Intel runs into thermal wall again with Conroe XE aiming at 130 watts
Intel demoed overclocked Conroe XE at 3.5GHZ with liquid cooling to journalists. David Tuhy at Intel said that the Extreme Edition would be able to run at up to 130 watts.
No matter how much Intel can overclock Conroe, it can't match the performance of 4x4. You see, overclocking Conroe from 2.9GHZ to 3.2GHZ is only a 10% improvement, but power consumption increases by 30%. Power consumption rises non-linearly as one increases clockspeed. From 3.2GHZ to 3.5GHZ is just another 9% increase, but liquid cooling is required. Intel is stuck with the old school of thought. The way they try to increase the number of compute cycles is again through jacking up frequency. Soon Intel runs into yet another thermal wall, deja vu all over again.
Intel never learns.
AMD's thinking has been different since 1999 when they designed the Opteron. AMD uses more compute engines to increase performance . With AMD64, you double the performance by doubling compute engines, thus multi-core and multi-CPU designs. The key to the success of this is to make sure when you double the engines, you have a delivery system that allows the compute cycles to be fully utilized, instead of being wasted in FSB collisions. That's the Direct Connect Architecture. IMC, ccHT, XBAR and HT technologies ensures near linear increase of computing power as one increases the number of compute engines.
AMD's 4x4 technology gives you near 100% improvement at the blink of an eye. 4x4 permanently pins Intel at half the speed of AMD. At 130 watt limit, AMD can put together two 65 watt AM2 X2 4800+ processors in a 4x4 PC, which almost double the performance of a X2 4800+. The beauty of Direct Connect Architecture allows AMD to build such a system at very low cost. No additional chipset is needed for 4x4, you just connect the ccHT wires. I estimate a 4x4 motherboard will cost $150-$200. 4x4 will be faster than a 2P 4 core Opteron, because it uses unbuffered RAM.
If Intel competes against 4x4 with a cheaper two socket shared 800MHZ FSB design, performance will be very poor, as each core will only get 200MHZ bandwidth. If Intel pulls Bensley/Blackford into desktop, the cost will be very high.
posted by Sharikou, Ph. D. @ 6/04/2006 09:00:00 AM
47 Comments:
Thats just the cpu, looks like extreme cooling is needed by the VMU's this is not good. gamers want a fast system without it being maxed out just to keep up with standard air cooled under clocked AMD's.
Sad so Sad
I hope the 4x4's are availiable sooner than later.
Are we to trust someone who only claims to have a PHD?
Your AMD fanboism is soon to be shot to hell as plenty of people are getting Woodcrest demo systems, most of which are quite used to the Opteron performance in their existing servers.
I can tell you that under Windows Server 2003 x64 with SQL 2005 64 bit, the Woodcrest demo we are currently running (dual processor system) beats the 2.4 Dual Core Opteron by over 40% in ALL categories used to benchmark SQL 2005.
FOURTY PERCENT.
Claim foul all you want, we installed the OS, we installed SQL, we installed the test suite.
The numbers speak for themselves.
I really don't get it why these 4x4 is going to help anybody but look
here.
This is first Conroe "vendor controlled benchmark". These numbers ARE NOT IMPRESSIVE. POV-Ray benchmark is done using single-core with 4MB of cache, PCMark 05 gives Conroe an 12% memory advantage (?), 17% CPU advantage. This is the best what Conroe can achieve with 2.67GHz.
AMD probably needs only to ship 3GHz chip to match these results...
I think that Conroe 2.67 is the best what Intel can give to people now. The rest is close to 2GHz which means that we can expect lower prices for X2 soon. Why not? 30-40% increased capacity with FAB36 means more space to satisfy the market!
If we discuss benchmarks NGMA is able to only to MATCH the performance level but nothing more.
And we still compare 65nm to 90nm!
I want to conratulate for the right choice as a customer. :)
we installed the OS, we installed SQL, we installed the test suite.
Dude, why don't you disclose your system configurations of the Woodcrest and Opteron systems and certify them? You understand database performance heavily depends on storage, don't you?
All semi-verifiable benchmarks show that Woodcrest is 10% slower than Opteron at the same clock.
Link - This is a dissection of the previousposters link, but no way was this "vendor controlled". This was a few european reviewers invited to Intel's shindig and same with IDF, they pushed buttons and wrote down scores, they even compared it to the same "FX-60" that is "overclocked" to "2.8GHz".
