Don't you love when people who have zero experience in the area of servers feel its their duty to (mis)educate the other equally inexperienced folks in threads like these?
I cringe when I read some of these posts, and knowing your area of expertise I bet you cringe even more so :laugh:
Zstream
Diamond Member
- Oct 24, 2005
- 3,395
- 277
- 136
Sorry mate but I have been laughing at most of your post and Phynaz for quite a while. It would appear you have limited experience in any server clusters or enterprise applications. Anyone who things dynamically overclocking a CPU in the middle of transaction is a good thing needs to sit down and reconsider your position.
piesquared, whether it will be turned on or off by default doesn't matter. A correctly implemented hardware overclocking feature will always be superior to a software feature. So it will be benfical to a bigger crowd by default, even if it's still a small crowd.
Personally going from the assumption "Intel engineers are smart and know what they have done", I'd say they want turbo-mode to run on most server-workloads too.
On the other hand maybe your understanding of turbo (or mine) is off, I suppose during one and the same transaction (i.e. workload) clock should stay the same if cooling is able to dissipate all the heat generated, no?
I think anyone challanging IDC needs to provide some proof, because he is our local guru. No just kidding, but seriously it would be nice if either side posted their arguments. I don't know what to believe, but want to be educated about the pecularities of server clusters and such.
Personally going from the assumption "Intel engineers are smart and know what they have done", I'd say they want turbo-mode to run on most server-workloads too.
On the other hand maybe your understanding of turbo (or mine) is off, I suppose during one and the same transaction (i.e. workload) clock should stay the same if cooling is able to dissipate all the heat generated, no?
Zstream
Diamond Member
- Oct 24, 2005
- 3,395
- 277
- 136
That makes no sense and shows your lack or experience with server clusters and major apps.
The technology is perfectly adaptable to home computer use. The instant this goes into a server environment I will place a wager most large company's (40k+) will have this option disabled.
In fact we will see another rise in having to disable HT unless Intel really changed the HT technology in the i7. Based on previews it seems as if it is very similar... Anyways, this has nothing to do with the preview of the Phenom II.
Originally posted by: formulav8
Well I built slightly over 1000 servers this year, how about you?
Jason
Slightly different logistics involved between the guy who works an assembly line building Ford F-150's and the guy who manages Enterprise Car Rentals north american fleet decisions on product mix and feature sets they want their customers to experience.
The discussion under contention here is the high-level decisions that go into deployment logistics of servers, and the features that are deemed assets versus liabilities when managing said fleet of servers.
I happen to know a little bit about what Phynaz does in TRW, ergo my comment to him about the wannabe's around here thinking they are schooling a newbie when they talk down to him. It's laughable, except the part where I cringe when I see how many people have no clue and at the same time have the arrogance to do everything they can to avoid being given a clue.
We are all born ignorant, to be ignorant is to be human, but it takes real arrogance to remain ignorant because we are swimming alongside legitimate experts in these forums.
Zstream
Diamond Member
- Oct 24, 2005
- 3,395
- 277
- 136
Originally posted by: Idontcare
Originally posted by: formulav8
Well I built slightly over 1000 servers this year, how about you?
Jason
Slightly different logistics involved between the guy who works an assembly line building Ford F-150's and the guy who manages Enterprise Car Rentals north american fleet decisions on product mix and feature sets they want their customers to experience.
The discussion under contention here is the high-level decisions that go into deployment logistics of servers, and the features that are deemed assets versus liabilities when managing said fleet of servers.
I happen to know a little bit about what Phynaz does in TRW, ergo my comment to him about the wannabe's around here thinking they are schooling a newbie when they talk down to him. It's laughable, except the part where I cringe when I see how many people have no clue and at the same time have the arrogance to do everything they can to avoid being given a clue.
We are all born ignorant, to be ignorant is to be human, but it takes real arrogance to remain ignorant because we are swimming alongside legitimate experts in these forums.
lol, right bud... Feature set of servers is nothing spectacular. They all right now have the same hardware and components. Very few large servers differ in this.
It is the installation and support of the Software on these servers that counts. How easy is it for someone to say we need a 4u server with 1tb of storage. They all have the same NICS, similar competing hard drives, backup PSU etc...
You can tell me what server you want but I am going to mad when the installed software accessed by thousands crashes because some new bug with HT or it dynamically overclocked and a thread crashed.... But whatever, apparently you are the expert lol.
Originally posted by: Zstream
Originally posted by: Idontcare
Originally posted by: formulav8
Well I built slightly over 1000 servers this year, how about you?
Jason
Slightly different logistics involved between the guy who works an assembly line building Ford F-150's and the guy who manages Enterprise Car Rentals north american fleet decisions on product mix and feature sets they want their customers to experience.
