Next-gen superdesktops: Nehalem-EX or EP

Can someone please tell me if this will make PC games any faster?

180W? I hope they are water cooled by default. Don't tell me that dopey little intel cooler can handle that. I doubt even a true 120 can handle that reliably. Otherwise I imagine that intel requires you to run your air conditioner at 50F.

Dave, I think you know it won't make a jot of difference to 99.999999999% of games out there.

Normally I'm excited about new technology, but I'm very happy with my Quad core for the last year or so. The limitations are still on graphics card and memory capacity. Which I'm sure will be dealt with largely through Larrabee if that ever makes to production! :)

@Jason
If so I hope they engineer a block for the entire top end of the motherboard, am tired of tubes running everywhere.

Was looking forward to higher clocks not more cores. (sigh.....)

@json: it says 180 when oc'd.

Uh, faster clocks not more cores?

2002 called, they want their marketing hype back ;)

800 gram solid copper heatsink X 2??
120mm fan X 2 ob the HS ??
Optimised case construction ??
Much prayer and crossing of fingers ??

Not possible with current stock cases and HSF's.

Will need water / vapour tech.

PC's will become increasingly a major fire hazzard in the home.

I will not be surprised if Governments look at imposing new standards on PC's with this sort of thermal compliance issue to address.

So, gaming is the concern, and we're talkin' CPUs.

Hmmm, something wrong here.

We all know the majority of games would love to have more GPU power, except for one, maybe two:
Flight Simulator X
Grand Theft Auto IV (possibly)

Flight Simulator X is so CPU and RAM bound it's stupid, years of building over the same engine that ran back in '98, and in fact, perhaps the CPU power wouldn't help, considering it's the PCI-e 2.0 interface that gets clogged up in the end.

And people say the PC version of GTAIV is quite CPU bound, but I don't have any first-hand experience with it.

So in the end, if you're considering having two Nehalems just for gaming, you're either rich and don't care, or you're just a moron who reckons more is better.

Well it is a very nice platform... but will it give me 100+ FPS in Crysis?

Someone above noted that games like GPU power... so this very expensive platform offers something like 4+ x16 PCIe slots.

Great, pay Intel $1000+ for a platform that can host 4 video cards. And pay AMD/Nvidia $1000+ for the video cards.

It is a nice job if you can get it eh? and cheers to Drashek.

Maybe the games programmers can write more efficient code?

No, I think I might as well laugh out loud while walking alone in public places.

absolutely agree:-

"Was looking forward to higher clocks not more cores. (sigh.....)"

QFT.

While multi cpu has its place I dont think the home market wants any more we just want higher clock speeds....after all multi cpu was only focused on when they hit their ghz limit....thats what they need to be focusing on not excessive mutlicore/threading.

Lynfield...supposed to hit 5 ghz on air...that will be enough for most domestic users for long time to come.

Give me threads any day and more cores.
What is the benefit of a 10Ghz single cored CPU over a four cored 3Ghz CPU?

Woopie, one does single threaded tasks quicker and one does not. Who is going to need a 10Ghz single cored CPU for there applications? Perhaps the people who insist on running poorly written applications, constructed by unintelligent programmers that insist on using only one thread. Why would you wish to run an old application on a single core at 12Ghz, that was originally designed to run on say a P3 etc?

Task Switching is passe, Multitasking is the way! Gawd didn't you guys own Amigas? or are PC people just not used to the term Multitasking yet and still believe in Task Switching??

That being said, shouldn't the same computing principals that apply to super computers, also apply to the humble PC? How is the PC special and how can it justify going against what has already been proven. That more cores and threads are better than only having one.

Reading most of the posts here remind me of the Pentium 4 era. It'll be a powerhouse, you shall need a fridge to cool it, is Intel mad? etc. Sounds more like jealous DAMMIT fanatics to me, whom haven't seen there idol put on any good shows for quiet a while.

I'm not advocating single cores but believe that multi threading has already gone far enough for the pc market...it is not only bad programing...some programs are able to benefit more from multi threading than others....just as the ghz limit was reached in the hardware race I believe the multi threading usefulness limit in respect of the pc has been reached and for that market progress now needs to be achieved by increased clock speeds.

as gpgpu increasingly comes to the fore and graphics core complexity enables them to deal with a greater range of parallel then the graphics card will become the natural home of parallel tasks leaving the cpu to do what it does best...hopefully at nice high clock speeds.

