only have to pay Windows OS license fees for running VMs, not for every single VM sitting on a hdd inactive. (I guess this one is only for enterprise customers though).

cheers, Skazz[/QUOTE]
No, I didn't know about buying 1 EE and having right for 4 more instances. But EE costs 4-5x the Standart server... and why mess with /PAE if 2003R2 x64 SS is cheaper? Besides most VMs we run are Linux... just don't ask me why the host is Linux... (easy guess - big boss wants it this way )

 

[KimB]

Sure. But I bet that (as usual) only 2nd incarnations of VT?pacifica will be "done right" , and right now VT lacks some features which Pacifica will have... so should we wait for pacifica? We wanted the job done now... and i do plan to percuade accounting department that we MUST get Xeons with VT no matter the price , these are our first x86 VM servers (there are several IBM "big irons") and we more or less do a testride session and in same time patching holes ...

Well, what can you do now on VT that you can't do on an Opteron? Because the performance is just a hell of a lot higher in the server benchmarks that I've seen.

 

[chavvdarrr]

Well, what can you do now on VT that you can't do on an Opteron? Because the performance is just a hell of a lot higher in the server benchmarks that I've seen.

well... right now - nothing. But when/if Xen/Vmware/MS make use of it... after all we need to evaluate both techs - there will be more and more VM-running servers in next years

 

[L233]

Dunno the statistics, but I don't know a single person in the last few years who bought an Intel CPU in retail. Not one.

Actually, I did. There were a few times when Intel was more attractive for enthusiasts not afraid of overclocking, e.g. the P4 Northwood.

The 2.4 GHz Northwood was an almost guaranteed 3.0 GHz overclocking success. All you needed to do was to up the FSB to 250 MHz, adjust the ratio to 5:4 and up the core voltage slightly. The beauty of this was that you didn't even need expensive RAM, just stock 400Mhz DDR. Plus, you got Hyperthreading and a more powerful SSE2 implementation. Another important factor for me was that it was much easier to find good mainboards that were passively cooled.

Best bang for the buck the since the heyday of the Celeron.

 

[Fox5]

Actually, I did. There were a few times when Intel was more attractive for enthusiasts not afraid of overclocking, e.g. the P4 Northwood.

The 2.4 GHz Northwood was an almost guaranteed 3.0 GHz overclocking success. All you needed to do was to up the FSB to 250 MHz, adjust the ratio to 5:4 and up the core voltage slightly. The beauty of this was that you didn't even need expensive RAM, just stock 400Mhz DDR. Plus, you got Hyperthreading and a more powerful SSE2 implementation. Another important factor for me was that it was much easier to find good mainboards that were passively cooled.

Best bang for the buck the since the heyday of the Celeron.

I'd still say bang for the buck would have been a $60 mobile AXP overclocked to 2.4ghz, performing roughly on par with a 2.8ghz to 3.0ghz p4. Oh, and that 3ghz 5:4 ratio p4 wouldn't have performed as well as a normal p4 due to the divider, you were probably better getting one of the 533fsb northwoods and overclocking it to 800fsb. (the beauty of the mobile axps was that they were cheap and completely unlocked, so they didn't have to suffer a performance penalty by having memory out of sync)

 

[L233]

I'd still say bang for the buck would have been a $60 mobile AXP overclocked to 2.4ghz, performing roughly on par with a 2.8ghz to 3.0ghz p4. Oh, and that 3ghz 5:4 ratio p4 wouldn't have performed as well as a normal p4 due to the divider, you were probably better getting one of the 533fsb northwoods and overclocking it to 800fsb. (the beauty of the mobile axps was that they were cheap and completely unlocked, so they didn't have to suffer a performance penalty by having memory out of sync)

The performance loss from going with a 5:4 divider was nearly non-existant. RAM still ran at 400 MHz. And going from FSB 533 -> 800 is pretty extreme, it's a 50% overclock, i.e. 3.6 GHz (or 3.4 GHz with a Northwood 2.26 GHz).

Probably wouldn't have worked with aircooling (if at all) and you'd have had to increase vcore quite a lot.. remember Sudden Northwood Death Syndrome?

IMO, the 2.4 GHz FSB800 Northwood was pretty perfect, especially since I could buy cheapo RAM and the chances that the CPU would not go to 3 GHz were pretty slim.

I wasn't aware of that $60 mobile XP and if it was really such a great and _reliable_ overlocker, then this would have indeed been a great deal.

 

[Fox5]

The performance loss from going with a 5:4 divider was nearly non-existant. RAM still ran at 400 MHz. And going from FSB 533 -> 800 is pretty extreme, it's a 50% overclock, i.e. 3.6 GHz (or 3.4 GHz with a Northwood 2.26 GHz).

Probably wouldn't have worked with aircooling (if at all) and you'd have had to increase vcore quite a lot.. remember Sudden Northwood Death Syndrome?

IMO, the 2.4 GHz FSB800 Northwood was pretty perfect, especially since I could buy cheapo RAM and the chances that the CPU would not go to 3 GHz were pretty slim.

I wasn't aware of that $60 mobile XP and if it was really such a great and _reliable_ overlocker, then this would have indeed been a great deal.

Well, I don't know what kind of performance hit the northwoods got from asynchronous memory, but their more robust chipsets and quad pumped fsb probably lessened the hit. I wouldn't be surprised to see that 3ghz performing like a 2.8ghz with the divider though. I know that on athlon xps, asynchronous memory had a horrible performance hit that basically made overclocking worthless. The memory latency increased so much that there was a good chance even if you overclocked several hundred mhz, the net performance would still be less.