The discussion under contention here is the high-level decisions that go into deployment logistics of servers, and the features that are deemed assets versus liabilities when managing said fleet of servers.
I happen to know a little bit about what Phynaz does in TRW, ergo my comment to him about the wannabe's around here thinking they are schooling a newbie when they talk down to him. It's laughable, except the part where I cringe when I see how many people have no clue and at the same time have the arrogance to do everything they can to avoid being given a clue.
We are all born ignorant, to be ignorant is to be human, but it takes real arrogance to remain ignorant because we are swimming alongside legitimate experts in these forums.
lol, right bud... Feature set of servers is nothing spectacular. They all right now have the same hardware and components. Very few large servers differ in this.
It is the installation and support of the Software on these servers that counts. How easy is it for someone to say we need a 4u server with 1tb of storage. They all have the same NICS, similar competing hard drives, backup PSU etc...
You can tell me what server you want but I am going to mad when the installed software accessed by thousands crashes because some new bug with HT or it dynamically overclocked and a thread crashed.... But whatever, apparently you are the expert lol.
You are so blistering mad at me right now you can't comprehend the message communicated by the text of my posts.
So let's just agree you are the expert on servers, I know nothing, and now maybe you could post say 10 links regarding these multiple applications that has failed to function on any P4, Power, or Niagara based server running their implemented variants on SMT for the past years.
(not to be confused with the performance degradation that is known to occur on some applications in a P4 HT environment...performance degradation is not the topic here, you have been very specific that you are talking about software crashing)
I can't post a link on what has never happened, but surely it is so rampant an issue that someone somewhere has documented it for specific cases of specific applications, so coming up with ten unique links to it will be a walk in the park for you.
I await my re-education process. Thanks in advance for this opportunity.
Originally posted by: Phynaz
IDC, I don't work for TRW. You must have been thinking of somebody else
I just looked them up, they do about $16B a year in sales. My employer has about $20B in sales.
I meant The Real World
But yeah, my bad on mixing acronyms there $20B? Is that all? Loser.
I'm sorry, I'm no server expert (I work at a company that support's several major companies photo labs, but we only have one sys admin and a couple of servers)...but I guess I just don't see how, unless there was some underlying stability issue with the chip (which there won't be in cases where turbo mode kicks in) the processor running say, 100mhz higher for a bit would do ANY HARM at all to anything running on the computer.
Does anyone have any *actual, factual proof* that turbo mode would be a bad thing for a server? It seems like tech like turbo mode is exactly what you'd want on a server. The chip takes less power until there is a really high load then if it's adequatly cooled it cranks up the juice to help deal with the demand.
cmdrdredd
Lifer
- Dec 12, 2001
- 27,052
- 357
- 126
It would indeed be on. Intel guarantees the functionality of turbo, and Intel builds many of the boards uses in Servers. So it would make sense that Intel would spec it to run ON by default. How many servers have you installed or worked with?
The point is, Intel has made a simple 200Mhz boost in clockspeed, by default, with no intervention on the part of the user, safe. There will be absolutely NO harm done to any platform by implimenting this. It will help more than anything. When there's a large workload comming in you can get a nice little boost in performance to offset the workload for absolutely free. No downside. I'm going to assume Intel's engineers know a hell of a lot more about their CPU technology than you or I and that there is no issue whatsoever leaving turbo on.
I would doubt that Intel defaults the Turbo on, though they may...while they guarantee functionality, they won't pay for turbo-caused damages to data on a mission critical system, they will just replace the system...
I can't think of a single one of my customers that would run their server in "Turbo"...there's nothing worse than dead air for a broadcaster!
I would also assume that the Turbo feature will add quite a bit of time to qualifying platforms. I would guess that while they may indeed default the Turbo on, it won't happen till the end of 2009.
IntelUser2000
Elite Member
- Oct 14, 2003
- 8,686
- 3,787
- 136
And I'm guessing you would be wrong.
Have anyone heard of the technology code named Foxton?? That's the dynamic clock speed increasing technology that was supposed to be present on the Itanium 2 "Montecito" and "Montvale" CPUs.
See these are the very high end CPUs sold to people who care utmost about reliability. Indeed the reason Foxton was scrapped was because it was pretty complex piece of tech and there was too much variance with all the analog circuits.
Tukwila, the next generation Itanium, is going to feature the very similar digitally controlled power management circuit that Nehalem has.
Simply put, no they won't disable it.
cmdrdredd
Lifer
- Dec 12, 2001
- 27,052
- 357
- 126
That's the thing, Intel guarantees no errors in the CPU from a 200Mhz clock increase. If overclocking did cause errors then most of the people who overclock and do prime95 testing and come back with stablility would have had errors in the calculations. These are 1Ghz+ overclocks as well.