Dude, stop talking out your ass because you have no idea what you are talking about.

"Lynfield...supposed to hit 5 ghz on air...that will be enough for most domestic users for long time to come."

Nothing is ever enough, you have no idea what you are talking about.

"multi threading has already gone far enough for the pc market...it is not only bad programing"

Have a programmer friend who can't find a job giving you advice? Wow.

"just as the ghz limit was reached in the hardware race I believe the multi threading usefulness limit in respect of the pc has been reached"

More bullshit, do you have any idea, for example, how many cores GPU's have? 128, 256... and it's climbing, its the ONLY way to go atm.

"the graphics card will become the natural home of parallel tasks leaving the cpu to do what it does best...hopefully at nice high clock speeds."

Man stop taking crack. So first you say multi threading is not good, then you compare how good gpu's are at parallel tasking? AHahaha.

I'm sorry the economy has left you without a job, enjoy working at McD's.

Wow what an extreme reaction...first let me clarify as I may not have used my terminology as correctly as I should.
I'm saying that the cpu as it stands can handle more than enough threads for the pc market....we are upto 4 core and 8 threads at the moment with nehalem...frankly I hardly ever see my quad core Q6600 maxed out on all cores. Yes I understand that programers can make their programs more multicore/thread friendly but far from all programs can be made in this fashion so increasing the cpu cores by a factor of X2 does not increase your performance by X2 in many programs and does nothing in others where as a X2 increase in clock speed would do so across the board.
So personally I would rather see the cpu prioritize clock speed rather than thread handling capacity...I also tried to point that if massive multi threading is your concern then the gpu is more suited to that sort of process than the cpu...the large numbers of stream processors have already outstripped the cpu many times over in data parallel processes and with the increasing complexity of the gpu (the next generation nvidia offering GT300 is going to be MIMD rather than SIMD)then the gpu will be able to take on more complex tasks.
So my point is that there is already hardware available in the form of the gpu for multi threaded processes and that as the cpu has already reached a point where it is sufficiently multi threaded for the pc desktop market that design should be refocused providing higher clock speeds to enable it to deal with those processes which are not so easily dealt with by a multi threaded approach.

I well remember in 1998 getting a P2 box clocked at 266 MHz IIRC. That's megahertz.

I thought at the time I could never need a faster machine. Then I discovered Quake II. There is always something else.

Yes we will always want more performance but I suppose the argument is how do you get it, through parallelism or clock speed and on which bit of hardware gpu or cpu.

Gpu and cpu's are heading for a clash where it will be possible to run a computer completely on the gpu OS included in the abscence of a cpu and conversley where you can run a computer completely with the cpu without any graphics adapter.

Possibly one could push the other out of the market.
Rather than this I just think it better that developers concentrate on the strong points of each and develop those accordinglyt...ie using the gpu as a massivily parallel processor while the cpu should be less heavily parallel and do out of order processes that it is better structured for but should concentrate on higher clock speeds.

how so many of the self proclaimed tech savvy folks around here don't have a clue what they are on about.
CPU designers have discovered that much past 3GHz power consumption increases rapidly. Because of this they started by improving the number of instructions processed per clock, widened the system to 64 bits and having reached diminishing returns on these factors they've now moved to multiplying the number of cores (Moore's law only ever talked about the number of transistors on a chip BTW, not how fast you could clock it). They're also finding it progressively harder to shift data on and off chip fast enough to keep these cores fed and running at full speed.
Software designers have a different problem. Making use of these cores. Much software is hard to break down into parallel units of work, and most operating systems don't make life any easier on this front either (Windows applications have a single GUI thread and don't like other threads updating windows for example). Some processing tasks (for example video and audio processing) are fairly well suited to being run on parallel hardware, which is where GPGPU comes in, but this kind of task is the exception rather than the rule.
So, clocks are not going to get much faster (BTW Lynnfield looks like it's going to be no faster than Core i7 without water or vapor cooling, which are not practical options for 99.99%+ of all users) so the only practical solutions are going to be improvements to OS, languages and software engineering practices to make parallel programming easier.

