Want an unlocked 12 core Extreme Edition?

[alkemyst]

power.

future sales.

 

[Homeles]

future sales.

Conspiracy theory.

 

[ShintaiDK]

12 cores overclockable? Ridiculous. I mean how big is the market really for users who would need 12 cores anyway? Ehhh I guess I can't be mad because there are folks out there in the world who do but I'd much rather have seen a more marketable product. A bad ass 8 core would do just fine.

The market for an unlocked enthusiast 12 core is so incredible tiny that Intel wont release it. And Francois is trying to create a twitter PR reason for one.

We get 8 cores next year with Haswell-E. Something that ca actually match performance expectations, unlike a 12 core.

 

[NTMBK]

The market for an unlocked enthusiast 12 core is so incredible tiny that Intel wont release it. And Francois is trying to create a twitter PR reason for one.

We get 8 cores next year with Haswell-E. Something that ca actually match performance expectations, unlike a 12 core.

Oh those 12 core parts can definitely meet performance expectations... just not for gaming.

 

[ShintaiDK]

Oh those 12 core parts can definitely meet performance expectations... just not for gaming.

Plenty of other things besides gaming too. Since not all apps can scale to 12 cores and 24 threads. But again, gaming is the main driver for the enthusiast class.

 

[NTMBK]

Plenty of other things besides gaming too. Since not all apps can scale to 12 cores and 24 threads. But again, gaming is the main driver for the enthusiast class.

I can't think of many apps which demand high performance (other than gaming) which don't scale well with core count. Rendering, image editing, video editing, video transcoding, scientific computing, software compilation (on any well laid out project)... Are there any good counter-examples? (Genuine question, I'm curious.)

 

[ShintaiDK]

I can't think of many apps which demand high performance (other than gaming) which don't scale well with core count. Rendering, image editing, video editing, video transcoding, scientific computing, software compilation (on any well laid out project)... Are there any good counter-examples? (Genuine question, I'm curious.)

As soon as you go away from the imagine/rendering segment, the amount of highly multithreaded applications drop like a stone. Other applications that can scale, might be slower with 12 cores at say 3.5ghz than 8 cores at 4Ghz.

Huge excel sheets for example demand high performance and doesnt scale. Client-server applications dont tend to scale, if concurrency can ofset it. While its an atrocity, one could simply mention Access. Compiling in VS2012 is also faster on a quadcore than a slower 8 core for example due to the scaling issue.

 

[NTMBK]

As soon as you go away from the imagine/rendering segment, the amount of highly multithreaded applications drop like a stone. Other applications that can scale, might be slower with 12 cores at say 3.5ghz than 8 cores at 4Ghz.

Huge excel sheets for example demand high performance and doesnt scale. Client-server applications dont tend to scale, if concurrency can ofset it. While its an atrocity, one could simply mention Access. Compiling in VS2012 is also faster on a quadcore than a slower 8 core for example due to the scaling issue.

Ah, massive spreadsheets- that eternal bane of the office worker. D:

But compilation in VS2012- it can actually be incredibly parallel, if you structure your solution file (and code) correctly. The more recent Visual Studios will not only compile multiple projects in parallel, but also multiple .cpp files within an individual project in parallel. Break things up into multiple smaller cpp files, instead of individual monolithic ones, and you will start hitting those cores hard. I know first hand that compiling a decent sized app will easily eat up 16 cores!

 

[ShintaiDK]

Ah, massive spreadsheets- that eternal bane of the office worker. D:

But compilation in VS2012- it can actually be incredibly parallel, if you structure your solution file (and code) correctly. The more recent Visual Studios will not only compile multiple projects in parallel, but also multiple .cpp files within an individual project in parallel. Break things up into multiple smaller cpp files, instead of individual monolithic ones, and you will start hitting those cores hard. I know first hand that compiling a decent sized app will easily eat up 16 cores!

Firefox compile in VS2012 is an example of a fast quadcore easily beating a slower 8 core Xeon. But where the 8 core Xeon got more computational power so to say. So while you see 100% usage, it might simply not scale 100%.