The beautiful thing is that they're comparing Conroe w/ 2 x1900's to FX-60 with like a 7800GTX and saying "Conroe pwnz" - Quite humorous.
This was a few european reviewers invited to Intel's shindig and same with IDF, they pushed buttons and wrote down scores
Intel is a fraud. I proved it.
"At 130 watt limit, AMD can put together two 65 watt AM2 X2 4800+ processors in a 4x4 PC, which almost double the performance of a X2 4800+."
Dailytech says that the 4x4 will accept only FX62s and higher.
If this is true, then 4x4 systems are going to cost a bomb. You don't think AMD is going to drop the prices of FX62s to match the E6600, do you?
And then, they won't really affect Conroe sales that much; seriously, how many people can afford $2400 for CPUs alone???
And if AMD does support less than FXs for 4x4, why would anyone buy their FX62s then? By your own logic, everyone would just buy 2 3800s which would be cheaper than an FX62.
By the way, this really looks like AMD is admitting that none of their chips will be able to match a single Conroe. So, they are hoping two of their top-end chips will be competitive to a single Conroe.
Link - Shows the 2.13Ghz Conroe not beating the FX-60 as promised by the Intel Rep.
Link - This shows that outside of Gaming, Conroe isn't the "killer" once thought.
Link - Shows how great these tests were, comparing DDR400 to DDR2-800 in the Conroe system. It also appears they're using 2 different graphics cards, and since I don't speak chinese and am too lazy to get it translated, I can't comment on the legitmacy of these benchmarks (probably Intel controlled).
No. I am absolutely certain that AMD will allow people to use lower speed AM2s in 4x4. Two FX62s is a overkill for Conroe. Two X2 3800+ at $290 each is more than enough to frag any Conroe. I personally would like to have a 4x4 with two 35 watt X2 3800+. The configuration will be
1) 4x4 MB: $200
2) Two X2 3800+ (2GHZ) energy efficient 35 watt max. Total 70 watts max, 8GHZ compute cycles total. Total cost $620.
3) One graphics card ( I am not a gamer)
This shows that outside of Gaming, Conroe isn't the "killer" once thought.
Games are very sensitive to cache size. Benchmarks between 1MB 939 and 512KB 939 show increasing L2 by 512KB increase gaming performance by up to 8%. Conroe has 4MB L2.
http://www.dailytech.com/article.aspx?newsid=2638
This says that the 4x4 will take FX62 and higher.
This brings me to another question. If, as you say, 2x3800 will be available; then why would anyone buy an FX62?
By your logic, the FX62 will be fragged by the 2x3800, and the FX62 will cost more, and consume more power.
Another question: In how many non-server applications will a dual core K8 outperform a similar-speed single core one? And, in how many games?
So, in how many games will a 4-core platform outperform a 2-core platform, by virtue of the additional cores alone?
For all applications that give good performance increases with greater number of cores, dual Opterons already exist, right? What makes 4x4 so special?
Sentinel, stop being a moron. People will buy the FX-62 because later on, guess what genious, they buy another!
If, as you say, 2x3800 will be available; then why would anyone buy an FX62?
Good question. But today a lot of people are still buying single core 939 4000+ which is priced higher than X2 3800+.
Even if 2x3800 eats some FX62 business, I expect the volume will compensate the loss.
In how many non-server applications will a dual core K8 outperform a similar-speed single core one? And, in how many games?
Since the introduction of Intel's hyperthreading and more recent addition of dual core, newer applications are written to take advantage of multi-threading. Most of the applications and benchmarks are multithreaded (media encoding/decoding, file archiving, web, content creation, SysMark, 3dmark06...). Newer games are also multithreaded (Quake 4, Call of Duty 2, Oblivion...). So you see the Pentium D 805 beats Athlon 64 3000+ in quite some gaming benchmarks. Going forward, I expect most games will be multithreaded to take advantage of multicore.
OK, I understand what you are saying.
But I disagree, and this is why:-
People who buy faster (clock speed) single cores over slower dual cores do so because they want better performance in single-threaded applications; mainly, games. IIRC, the dual core FX was released long after other dual core A64s, and still AMD marketed the FX as the better choice, rightfully so, simply because of this. Even now, an old single core FX will perform better than slower dual core X2s in most games and other single-threaded applications.
I think this will be the case with the 4x4 too. I think gamers will mostly choose an FX62 over 2x3800, simply because in 99% of games, a single FX62 will be way better.
And, I think Dailytech are right, AMD will not support 4x4 for low-end X2s, or they will end up seriously hurting the sales of their own high-end processors.