Dude.. I am not an AMD fanboy. Haven't had a favorite AMD system since K-6-III.
I just wanted to point out that neither AMD or Intel is now or ever will be YEARS ahead. Things change way to rapidly and 6 months from now either one can come up with a hefty surprise.
I have chose Intel for myself last few builds.. I did that because I am PC enthusiast and must have every bit of performance I can. I don't care for i7 myself, it maybe worthwhile in a year or so when software company's catch up with the synthetic benchmarks core i7 is currently generating. In the meantime if I think Phenom II is a better choice, then that's what direction I will head. Intel has tried hyperthreading before and it didn't pan out. Sorry for the rant.. but sometimes Intel fanboys can get annoying as AMD fanboys.
Has any one here ever thought about why would Intel on its new architecture will include a feature like turbo boost. I mean isn't a new a architecture suppose to be provide lots of improvement in performance per clock over the previous GEN without having TURBO BOOST.What I think is that TURBO BOOST is there to make up for performance slowdowns caused by HYPER THREADING in certain scenario's as we had the same issue with P4 HT even though implementation of HT on i7 is definitely better than on P4 .If not for that I can't think of any other reason for having a feature like TURBO BOOST on 7i.
cmdrdredd
Lifer
- Dec 12, 2001
- 27,052
- 357
- 126
If I owned AMD and you worked me as a viral marketer you'd be fired.
Seriously now, do you believe Turbo is only there because the i7 sucks? Really now...and here the i7 is at least as fast as a Yorkie and those are faster than AMD's current lineup so...
I'll take the free turbo boost. Thank you.
To be fair I have seen examples on high end Intel servers having thousands of connections where the software vendor recommended turning off hyperthreading for better results in enterprise class applications.
But on these desktops hyperthreading would not be a liability with video manipulation, editing, recoding and similiar applications. My own informal testing with Nero Recode shows with hyperthreading off the same recode takes longer.
So yes, thank you, I'll take the turbo, I'll take hyperthreading. For those cases it hinders rather than helps, I do have the option of disabling it. My last rig was a 2.1 p4 on a asus 533 mb. It was overclocked to 2.7 ghz for 7 years. At this point, I'm not interested in overclocking this system, but I find it interesting the goodies Intel put on the chip are touted as liabilities rather than assets. I guess the glass half empty crowd is in the house.
To be fair I have seen examples on high end Intel servers having thousands of connections where the software vendor recommended turning off hyperthreading for better results in enterprise class applications.
But on these desktops hyperthreading would not be a liability with video manipulation, editing, recoding and similiar applications. My own informal testing with Nero Recode shows with hyperthreading off the same recode takes longer.
So yes, thank you, I'll take the turbo, I'll take hyperthreading. For those cases it hinders rather than helps, I do have the option of disabling it. My last rig was a 2.1 p4 on a asus 533 mb. It was overclocked to 2.7 ghz for 7 years. At this point, I'm not interested in overclocking this system, but I find it interesting the goodies Intel put on the chip are touted as liabilities rather than assets. I guess the glass half empty crowd is in the house.
run their server in turbo wtf? the PCU control is automatic and binning includes turbo tests. a failure during turbo mode is regarded the same as any other failures.
and let's be honest here, how much more time can turbo mode add to OEM qualification? what basis do you have to make your assumption? why wouldn't the builders just leave the feature on and proceed as normal. it's a transparent feature for that reason.
IntelUser2000
Elite Member
- Oct 14, 2003
- 8,686
- 3,787
- 136
The exact reason is they are starting to exhaust single thread performance increaing options.
Back when Core 2 Duo first came and Johan said Free Lunch is Back: http://www.anandtech.com/showdoc.aspx?i=2748&p=6
"The result was that the reputed Dr. Dobbs journal headlined : "the free lunch is over" [1] claiming that only larger caches would increase IPC a little bit and that the days that developers could count on the ever increasing clockspeeds and IPC efficiency of newer CPU to run code faster were numbered."
Core 2 didn't use new ideas to increase single thread performance. It refined known ideas and expanded on them. Excellent prefetchers, Memory Disambiguation, better prefetcher, 4-issue, these aren't new ideas. There is no free lunch after all.
Turbo Mode was implemented because taking advantage of extra TDP headroom when running cool or not running all the cores is a good idea when implemented well. Clock speeds increase performance uniformly, unlike other architectural changes.
The next Tock, Sandy Bridge is only enhancing the Turbo Mode further: http://en.wikipedia.org/wiki/S...ge_(microarchitecture)
"Dynamic Turbo allows the CPU power to exceed the TDP value when the rest of the platform is relatively cool. The frequency gain can be up to 37% for one minute, and over 20% in most cases."
10-15% performance single thread gain will come from clock speeds.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter