Xeon X5570 "Nehalem" benchmarked.

[yomamafor1]

Look guys take it easy on keithlm all he had was the server chips,
now all of that gone he has nothing else.
So grasping at straws and hoping the worst for intel is better then
excepting the complete ass pounding AMD is getting rite now.

Although I know I'm not supposed to flame other users....

...but damn that's a good summary you have there 😀.

 

[gallag]

Its funny, from AMD's senior vice president on the release of Nehalem:

"After nearly six years of Intel telling customers that the AMD Opteron processor architecture was the wrong answer, this week our competitor has finally delivered “Nehalem” ― which some might call a copy, at least as far as the architecture is concerned."

We already know Nehalem architecture is basically an Opteron architecture. Wow, thats a technological breakthrough for big Intel.

If nehalem and opteron are the same architecture why is nehelam about 50%faster? Why can people now replace quad systems with duels and keep the same processing power? Why are companys looking forward to halving there per socket software bills?

 

[keithlm]

Actually, there is a complete guarantee. It's called Intel guaranteeing the chip. That statement of yours would be like if I claimed that all SSE extensions should not be used on AMD chips because they might hurt stability. It's complete garbage and it is utterly ridiculous for you to be spreading this FUD.

You Intel people just keep telling yourselves that. Perhaps if you keep it as a mantra you can actually make it true. At least in your own minds.

But in that little place I like to call "reality" this wouldn't even be an issue. Some junior admin in a company would suggest leaving these features enabled and the senior staff would give it a resounding "Not going to happen for at least another two years." (That is: If it gets that much of a positive response.)

Guaranteeing a chip is not going to help out a company that loses money because of a failure on an important 24/7 system. Intel may guarantee the chip; they also have clauses that do not leave them liable; caveat emptor.

But as I said you guys are probably correct... this issue would never come up on a serious mission critical system because nobody will buy this chip for one of those systems until the 3rd or 4th publicly available stepping. But regardless... this is going to be true on many non-mission critial systems. It all comes down to whether it is an important system used by many people or just a system that won't be harmed by a reboot. So on a few test systems it may be enabled for a few years. (And then disabled after it is found that it doesn't really provide all the performance that people on many forums actually believe.)

Luckily I am no longer an admin; I hate the whining users; and I like being a developer much better. But 15 or 20 years ago when I was an admin I can guarantee these toys would not have been used in a production system. Sorry if some of you don't have enough experience to actually realize what I am saying is true; but that doesn't make it "FUD".

EDIT: But then again, I have to say that luckily most extreme mission critical systems would never be put onto a a Wintel box of any type.

 

[yomamafor1]

Its funny, from AMD's senior vice president on the release of Nehalem:

"After nearly six years of Intel telling customers that the AMD Opteron processor architecture was the wrong answer, this week our competitor has finally delivered “Nehalem” ― which some might call a copy, at least as far as the architecture is concerned."

We already know Nehalem architecture is basically an Opteron architecture. Wow, thats a technological breakthrough for big Intel.

:lol: :lol: :lol: :lol:

You're kidding me right? Having IMC and native quad is a "copy of Opteron"? I thought after the management changed from spinster Hector Ruiz and Randy Allen to Dirk Meyer, AMD will be a much better company. Guess I'm wrong.

Well AMD also said that they want to change the way battery life is measure on the laptop, and the first program they suggested is 3DMark06.

 

[dattimr]

But of course any admin would deploy a TLB'ed Opteron right away, huh.

 

[exit2dos]

Ummmm....I guess you missed the part where Intel had to pay Transmeta 150 million dollars huh. Intel didnt allow Transmeta anything, they had no choice.

You are aware that AMD violated the same patents in question?

Seems Transmeta didn't sue AMD after AMD purchased nearly 10M shares of Transmeta. Coincidentally, AMD also had to have Transmeta develop the basics of Hypertransport for them.

So, please stop with the FUD. Everyone is entitled to a preference, but not an agenda.

 

[keithlm]

But of course any admin would deploy a TLB'ed Opteron right away, huh.

You should never worry about the shark you can see and already know how to deal with.