After all, people like LMM will think, they can always add another 3800....
"Going forward, I expect most games will be multithreaded to take advantage of multicore."
I completely agree, and all applications, actually.
But four threads?
How many applications now have 4 threads? Many don't even have 2.
I think it will take a while for games and other applications to use 4 threads; and by that time, K8L and Nehalem and who-knows-what-else will all be common.
In most games present today, or will be released this year, which would give a better performance? A dual FX62 4x4 system; or a system with a single FX62 or a Conroe XE with the remaining 1000 dollars spent on exotic overclocking to clock the FX62 to 5GHz or the Conroe to 6GHz?
Sentinel is kinda dumb.
All 3D Modeling software supports up to 6 threads or more. Encoding software is 4 Threads or more for the norm. MultiTasking, last time I checked, uses multiple threads.
4x4 is not for single applications my small minded Sentinel, it is to preserve gaming performance while doing a million other things.
How many applications now have 4 threads? Many don't even have 2.
Back on the FX62 question. I think Mad Mod Mike made a good point. There will be people buying 2x 3800s. But there will be people buying 2x FX62s. People never get enough computing power.
As for threading. Once the code is designed for multithreading, it's usually for unlimited number of threads. Making the code 2-threaded is the same as making the code N-threaded. So called "1,2, infinity" principle of computer science.
"All 3D Modeling software supports up to 6 threads or more. Encoding software is 4 Threads or more for the norm. MultiTasking, last time I checked, uses multiple threads."
Is AMD targetting 3D modelling and encoding with 4x4? I do know about multithreading performance in encoding and related apps.
I thought AMD was targetting the extreme gamer with 4x4.
And my points were based on this assumption; from the POV of a gamer.
I thougt people who did 3d modelling and encoding already used 4-way and 8-way Opteron systems... Such people won't find the 4x4 that attractive, will they?
"As for threading. Once the code is designed for multithreading, it's usually for unlimited number of threads. Making the code 2-threaded is the same as making the code N-threaded. So called "1,2, infinity" principle of computer science."
Well, I am not a software engineer; but does performance really increase linearly with the rise in number of cores? In all applications?
Will an application show the same kind of increase from 2-cores to 4-cores as from 1-core to 2-cores?
"You can't fix stupid" - Too true for Sentinel.
4x4 will offer the best performance for gaming using as many processors as it is capable of while also allowing them the comfort of being able to multi-task. Who says AMD's reverse-threading won't be enabled on a firmware device inside a 4x4 board allowing 2 FX-62's to chug away at a Single Threaded game?
Again, what makes the 4x4 so special? What makes it such a big deal when you can have 2P Opteron systems? Or 4P Opteron systems?
I thought AMD was targetting the extreme gamer with 4x4.
And my points were based on this assumption; from the POV of a gamer.
Again, when a game developer figures out how to divide the game into concurrent threads, it won't be 2 threads. Instead, the code will supply unlimited threads. This is "1,2, infinity" principle in computer science. Solving the problem of 2, is a general solution of N. For instance, suppose one changes the game code to run each object in a thread, then the system will spawn a lot of threads. Four cores will run 4 threads at the same time.
Either Sentinel is trying to condensation me, or he's just that stupid.
2P Opteron boards that provide SLI, are $400 or more and $500 for good ones. 4x4 is going to be 1/3 of that and instead of using buffered DIMM's, it will use non-registered memory and DDR2. A 4P Opteron board doesn't even provide SLI, and a good board that provides a single PCI-E that is 4-Socket for s940 is $1,000 range.
Again, what makes the 4x4 so special? What makes it such a big deal when you can have 2P Opteron systems? Or 4P Opteron systems?
4x4 looks like an old idea, but it actually brilliant. The difference lies with cost effectiveness and target market. Intel would need a tweaked Woodcrest server to compete against 4x4. I have analysed this elsewhere. For a user, he doesn't care about how many sockets are there, all he really cares is price, performance, power consumption. 4x4 simply gives more computing power at better price/performance.
Reverse hyperthreading in 4x4?
I really don't have time now to explain what I feel about the reverse-hyperthreading rumor. Anyway, I will say just this. I think there is a greater probability that the 4x4 is actually a quantum computing machine than for the 4x4 to have hardware-based reverse hyperthreading.
Now, LMM has already made his position clear.