"CPU designers have discovered that much past 3GHz power consumption increases rapidly."

And try as they might to increase performance by changes in other areas the bottom line is that the above is the main problem that needs to be addressed by whatever means be that an overhaul of cpu design and materials or whatever...that is where they should be focusing their energies as the other methods you outlined only offer limited performance increases.

We don't believe that the future of search is dead, neither inquery nor deed.

And the bandwidth played on...

@ Technogiant
I concur Sr.!

Was going to jump in before and debate on the higher clocks issue but your last post hit the nail right on the head.

@ Steve
Don't care what degree you have, at some point in time they are going to have to up the clocks! = True Progress

"at some point in time they are going to have to up the clocks"
If this was easy they would have done it already. LOTS of research is being pumped into solving the technical issues, but the gains have been fairly modest to date. Barring some unforeseen breakthrough this trend is likely to continue - modest improvements in clock speed and increasingly more parallel units.
Complain all you want, this will have no effect on what is technically possible. Software is not going to be able to rely on simple clock increases for performance, it's going to have to evolve to make use of these parallel cores, and that isn't easy either.

You accept too easily this approx 3ghz limit on clock speeds..this issue will be and has to be overcome to allow for true and continued progress all the rest is just temporary diversion.
You can only make your OS multithread friendly once and programs can only be multi threaded to a certain extent so you are rapidly going to hit a ceiling where by further parallelism of the hardware has no further benefit.

Continual increases in clock speed are the only long term sustainable answer.

Short of going into semiconductor research how can you do other than accept the limits that technology defines? You can stand on the sidelines shouting "want it, want it, want it" all you like, it doesn't solve the problem.
I want a car that will do 200MPG and 300MPH, the chances of me ever getting one are vanishingly small. The same kinds of problem are in play for this. The amount of power needed doesn't increase linearly with top speed. You need maybe 80 BHP to reach 100 MPH, 500 BHP to reach 200 and 1000 BHP to reach 250. That kind of power isn't conducive to fuel efficiency. Manufacturers have been trying to give us both power and efficiency, but they are constrained by technology and the laws of physics.
The same problems are faced by processor designers. In order to sell to the mainstream market their devices must be air cooled. Air cooling a 130W CPU becomes noisy and expensive (big lumps of copper, fast fans, lots of air to shift) and the results aren't twice as fast as a 65W model. They work within the limits imposed by their fabrication processes, and manufacturers are spending billions on trying to improve them also.

I do understand there are many technical problems associated with increasing clock speeds...I certainly don't believe we are anywhere near the ultimate constraints impossed by physics at an atomic level but the problems are those of manufacture techniques materials and design.
Perhaps I am just more optomistic than you as I fully believe that they will be overcome so that we can get back on the main road towards progress rather than this ultimately evolutionary dead end of parallelism.

You obviously don't understand what is going on here. Multi-core design is a response to the limitations of process shrinkage. Nobody likes writing parallel code, and we'd all love to avoid it if we could, but it's a necessity especially when there are severe limitations in pushing individual processor performance.

Even with perfect lithography and manufacturing, you're looking at devices which simply do not perform very well because of their atomic constraints. It gets worse with each generation, and will probably peter out around 10nm. How are manufacturing techniques going to help when even the metal wire conductors are contributing to slowdown, heat, and ultimately *cost*? There is only a benefit to increasing clock rates if you maintain the same or better architecture. Otherwise, a higher clock rate usually means reducing the amount of work performed per cycle (like the Pentium 4), ultimately gaining you nothing more than marketing material. The last thing we need is a "Ghz bubble" because people are stupidly infatuated with clock rates.

So really, do a little reading before you talk out your ass. Optimism is one thing, naivity is another.

You guys just got me shaking my head, have you guys seen/read some of the work they are doing with carbon?

To BB the world is flat and Steve would have us driving model T's. There are always solutions out there.