Besides... if my memory serves me correctly the manufacturer revealed that problem as soon as possible and also provided a fix; making it a non-issue. (Except for use in forums as flame bait.) New opterons don't have the problem anyway; making it even less of a non-issue.


======================================================
EDIT: Actually for all the "It would be stupid to disable HT/Turbo" people:

An admin that has older Xeons that have hyperthreading enabled might actually advocate buying these new Xeons and leaving HT/Turbo enabled.

BUT

An admin that has older Xeons that have hyperthreading disabled will probably not advocate buying these new Xeons. If they ever do get some systems with these chips then they will disable HT/Turbo immediately.

So the only real question is: What percentage of admins fall into each category? (I've already told you the answer.)

 

[fazers_on_stun]

You mean people like your sacred AMDzone The_Ghost, who deletes everything that's not "AMD propaganda"? Or by "qualified and experienced", you mean yourself?

Seriously, you guys at AMDzone needs to learn the meaning of "reality". Nehalem operating at 200+ TDP? LOL!!

Speaking of which, it seems Sci's AMD bias was too mild to pass muster, and so UAEZone went and made Abinstein a moderator. Somebody must have given him a scorecard to tell which posters were OK to attack and which posters were actually pro-AMD and thus not OK to attack. A required skill for mods there is to be able to aim the flame and not scorch the shorts, even though most of them appear a bit cross-eyed with Intel-envy.

Interestingly enough, Keith here uses the same "toy marketing" disparaging labels for hyperthreading and turboboost that Abinstein uses. Perhaps there's a modship waiting for him over there before too long .

 

[keithlm]

Interestingly enough, Keith here uses the same "toy marketing" disparaging labels for hyperthreading and turboboost that Abinstein uses. Perhaps there's a modship waiting for him over there before too long .

No thanks; that is too much like work.

 

[fazers_on_stun]

You Intel people just keep telling yourselves that. Perhaps if you keep it as a mantra you can actually make it true. At least in your own minds..

Well if it's alright with you, we'll just wait for the 2P marketshare reports to see what actual system buyers think, mkay?

Luckily I am no longer an admin; I hate the whining users

Now there's a great big surprise - LOL!

But 15 or 20 years ago when I was an admin I can guarantee these toys would not have been used in a production system. Sorry if some of you don't have enough experience to actually realize what I am saying is true; but that doesn't make it "FUD".

Hmm, I must have been asleep and missed the announcement back in 1994 or 1989, about hyperthreading or turboboost being used in Intel CPUs back then. 😀

 

[yomamafor1]

You should never worry about the shark you can see and already know how to deal with.

Besides... if my memory serves me correctly the manufacturer revealed that problem as soon as possible and also provided a fix; making it a non-issue. (Except for use in forums as flame bait.) New opterons don't have the problem anyway; making it even less of a non-issue.


======================================================
EDIT: Actually for all the "It would be stupid to disable HT/Turbo" people:

An admin that has older Xeons that have hyperthreading enabled might actually advocate buying these new Xeons and leaving HT/Turbo enabled.

BUT

An admin that has older Xeons that have hyperthreading disabled will probably not advocate buying these new Xeons. If they ever do get some systems with these chips then they will disable HT/Turbo immediately.

So the only real question is: What percentage of admins fall into each category? (I've already told you the answer.)

Actually, the question becomes..

...which admin is lazy and stupid enough to not do their homeworks before spending hundreds of thousands of investment on new server equipments? A lot of benchmarks are already been done, and companies can usually just look them up and compare the differences.

Maybe I should also make the assumption that any server administrators are going to turn off Shanghai's split power plane, because that's going to slow down the processor. Does that make sense? Yep, that's how ridiculous you're sounding right now.

 

[yomamafor1]

Speaking of which, it seems Sci's AMD bias was too mild to pass muster, and so UAEZone went and made Abinstein a moderator. Somebody must have given him a scorecard to tell which posters were OK to attack and which posters were actually pro-AMD and thus not OK to attack. A required skill for mods there is to be able to aim the flame and not scorch the shorts, even though most of them appear a bit cross-eyed with Intel-envy.