But this is what I would do to build an AMD gaming system: I will get a single FX62, and spend the other $1200 for the best cooling system money can buy, and then cool it to 5GHz. I think that will give me better performance in 99% games, past, present or near future, than 2 stock FX62s.
For an AMD encoding/3d modelling system, I will buy a 2P Opteron system...
I think I will sign off now, till the 4x4 news gets clearer. Specifically, whether it will accept low-end X2s, and probably benchmarks against 1P systems. I think 4x4 will NOT accept non-FX X2s, and that it will NOT offer a performance increase to warrant a $1200 extra processor...
Sentinel, I really don't understand - what kind of computer do you have?
I have A64 3000+ and it's enough for most of the tasks. I will buy a dual-core but 4x4 is a valuable proposition. During analyst day AMD representative has said that X2 will be supported in 4x4 configuration - you can check it. Second, quad-core applications are coming, games are coming. Talking about single thread best performance is not an option. What kind of application requires super-performed single thread? Games. Only games. The rest of the software can be written to support multicore and new applications are designed to do so.
I write applications and I would like to have quad-core. One for database, one for system and minor tasks, two for application execution. Why not? I hope that 4x4 should give a chance to buy single-thread cheaper CPU to just build "artificial X3". How about cheap Sempron for 70$ and X2 4000+ as a heavy duty?
Just saw this:-
http://arstechnica.com/news.ars/post/20060602-6977.html
Article by Hannibal at ArsTechnica:-
"So is the gaming world ready for quad-core? In a word, no. My prediction is that for most games in the next two years, you'll be better off with a dual-core machine where each individual core has superior single-threaded performance than with a quad-core machine where the single-threaded performance is lacking. I'll even go ahead and predict that a production dual-core Conroe system will still beat a production quad-core AMD system when the two are benchmarked against each other on popular games. (Note the words "production" and "popular" games. This means systems and software that are actually available to ordinary gamers in the retail channel, and not some combination of unreleased/beta/demo hardware and/or software.)"
Article by Hannibal at ArsTechnica...
The author is definitely not up to the trend. Multithreaded games are a MUST for the next two years if you want to be good. Right now, those multithreaded games such as Call of Duty 2, Quake4 and Oblivion definitely have a massive edge, as they full utilize the power of multicore. Newer games will be mostly multithreaded. Remember this, the Xbox360 and PS3 require multithreaded games. The game developers usually develop games for PC and consoles. They are not stupid.
Sentinel,
You are amazing !
What you predicted can not be true in any cases. If application is not multi-threaded, single core CPU will be the best one for use. If the application is multi-threaded, quad-core CPU will be fast than dual-core or single-core. There is simply no case that dual-core will be the fastest.
Reading your posting, I can tell you do not much about programming. In general, the applicatin is either multi-thread or not multi-thread. No programmer will programming exact 2 threads, but N threads. Once the application is multi-threaded, the more the core, the better the preformance in general (particularly when we are talking about 2-4 cores)
this is true that 4x4 will not give immediate solution for all performance needs for today gamers.
however, lets be honest MT software is the future of microsoft (as it reality today for all unixes including linux and macosx). Intel sends same signal here - projecting NO single core cpus next year.
so, it just a matter of time when all games will utilize MT on full speed. today we have some games. tomorrow we will have all. besides that, even single threaded games will benefit of, say, MT drivers for GPU, MT operating system and so on.
windows is already multithreaded today. anyone can easily check amount of thread in any windows application using task manager column. no reasons to think that vista will not use MT at full scale keeping in mind cpu indistry signals.
looks like that AMD is better prepared to MT future with 4x4 and next gen opterons.
now about 40% of Woodcrest vs 2.4 (slowest BTW:)) opteron.
Im working on company which provides IT solutions for financial companies (ML as well btw, hehe). we constantly test intel vs amd using redhat/suse linux.
last time we did it, using our daemons, 2P 3.6 xeons were slower than 2P 2.4 opterons by 4 (!) times. xeons just have insufficient bandwidth for real, big scale application. new Woodcrest will be better of course, but I wonder they will cut all 4X distance.
majority of our customers used 2P systems 2y ago. today most use 4P X2. thus 8 logical real cpus (not 4 real HyperT-cpus as for Intel). I just see no way to go back to 2P, most will go rather 8P 2X next year, when total solution will become cheaper. intel share besides our customers is close to 0% today.
4P 2.4 opterons is quite cheap, very fast and energy efficient solution. and its still relativelty low end server solution based on amd :) no need for any fanboism, anyone can measure and decide himself.