@Technogianr & mario
As Steve T says, we have more or less reached the limit of what's possible with today's materials. It isnt impossible to up the Hz, just uneconomical. We are now reaching the upper limit of the Hz envelope, back in P2 era we was at the bottom. So any new Hz gains must come from new tech, materials and/or processes, and for the next ten years we are stuck with what we have today. Simply because it takes about that long to get it from idea to retail product.

@Mario

"To BB the world is flat and Steve would have us driving model T's."
A ridiculous analogy and failed assumption. Neither says we wont move forward, but it is going to take time. Cars have evolved in 100 years+, ic's in 30? We are getting there, but not in at least another 10-15 years. Have patience, billions of euros is spent in R&D every year, and it is not going faster just because you are making a fool out of yourself. ;)

Ok we now have the hardware so where oh where is the software that can use all this power. Since most ppl NEED a OS they can use without knowing anything. Most computers are "FOR IDIOT" Windows boxes so we can rule out any OS that can in actuality use this hardware. Kick Ass stuff for thoughs who can use the processing power. Can you imagine how much this would have cost just 5 years ago.

"So really, do a little reading before you talk out your ass. Optimism is one thing, naivity is another."

I believe your thinking is rather stuck in a box as another comentator said.

Perhaps we are coming to the end of the GHz possible with current materials and manufacturing processes....but this is not a point at which to say... "Oh well thats it then".

This is the point where step change is required to perhaps totaly different technologies and materials.

I'm sure Intel would love to continue selling Nehalems of different flavours with more and more cores but with continualy diminishing returns in terms of performance so long as they can make a profit rather than invest in massive R&D and massive restructuring to move to totaly new technologies.

So long as people like you go on believing the mantra that the limit has been reached and excusing them they will continue to flog this dead horse for all it is worth.

So I'm not talking out my ass...just try and take a wider view.

the above was @BB not @Steve

honestly, I think both sides are right and wrong at the same time. Unless there is a new, standard, multi-threading programming platform/language, programmers are not going to be able to get much benefit from more than 4 cores on current tools for every day applications.

People! Intel and AMD knows this, you are not seeing intel churning out 24 core desktop processors now, not because they can't, its because is completely useless.

But then again, the Gigahertz race represents another problem. Higher frequencies usually mean more power, inherent problems to current chip designs, parasite capacitances building up inside, leakages, etc. Sure, we can always lower manufacturing process, but then what? we'll reach physical limits that are unavoidable for current semiconductor technology.

I'm sure the future of microprocessors is not gonna be trillion-core, multi terahertz processors, but optical switching logic processors and systems, on which these terms (threads, cores, cycles) won't have any significance.

It's amazing how stubborn (and unobservant) some people can be.
Firstly on the unobservant front, you were replying to a post by BB and attributing it to me.
Secondly, as has already been mentioned by Dudler, any new (non-silicon) process technology on the horizon is some way from mass production. At a minimum we've got two generations of new silicon based processors to come before anything else is ready to replace it. Semiconductor manufacturers aren't ignoring the other possibilities, but it will take at least that long to come up with the fabrication equipment and a new CPU design to build on it.
Until then you are, as has been said, talking out of your ass.

And you are incredibly rude and condecending...pity some people can't talk without the need to put others down.

Also you are now being naive...

"Semiconductor manufacturers aren't ignoring the other possibilities"

Rest assurred manufacturers will bleed current technology dry before they make the necessary investment for a complete revamp.

Naive? That's what you have been demonstrating my friend.

At each process size Semi manufacturers need to buy a whole set of new hardware. They don't spend the kind of money involved unless (1) it gives them a significant advantage and (2) they can make a profit. If, rather than buying a whole lot of new, VERY expensive, 32nm manufacturing kit they could buy into some other non-silicon technology which gave them significant advantage (e.g. much faster clock speeds) then they'd do so.

If AMD, for example, could suddenly make 6GHz+ air cooled CPUs it would be a disaster for Intel. They can't allow that kind of technology gap appear. Worst case they can move the older equipment over to making less demanding parts (which won't make the same kind of money, but will help pay for the costs), but whatever happens the top end manufacturers MUST keep buying new process equipment to stay at the top. If some other process will give them significant advantage then they'll buy it like a shot.