Interestingly enough, Keith here uses the same "toy marketing" disparaging labels for hyperthreading and turboboost that Abinstein uses. Perhaps there's a modship waiting for him over there before too long .

You are kidding me... Abinstein a moderator?

I thought the bias Ghost posses is horrible enough.... now Abinstein? No wonder there are more bots visiting the sites than users :lol: :lol: :lol:

 

[jed]

Everyone in this forum wheather you like AMD or not have
said phenom II's are actually good chips, and argue with
anyone claiming their great chips.
Now Intel has a few great Xeon chips out, and all i see from
the AMD boys is if you turn everything off it's not that much
better then AMD's best, which is complete nonsence.
No one buy these kind of chips to disable and cripple them.
Now the other AMD cheerleaders one can get to stop talking
all of a sudden have nothing to said at all, i think all of this is
pretty funny. just my 2cents.

 

[someguy7]

Even with HT and turbo off its still faster by a large margin.

With the guys its always the same. AMD=great/user experience....

Intel=spintel/marketing gimmick/unstable/Intel cpus are made to exel in benchies but suck in real world/The review sites are paid off by Intel.

Nobody cares about overclocking(when the AMD chips had no headroom). Now look at how well the PII's overclock.

It's always gotta be something with them. DOUBLE CHEESEBURGERSSS!!!!

 

[keithlm]

Well if it's alright with you, we'll just wait for the 2P marketshare reports to see what actual system buyers think, mkay?

But 15 or 20 years ago when I was an admin I can guarantee these toys would not have been used in a production system. Sorry if some of you don't have enough experience to actually realize what I am saying is true; but that doesn't make it "FUD".

Hmm, I must have been asleep and missed the announcement back in 1994 or 1989, about hyperthreading or turboboost being used in Intel CPUs back then. 😀

You are correct. We only have to wait for the 2P and 4P and 8P results. (You know... the small machines that don't really do any serious work anyway.)

On the other hand your posts won't make a bit of difference when it comes to the HT issue. This toy will be disabled by most experienced administrators.

Pop quiz: How many of you people that keep posting about how an admin would stupid to disable HT have actually worked in an admin job? And I don't mean running a single server.

I must admit.... back in the day I only had 50 servers that serviced about 1200 workstations all over the country. Actually since I wrote the admin manual used by everybody in the country and decided how all the servers and workstations would be configured that number should actually be more like about 600 servers and 15,000 workstations. But hey... apparently that means I know nothing about how business and government work... since I only did that job for about 5 years. And since it started in about 1989... that must mean I'm out of touch with reality. (You wish.)

I actually know how these decisions are made... it takes about 6 months to get somebody to change UNIX kernel configurations to optimize a system. Even when the person telling you to do it previously worked specifically in a VERY high paying job advising people how to optimize their systems for database use. But hey... you people without any real world experience obviously are going to change how these things work instantly in the business and government world because of Intel marketing. (Only in your dreams.)

Delude yourselves all you want to. The toy marketing tools will be disabled in most real world situations. You can all claim otherwise; but it won't really matter in the real world... only on this forum. And yes... I will read all the replies and find it amusing.

ADDITION: And yes... I admit... when I first started doing admin work I would have been one of the people that would have advocated using these types of things. Over time that was squashed out of me. I kicked and screamed... but I finally learned. But after actually having experience at that job AND ALSO working as a consultant for Oracle for years... I now would no longer be stupid enough to do so. It won't happen in most situations regardless of how many of you posters want to delude yourselves. Have fun living in the fantasy world.

 

[jimmysmitty]

Its funny, from AMD's senior vice president on the release of Nehalem:

"After nearly six years of Intel telling customers that the AMD Opteron processor architecture was the wrong answer, this week our competitor has finally delivered “Nehalem” ― which some might call a copy, at least as far as the architecture is concerned."

We already know Nehalem architecture is basically an Opteron architecture. Wow, thats a technological breakthrough for big Intel.

Um.... what? Sam architecture? Ok so having a IMC and it being native quad makes it the same architecture. Even though everyone knows that Nehalem is basically a Core 2 with IPC enhamncements, a IMC and a super fast interconnect.