Let us see woodcrest performance when reviewers get to run it in their environment with specs they have to put together and not in intel controlled setup.
We should not forget we are still comparing 90nm parts to 65nm parts. How many more months before AMD 65nm parts r available in the high end category? Ins't woodcrest and conroe targeting the highend market intially?
The fact that AMD is ramping up production and even thinking of putting a new fab in Newyork state means they r confident of grabbing more market share in the next 2 to 3 years.
It is high time intel did some soul searching and went back to doing great chips instead resorting to the gimmicks they r busy with right now.
I see that every company where management changes to being lead by marketing gurus rather than techinal leaders suffers in the long term.
If we take what we know at the moment and that is that AMD have said it is for FX62 and upwards and taking as we know at the moment the price of the said chips then this is going to be a very expensive way to get performance.
AMD have got to cut prices somewhere. In an earlier entry Sharikou said that AMD got get the students to continue to buy AMD with a marketing campaign, but I tend to think students buy mainly on cost.
AMD HAS TO REDUCE PRICES rather than forcing the cost up with these 4x4 systems.
"Dailytech says that the 4x4 will accept only FX62s and higher."
Wrong. Dailytech says 4x4 will accept ONLY AM2, such as FX-62 and higher. That doesn't mean ONLY FX-62 and higher.
What prevents 4x4 to accept slower CPUs when 1) socket is the same 2) HT speed is the same?
"How many applications now have 4 threads? Many don't even have 2."
How good is it to limit the number of threads to 2 once you adopt multithreading programming? It's like... how good is it to limit the number of functions to 2 in C?
It wouldn't be called procedural programming if there are only 2 functions. Similarly, it wouldn't be called multithreading if there are only 2 threads.
"I will get a single FX62, and spend the other $1200 for the best cooling system money can buy, and then cool it to 5GHz."
You can't do that. Physics doesn't work that way. At 5Ghz speed, the voltage would easily punch through thousands of transistors on the die, and your cooling system might work better as a fire extinguisher.
Let me tell you why 4x4 is good.
1) It doesn't need server-grade stability features.
2) It's specialized for dual graphics, something the servers don't need.
3) It might not match a 3.7Ghz Conroe's peak performance, but will behave smoothier and more responsive.
4) You can upgrade and get double-boost. Don't dream of upgrading your Conroe... Intel will at least sell you another pair of MB + CPU, for sure.
There are others if these aren't enough for you.... :D
"My prediction is that for most games in the next two years, you'll be better off with a dual-core machine where each individual core has superior single-threaded performance than with a quad-core machine where the single-threaded performance is lacking."
I don't agree with that at all.
Any title that works well on PS3 or XBox 360 will work better for multicores.
Also, there's no production Conroe system yet, not until August, probably later. And he's wrong to think game writers won't do multithreads to take advantage of Conroe's dual cores at that time. He's even more wrong to think those multithreads are limited to two only.
I don't agree with all your comments regarding multi-threaded approach.
As a programmer I understand that games can be designed only for dual-core. Why? Because if look into the game we can see two main elements: first is a game mechanics, enviromental reactions etc. second it is enemy simulation/decision. Now if we take FPP (Quake) game, there can be single thread for every artificial enemy or one thread for all of them. More threads means more synchronization points and slower performance so to achieve best performance the programmer can design it for two thread architecture only.
If we look at xbox and ps3 architecture, it is MT for sure and this is a very important point for 4x4 architecture. We also must remember about physics which needs a whole CPU or maybe two only for calculations...
4x4 is good for gaming!
For sentinel and others who think multithreading doesn't matter to general users - on my current Windows XP laptop I have 51 processes and 445 threads. That includes a bunch of Windows background services, my IM program, Mozilla/Seamonkey browser/email client, Skype, and a few other random apps. The fact is that Windows programs have been heavily threaded since the Win95 days, and multi-tasking performance has been important on PCs since at least that long.
As for games and single-threading - the fact is that single-threaded games are a historical anomaly. All of the arcade games of the 80s were multi-processor multi-tasking computer systems. It was only when porting them to home PCs that they were rewritten to operate in a single-processor single-threaded environment. (That's partly why so many ports of popular games of the era basically sucked on PCs.)