I understand what you are saying but we are not talking about an in comparison simple change of process size...I'm talking about a complete change of the technology and materials from the ground up.

We have all seen several spectacular news reports over recent years of things that promise to revolutionise computing...quantum computing...multistate transistors...I recall one report of a transistor of sorts able to switch at 500GHz...and countless others that I can't recall.

Yet these items come and go without further mention...and why.... because the cpu industries are so heavily invested in doing things the way they do now that they will not make any radical change unless they absolutely have to. The industry is an oil tanker moving in one direction that will not change course.

Perhaps it isn't even down to them to research these matters as they have a responsibility to their share holders not to thrown millions down the pan in often fruitless research....but even if a university boffin was to come up with a promising discovery the huge investment that has been made pointing the industry in it's current direction would prevent it taking that discovery forward.

This is all getting a bit vague and crystal ball gazing now....perhaps we will just have to wait and see....but tell you what...I will only be upgrading from my quad core if something comes out with high clocks not just more cores.

Off to bed now...goodnight.

I understand it we should have another 5+ years before we get to where the current shrink and add transistor's philosophy stops working due to atomic limitations. At that point we'll see further process improvements, design optimizations, increased ipc, etc but progress will likely slow down considerably.
There's many alternative technologies, but none are nearly as mature or well understood as what we use now so it's hard to say. But if you look at other parts of computer history, the floppy disk was eventually replaced by cd's/dvd's and flash drives, hard drives are beginning to be replaced by flash drives as well, crt monitors were replaced by lcd flat panels, etc. you can go way back to punch cards, vacuum tubes, analog modems... There's very likely something that will replace our current technology once it's gone as far as it's going to. But I think we have 5-10 years maybe more before that will happen. But if you're desktop is a 16 core with 16gb of ram which is probably going to be commonplace in 5 years, maybe computers don't need to get faster. Sure some people can benefit from 96 core 512gb of ram etc and those should be available, but I'm guessing most of the world will be very happy with 16 cores and 16gb of ram :-), should run windows 7 ok, verdict is still out for windows 8 though :-)

You're guilty of reading only the headlines and not the details. Lets take two of your examples.

Firstly Quantum computing. Why don't we have those on our desktops? The technology is still in the labs and far from complete. Researchers have managed to build a 2 bit quantum processor that can handle some simple instructions, but for no more than a microsecond. That sounds like something you can have on your desktop tomorrow doesn't it? (NOT!) There's LOTS of research needed before this technology is anything like ready for practical use. I'd guess at at least 10 years worth. See http://www.theinquirer.net/inquirer/news/1431915/quantum-computing-closer

OK, what about the 500GHz transistor then. Firstly it can only manage that speed if you supercool it (-268C is rather beyond the capability of even vapor cooling). At room temperature it can reach only 350GHz. So, you say, I'll take a 350GHz CPU, that will do nicely. the reasons are as follows :

The speed of an individual transistor, when pushed to extremes, does not give you anything like the clock speed of a processor. Conventional CMOS transistors have been demonstrated running at speeds of 150-200GHz and yet here we are with CPUs running around the 3GHz mark. Millions of transistors must run in synchronization in a modern CPU. In each computational element signals must propagate through multiple transistors (including the delay of traveling BETWEEN transistors over metal wires). Your actual clock is a small fraction of the maximum switching speed of an individual transistor.

The second problem is that it is the wrong type of transistor. The transistors in question are analog bipolar transistor designed for communications work. Bipolar transistors are not well suited to high density, high speed digital logic. You CAN use them for it, but they consume much more power at a steady state, so the whole chip becomes very hot, not just the areas that are doing work.
See http://episteme.arstechnica.com/eve/forums/a/tpc/f/174096756/m/250005579731?r=150006789731#150006789731

So beware the headline grabbing news release from a lab. Even if what they say is practical and applicable, it is most likely many years until the technology can be used in mass-produced systems.

If I'm wrong then I'm big enough to stand corrected...I'm not invovled in IT in anyway other than having an interest in following the latest hardware developments.
It's just my opinion that the current pursuit of ever increasing parallelism is ultimately a dead end...perhaps it is just the way things have to go as a particular process nears the end of its development capabilities.