While the ideas have been around forever (and implemented into various chips BEFORE AMD had done it with K8) AMD was the first to produce it on a massive level. Intel has had various IMCs and has had a plan for the interconnect for a while.

And actually the Vice President sounds a bit jelouse and bitter to me. Its probably because he doesn't want to admit that what Intel said was that a IMC and interconnect is not needed right now. That and that Intel implemented it much better.

You are aware that AMD violated the same patents in question?

Seems Transmeta didn't sue AMD after AMD purchased nearly 10M shares of Transmeta. Coincidentally, AMD also had to have Transmeta develop the basics of Hypertransport for them.

So, please stop with the FUD. Everyone is entitled to a preference, but not an agenda.

Didn't know that. But of course if a company actually buys your stock out then why would you sue them? That means they basically invested tons of cash into you.

You should never worry about the shark you can see and already know how to deal with.

Besides... if my memory serves me correctly the manufacturer revealed that problem as soon as possible and also provided a fix; making it a non-issue. (Except for use in forums as flame bait.) New opterons don't have the problem anyway; making it even less of a non-issue.


======================================================
EDIT: Actually for all the "It would be stupid to disable HT/Turbo" people:

An admin that has older Xeons that have hyperthreading enabled might actually advocate buying these new Xeons and leaving HT/Turbo enabled.

BUT

An admin that has older Xeons that have hyperthreading disabled will probably not advocate buying these new Xeons. If they ever do get some systems with these chips then they will disable HT/Turbo immediately.

So the only real question is: What percentage of admins fall into each category? (I've already told you the answer.)

I agree with the fact tat a server admin would rely more on actual facts being posted about a new CPU than their experience. If they went on their experience they could have said "AMD is not reliavle...." blah blah blah. But instead if they go based on what they are seeing from results and test done by professional sites then they can make a more informed decision.

BTW, it does seem like you have no argument left. I mean now that Intel isn't bottlenecked in the server market what excuse will you use? Or are you willing to bite the bullet and get one and test it yourself.

Hell maybe it will be "more responsive" or whatever you used before. Or are you afraid you might actually like a Intel chip thus backstabbing your AMD?

Even with HT and turbo off its still faster by a large margin.

With the guys its always the same. AMD=great/user experience....

Intel=spintel/marketing gimmick/unstable/Intel cpus are made to exel in benchies but suck in real world/The review sites are paid off by Intel.

Nobody cares about overclocking(when the AMD chips had no headroom). Now look at how well the PII's overclock.

It's always gotta be something with them. DOUBLE CHEESEBURGERSSS!!!!

I still love my double cheese burger Q6600. It was better than waiting it out and finding that the first Phenoms were a not so stellar performer for me. Plus for what I paid so far its been a damn good investment.

But of course. Intel is evil. Always will be because they are the biggest company. Just like to Mac owners M$ is pure evil, unstable, no good, a business only machine and so on and so forth.

Man the world suprises even me sometimes.

 

[keithlm]

Hell maybe it will be "more responsive" or whatever you used before. Or are you afraid you might actually like a Intel chip thus backstabbing your AMD?

Actually I have jobs that take more than 48 hours on the Xeon production systems we currently have. Hyperthreading actually slow things down and the jobs take about 4 or 5 hours longer to run with it enabled.

The jobs take about 35 hours on the Opteron test systems we use. It is interesting that the Xeon systems run at a higher clock speed to produce the slower results. (Yes: Same amount of CPU and cores.)

They are now swapping out the newer Xeons with the older Opteron machines.

Humorously I have nothing to do with the decision making process. But it definitely does affect my opinion of Intel and Hyperthreading. (The answer is NO and NO.)

But even more luck: The new Corporate plan is on migrating totally away from the Intel/AMD architecture completely. (And going back to UNIX.) The move to Intel/AMD on Windows was a bad decision that will now be corrected over time.

Eventually I won't care about the entire Intel vs AMD issue because my personal workstation will only be a terminal.

 

[fazers_on_stun]

You are kidding me... Abinstein a moderator?