Anyhow, I think the mobile athlon xps hit the market about a year before the athlon 64 came out, and peaked in popularity about 6 months before the athlon 64 came out. They were just normal Bartons (well, some were tbreds, but most were bartons to make up for the 133fsb the mobile cpus used by default) but were cherrypicked and able to operate at 1.35v or 1.45v instead of the standard athlon vcore of 1.65v. For instance, I'm currently running a mobile barton at 200fsb 2.2ghz at 1.45v. When taken up to 1.65v, it was pretty typical to be able to get either 220fsb at 2.2ghz or 200fsb at 2.4ghz. In some extreme cases, people were hitting 2.6 to 2.7ghz, and upwards of 235mhz on the fsb. And since these were all multi unlocked for cool n' quiet (and amd had just locked multis on the desktop athlons, which overclocked worse anyway), they were quite nice to have. And despite being AMD's best silicon (though I think the 3200+ was only like $140 at this point in time) they ranged in price from $60 to $100 since they were priced at what their speed rating was. While it was disappointing to see the Athlon 64s come out shortly after and blow everything away, it was nice to have a system that could compete with the top P4 cpu at the time for less than the cost of the P4 cpu by itself. (minus the price of the video card, though there were quite a few $50 nforce2 boards with integrated gf4mx graphics) Man, I wish those days of super cheap overclocking, though something similar can be attained currently by picking up a sempron and a socket 754 board, but overall it's more expensive and doesn't come as close to the top performance.

 

[L233]

I did some before/after benchmarking back then and the performance increase from going to 2.4 to 3.0 was pretty much exactly what was to be expected.

I can remember the discussions back when the first CL3 DDR 500 modules appeared and people preferred running a 1000 FSB with CL2 DDR 400 RAM at 5:4 because it was faster than CL3 DDR 500 at 1:1. So if there is any performance drop from using an async 5:4 divider it must be smaller than that the performance drop from CL2 -> CL3. We're probably talking about something in the ballpark of 5% here.

 

[Fox5]

I did some before/after benchmarking back then and the performance increase from going to 2.4 to 3.0 was pretty much exactly what was to be expected.

I can remember the discussions back when the first CL3 DDR 500 modules appeared and people preferred running a 1000 FSB with CL2 DDR 400 RAM at 5:4 because it was faster than CL3 DDR 500 at 1:1. So if there is any performance drop from using an async 5:4 divider it must be smaller than that the performance drop from CL2 -> CL3. We're probably talking about something in the ballpark of 5% here.

That'd be 150mhz, which is basically a whole speed grade drop. (which is virtually nothing, but still that would be quite a large difference in cost if you were to buy the faster cpu)

 

[Frank]

Both Dell and HP sell AMD nowadays (although you still have to look hard with Dell), because people ask for them and you just get a lot more computer for the same money.

 

[dizietsma]

3dmark has always favored Intel for the 3dmark score, I'd imagine they're able to eek slightly more performance out of their AGP/PCI-Express buses than AMD mobos do.

Incorrect, the last Intel processor to be at the top of 3dmark 2001 was the Intel P4 Northwood on 875 chipset. Since then it has been AMD K8 all the way. It's only since Dothan and now Yonah has the balance been addressed.


I'm not sure where you got your Pentium M's were supposed to hit 3Ghz on air from and some cap out at 2.2 .. most Pentium M's on air hit the 2.5-2.7 level when ones with the corrct multipliers are chosen for use with appropriate motherboards and / or 479 converters.

 

[Fox5]

Incorrect, the last Intel processor to be at the top of 3dmark 2001 was the Intel P4 Northwood on 875 chipset. Since then it has been AMD K8 all the way. It's only since Dothan and now Yonah has the balance been addressed.


I'm not sure where you got your Pentium M's were supposed to hit 3Ghz on air from and some cap out at 2.2 .. most Pentium M's on air hit the 2.5-2.7 level when ones with the corrct multipliers are chosen for use with appropriate motherboards and / or 479 converters.

3dmark2001 is quite old.....and very cpu limited.
I was talking more of the latest 3dmarks, which tend to be gpu limited, where the Intel platforms will usually slightly edge out the AMD platforms, despite having the same gpu and a weaker cpu.

As far as the P-M, there seemed to be quite a few people claiming they had one that hit 3ghz, yet in nearly every professional review I've seen the P-M's cap out around 2.2 to 2.3ghz.

 

[dizietsma]

As far as the P-M, there seemed to be quite a few people claiming they had one that hit 3ghz, yet in nearly every professional review I've seen the P-M's cap out around 2.2 to 2.3ghz.

A lot of the reviewers used the 855 chipset, when used with the old 875 chipset the average air cooled Dothan was around the 2.5 to 2.7 mark when the way to boot at 200Mhz FSB was worked out.

 

[Dandybear10]

It doesn't really matter that the Conroe isn't out yet because what is known is that it performs extremely well, will be out in a matter of a few months and the only update for the Athlon 64 is a new socket with DDR2 support until at least next year. Price is a non factor as well, the high end Conroe (E6700) will retail for about $530 while the next down will go for a mere $316 and these things overclock like crazy. So for about $209 you will be able to get noticably better performance than an FX-60.

Would be nice.