I don't just restrict myself to the reading of headlines but admit that often the detail is beyond my level of knowledge...you are obviously much more knowledgeable in the subject than I and I respect your opinions....but I still won't upgrade my quad core for a while or at least until the software environment evolves considerably. :)

Chips like this just seem to prove that the industry is still trying to make the perfect PDP-11.

..running on this kit was the Windows 7 "3D Text" screen saver. Sorry, kids. Still choppy as hell. Try again in 20 years! ^__^

to Technogiant,

How many programs / browser tabs you have open? Multiply that like 10 and you get the core count you might need for good performance. Even then some background programs/processes could use more cores...

Frequency is stalling because of silicon, for better frequency we need graphene.

In short: More cores please!

Just to second HowMany's response and to add that I see no reason not to have as much parallelism on the CPU side as on the GPU, there ae precious few tasks that can't be converted to run in parallel assuming you have full access to main memory and no restriction on code size which are the problems with programming for GPUs.
Personally I don't care either way as long as things get faster overall - my main interest here is rendering fractals and that can generally just as easily be accelerated by using more cores/threads as having higher clock speeds.

Some tasks lend themselves readily to parallelism, others do not. It's not by chance that graphics processors have successfully employed hundreds of cores in parallel. If you have an image with 3 million pixels and you need to change the shading on it, all those cores can divvy up the image and work on the individual pieces independently.

Conversely, if you have a spreadsheet and cell B1 depends on A1 and C1 depends on B1, etc, it doesn't matter how many CPU cores you have. Calculating proper results is a sequential process. Only a faster CPU will help.

I'm a software developer and don't play games. My main system is a Pentium 4 3.4 GHz dual core with 4 GB. With 2 email clients, some browsers, WinAmp and an IDE (programming environment), I'm only using between 2% and 4% of CPU. When I kick off a build on a large project, that goes up to 50%. I doubt the average non-gamer home user would ever get above 5% CPU usage on a similar system.

Even with today's fastest memory, the CPU is doing nothing for 10-15 clock cycles waiting for memory to respond. That's why all new processors have multiple memory channels. That's no panacea; orchestrating all those channels to keep the CPU busy is an inexact science.

Probably the biggest boost to performance with existing technology would be to figure out how to get memory to run at the same frequency as the CPU. Easier said than done, obviously.

There are several things I find objectionable about your reply.

"Conversely, if you have a spreadsheet and cell B1 depends on A1 and C1 depends on B1, etc, it doesn't matter how many CPU cores you have."

You're not likely to find useful spread-sheets that need to iterate millions of times on a single input. Particularly ones where the loop carried dependency isn't an artifact of bad programming.

If you're running a time consuming spread-sheet application it is most likely because you have a large data set and you want to run the spread-sheet once on each element. This is easily ammenable to paralellization.

"Even with today's fastest memory, the CPU is doing nothing for 10-15 clock cycles waiting for memory to respond. That's why all new processors have multiple memory channels. That's no panacea; orchestrating all those channels to keep the CPU busy is an inexact science."

This is incorrect for several reasons.

RAM memory is more like 100-200 clock cycles away; 10-15 cycles latency is ~L2 cache latency. The CPU is pretty good at hiding latency for most kinds of tasks by using multiple layers of successively larger, slower caches, by taking advantage of instruction level paralellism with OoOe and register renaming.

Multiple memory channels and newer DRAM types increase bandwidth; they do not appreciably lower the latency. DDR2 and DDR3 use more multiplexing to increase bandwidth(essentially the same thing as adding more memory channels) but they don't appreciably increase the core frequency or lower the latency of the memory cells themselves.

Higher bandwidth is very useful for a particular class of tasks that have high degree of data locality and high degree of data parallelism. When you have both something wonderful happens; you can stream data from memory in a sequantial fashion and memory latency becomes completely irrelevant. In stream processing you are limited by either the arithmetic resources or memory bandwidth. When the compute intensity is below roughly 50 to 1 arithmetic ops to I/O ops you will benefit from higher bandwidth.

Most types of signal processing benefit very much from stream computing.