I thought the bias Ghost posses is horrible enough.... now Abinstein? No wonder there are more bots visiting the sites than users :lol: :lol: :lol:

Unfortunately, yes - his username is in green, which either means he's got fungus or is an AMDZone moderator - maybe both

I predict that some of the more moderate long-time pro-AMD posters there will depart for other pastures - the fellow who runs a render farm for instance. Abi was trying to equate his P1 machine video compression times to this guy's experience in CGI, and flaming him as being a "FUDster" or some such... Despite many other long-time forum members telling Abi to lay off, he never did really.

UAEZone is trending to more and more users like Zooty and Abi - full of themselves, more opinions than knowledge, and simply intolerant of anybody else's viewpoints. So that's why you get mostly flame threads against Intel rather than any intelligent discussions. Sharikook would be 100% at home there.

 

[fazers_on_stun]

You are correct. We only have to wait for the 2P and 4P and 8P results. (You know... the small machines that don't really do any serious work anyway.)

Isn't 2P the majority of the server market? And isn't 4P just around the corner?

On the other hand your posts won't make a bit of difference when it comes to the HT issue. This toy will be disabled by most experienced administrators.

Maybe. But it occurs to me that perhaps an admin with more recent experience might think otherwise. Esp. in light of the SAP scores. I guess one of us should Google and find out 😀

Pop quiz: How many of you people that keep posting about how an admin would stupid to disable HT have actually worked in an admin job? And I don't mean running a single server.

Not me - I'm one of those "whining users", and proud of it! LOL

I must admit.... back in the day I only had 50 servers that serviced about 1200 workstations all over the country. Actually since I wrote the admin manual used by everybody in the country and decided how all the servers and workstations would be configured that number should actually be more like about 600 servers and 15,000 workstations. But hey... apparently that means I know nothing about how business and government work... since I only did that job for about 5 years. And since it started in about 1989... that must mean I'm out of touch with reality. (You wish.)

Now, now - I was just pointing out some inconsistencies in your previous statement. However, what "toy features" in CPUs got disabled in 1989 (circa the 486 era IIRC) and in 1994 (Pentium MMX era)?

 

[someguy7]

Great, Base an opinion on Intels Nehleham Xeons with HT on old Xeons(that everybody with a brain knows sucked compared to AMD's opertron). AMD troll logic at its best.

 

[fazers_on_stun]

Actually I have jobs that take more than 48 hours on the Xeon production systems we currently have. Hyperthreading actually slow things down and the jobs take about 4 or 5 hours longer to run with it enabled.

The jobs take about 35 hours on the Opteron test systems we use. It is interesting that the Xeon systems run at a higher clock speed to produce the slower results. (Yes: Same amount of CPU and cores.)

They are now swapping out the newer Xeons with the older Opteron machines.

Humorously I have nothing to do with the decision making process. But it definitely does affect my opinion of Intel and Hyperthreading. (The answer is NO and NO.)

But even more luck: The new Corporate plan is on migrating totally away from the Intel/AMD architecture completely. (And going back to UNIX.) The move to Intel/AMD on Windows was a bad decision that will now be corrected over time.

Eventually I won't care about the entire Intel vs AMD issue because my personal workstation will only be a terminal.

So what exactly are these "newer" Xeon systems that are being swapped out for Opterons? Weren't the last Xeon CPUs to support hyperthreading based on the old P4 designs?? IIRC no recent Xeons before Gainestown supported HT, just SMT in multi-sockets.

Beckton, with 8 cores, 4 QPI interfaces and support for 4 sockets is supposed to be out later this year. With HT that's 16 threads per socket, or 64 per server. Maybe you should suggest your company look at those, eh? Oops, I forgot your previous edict, that serious businesses stay away from Wintel.

 

[yomamafor1]

Great, Base an opinion on Intels Nehleham Xeons with HT on old Xeons(that everybody with a brain knows sucked compared to AMD's opertron). AMD troll logic at its best.

He IS a regular at the AMDzone... what do you expect?

 

[cjl]

Actually I have jobs that take more than 48 hours on the Xeon production systems we currently have. Hyperthreading actually slow things down and the jobs take about 4 or 5 hours longer to run with it enabled.