Gaming has always been inherently a multitasking problem - you have to run AI and control code for all of the objects/actors in the game. Decision making for each enemy, collision detection for every onscreen object, input processing, all of these make a game inherently multi-threaded. The fact that games are becoming multi-threaded *again* isn't anything new, it's just a return to gaming's original roots. Gaming performance, smoothness and playability will only improve as a result.
http://www.bit-tech.net/hardware/2006/06/04/intel_conroe_performance_preview/7.html
"We found that it was faster than the current flagship Pentium Extreme Edition 965 processor in nearly every single-threaded scenario, but there were times where Conroe fell behind in multi-tasking scenarios."
So we have a independent reviewer finding that Conroe is not good for multithreaded applications. And as he says, its still not the production hardware that'll be shipped to customers.
How many more skeletons are waiting to fall out of the Intel cupboard before Conroe finally ships. I think Sharikou, was right all along, Intel is really in deep trouble
On the subject of power consumption, have you seen that the latest AMD Athlon 64 FX-62 consumes MORE power than the Intel Pentium Extreme Edition 965 under a full load? Anandtech has the results right here: http://images.anandtech.com/graphs/amd%20am2%20launch_05230690506/12056.png
Then consider that the Conroe processors running at 2.4Ghz and 2.67Ghz will offer much better performance than an FX-62, while using a lower power design that will consume far less power than AMD's processors made on the old, pathetic, 90nm process.
4 x 4? Sounds like a scam to me! AMD knows that their processors are going to be so soundly beaten by Intel's next gen stuff that they need TWO processors to fight against ONE! No matter, as Intel's quad core Kentsfield desktop processor will be ready for release early next year, soundly defeating this pathetic hack while using only a single processor!
I predict that Intel will ramp up production of Conroe, Merom and Woodcrest quickly, and AMD's processors will be left in the dust. No one will want to buy them. Intel is setting competitive pricing on Woodcrest, Conroe and Merom; thus requiring AMD to cut prices on everything. Who wants a $1000 FX-62 when a 2.4Ghz Conroe that costs $300 runs much faster and cooler!? Intel can use it's remaining stocks of Pentium 4s and Pentium Ds and sell them at Celeron level prices, soundly beating anything AMD has to offer in the budget sector. Once all the old inventory has been cleared out, Intel can release Celeron processors for desktop PCs and notebooks based on it's next gen architecture.
AMD will be posting operational losses by the end of the year, and their stock will be valued at under $10 by that time also!
Intel can use it's remaining stocks of Pentium 4s and Pentium Ds and sell them at Celeron level prices
So buy one. The only viable offer from Intel now is D805 which you can overclock up to 4GHz. AMD is targeting 30% marketshare so 7 out of 10 people is going to buy this crap after all.
My last Intel was Celeron 300A. From this time, Intel was always worse and more expensive choice.
"AMD will be posting operational losses by the end of the year, and their stock will be valued at under $10 by that time also! "
i really doubt AMD will be in that situation again, considering they still have the performance crown in desktop / server area (NGMA hasn't released to the public yet), and WILL still have the performance crown in server 4P and up. intel still needs to "sell" their pentium chips to regain capitals and monetary fundamentals to build NGMA, which are pretty much proved to have low yield.
in my opinion, intel is in a more desperate condition than AMD.
"AMD will be posting operational losses by the end of the year, and their stock will be valued at under $10 by that time also! "
Really? Please let me know then, I could buy tons of AMD stock!
"Again, what makes the 4x4 so special? What makes it such a big deal when you can have 2P Opteron systems? Or 4P Opteron systems?"
Thats exaclty AMD's point!
A new architecture, using low-cost CPU's (Athlons) in order to allow Opteron-like functionality (2P/4P), but without it ever being compared to the Opteron due to its own unique branding and marketing.
Fact 1: Threaded applications are NOT comming soon... they are already here!
Fact 2: There are fewer non-threaded applications than there are threaded ones.
Two threads? Most apps already use dozens of threads... some even hundreds (servers).
Threads have several performance issues (in regards to your 4x4, see #3):
1) Synchronized blocks (very bad)
2) Context switching (from thread to thread - the smaller the time-slice for each thread, the worst off. Usually 1000 context switches per sec, or 10ms time-slice each).
3) Cacheing - immagine this: thread 1 executes function A on processor 0 and then it is interrupted in the middle of the function. It is next scheduled to run on processor 4. What about the variables that are in L1/L2 cache of processor 0? They are not in the L1/L2 cache of processor 4?
That is why most threaded application are usually NOT scheduled to run over multiple processors.
It is better to schedule a web server on processor 0, the DB on processor 1, application code on processor 2, etc...
Post a Comment
<< Home