The jobs take about 35 hours on the Opteron test systems we use. It is interesting that the Xeon systems run at a higher clock speed to produce the slower results. (Yes: Same amount of CPU and cores.)

All HT Xeons up until Nehalem were P4 architecture, so it's no surprise that they sucked compared to opterons. Also, the HT was implemented differently with them. The current implementation of hyperthreading has basically eliminated that performance penalty that sometimes occurred with the older design, and the core itself is basically taken from a Core 2, but without the bottlenecks in the FSB and old memory controller. Basically in every way, it will wipe the floor with the Opterons. That isn't just an opinion either - the current benchmarks show that a dual socket Nehalem can nearly keep up with a quad socket Opteron setup (and yes, I am referring to the latest Opterons).

Don't get me wrong, I like Phenom II quite a bit, and it definitely is a great CPU for most home users. It games nearly as well as i7 (well enough that I would recommend Phenom II in many cases, depending on budget), and is a vast improvement over the disappointment that was Phenom I. However, to claim that it beats Nehalem in server tasks, a market which Intel specifically designed Nehalem to excel in? Ridiculous. Luckily for AMD, even low end server parts tend to command fairly high margins, and the AMD top end parts are still competitive with mid range to lower end Nehalem parts (both in pricing and performance, at least as of the last time I saw), so it isn't completely hopeless.

 

[MU_Engineer]

Whoops double same post.

Anyways. While we cannot say what a server admin will do I do not see a reason to disable HT since it actually does not change the power usage much at all.

The server admin might shut it off if it negatively affected performance due to threads "bouncing" between real and logical cores or loading up logical cores before real cores (like what happened with HT on P4s in some cases.) That's not Intel's fault, it's whoever made/configured the OS kernel scheduler, but it still might result in HT being shut off.

In fact a Nehalem based Xeon only uses about 10-15w more than a Penryn based Xeon due to the IMC being on the CPU. But that means the chipset will use less power since it will not have the MC anymore thus making the power jump and loss even out.

There is also a considerable power savings to be had in moving from DDR2 FB-DIMMs to normal RDDR3 RAM, particularly as the number of RAM modules increases.

this is even shown in a Core i7 compared to a Core 2 Quad. The CPU itself only uses 10-15w more due to the IMC but overall system power usage has not changed much at all.

The Bloomfields are more efficient at idle due to the clock gating and the smaller-transistor, lower-voltage on-die memory controller. But, they take more juice at full roar due to the larger number of transistors and higher clock speeds allowed with Turbo Mode. It's not a big difference and few that would buy those parts care as their GPUs suck far more joules than their CPUs do.

Add in this and the 32nm parts that are going to have a newer versionof the HK/MG and more than likely backwards compatability it does present a threat to AMDs bread and butter server market.

It does, but AMD isn't sitting still. The quad-core Shanghais have quite a bit of headroom left, so AMD can simply goose the throttle to reach the same power consumption numbers as the Nehalem-EPs and narrow the performance gap (but I highly doubt they'll eliminate it.) AMD has apparently pushed up the six-core Istanbul processor to June or July, and Istanbul brings HT 3.0, six cores, and a few other tweaks to the table. Unless AMD runs into thermal issues with Istanbul and has to keep clock speeds low, Istanbul looks like it could be very competitive with the Nehalem-EPs.

I guess the only down side here is that now there is no way to say that a Intel chip will not keep up with or beat a AMD chip in the server market. Thats how it has been in the past but this is now. Things change. Much like AMD did with K8 to whomp Intel, Intel is doing with Nehalem to whomp on AMD.

The real unknown quantity is what's AMD's K11 is going to look like. Intel isn't supposed to do much work on the Nehalem's microarchitecture for a couple of years, so performance will mostly be due to clock speed, core number, and cache size. Those tend to increase in a pretty slow and predictable manner. AMD supposedly has a significant rework of the K10 microarchitecture in store with the K11, so we really can't say what they have up their sleeve as they are notoriously quiet.

 

[someguy7]

He IS a regular at the AMDzone... what do you expect?

Sorry. I didnt actually know he is a reg there. That site is a joke. At least the forums are.

Thanks yomamafor1