 

[Nothingness]

As soon as you go away from the imagine/rendering segment, the amount of highly multithreaded applications drop like a stone. Other applications that can scale, might be slower with 12 cores at say 3.5ghz than 8 cores at 4Ghz.

Huge excel sheets for example demand high performance and doesnt scale. Client-server applications dont tend to scale, if concurrency can ofset it. While its an atrocity, one could simply mention Access.

I definitely agree. There also is a point at which even for multi-threaded apps one will be slowed down by some shared resource. The main example that comes to mind is memory controllers and memory itself. You'll also have more threads competing for level 3 cache.

Compiling in VS2012 is also faster on a quadcore than a slower 8 core for example due to the scaling issue.

Do you have a link about such a study, I'm very interested I wonder where the sweet spot for number of cores vs frequency is.

 

[NTMBK]

Firefox compile in VS2012 is an example of a fast quadcore easily beating a slower 8 core Xeon. But where the 8 core Xeon got more computational power so to say. So while you see 100% usage, it might simply not scale 100%.

Different for different solutions, depending on how well it's been structured. Certainly the stuff we compile, a 16 core machine finishes _much_ faster than an 8 core machine. Can't comment on how common that is though. *shrug*

 

[ShintaiDK]

Different for different solutions, depending on how well it's been structured. Certainly the stuff we compile, a 16 core machine finishes _much_ faster than an 8 core machine. Can't comment on how common that is though. *shrug*

Exactly, it was just an example tho. But again, it erodes the foundation for the "enthusiast" 12 core. And where you would just buy Xeons if you had the need.

 

[SunRe]

I can't think of many apps which demand high performance (other than gaming) which don't scale well with core count. Rendering, image editing, video editing, video transcoding, scientific computing, software compilation (on any well laid out project)... Are there any good counter-examples? (Genuine question, I'm curious.)

You will be surprised how badly image processing software scales with no of cores. I'm talking Photoshop and Lightroom included, not to mention other garbage like proprietary raw processing software, etc. This kind of software actually scales better with frequency, although there are tasks where multiple cores could be used. I can link to data if interested in on this topic.

 

[Nothingness]

You will be surprised how badly image processing software scales with no of cores. I'm talking Photoshop and Lightroom included, not to mention other garbage like proprietary raw processing software, etc. This kind of software actually scales better with frequency, although there are tasks where multiple cores could be used. I can link to data if interested in on this topic.

I'm interested

 

[SOFTengCOMPelec]

But it is also a kind of chicken and egg situation.

If Intel made their top (consumer) chip a combined, quad core, high freq (3.5 GHz), AND also a 64 core (256 thread!), 1.1 GHz set of cores AS WELL, i.e. total of 4+64 = 68 cores. (At a MASS MARKET PRICE, i.e. <=$300..$500???, if that is even possible)

Then (chicken and egg), eventually software (where it is possible), would be produced to use such a beast, and some of the technical problems in scaling software up to use 64 cores would be solved.

N.B. The above part (if you are NOT already aware of it), is actually something which already exists!

Disclaimer: NOT on a single piece of silicon, yet!
But as 2 separate parts, which "COULD" be combined into a chip, in theory, at least.

Proof (they do more expensive, better ones as well) :-

Before buying these, make sure that your lottery ticket numbers have come up!

 

[ShintaiDK]

Even if Intel released a theoretical 256 core 4Ghz Haswell tomorrow at 200$. You still suffer the same issues of serial code and Amdahls law.

Even with 90% parallel code, a 1024 core CPU is only 10 times faster than a singlecore. And roughly 4 times faster than a quad.

Its not the chicken and the egg situation. Its simply software limitations.

 

[SOFTengCOMPelec]

Even if Intel released a theoretical 256 core 4Ghz Haswell tomorrow at 200$. You still suffer the same issues of serial code and Amdahls law.

Even with 90% parallel code, a 1024 core CPU is only 10 times faster than a singlecore. And roughly 4 times faster than a quad.

It can and is efficiently done, in some cases already.
A good example (except that it is somewhat naturally parallelizable, weakening my arguments) is gaming, and graphics cards.
I.e. (if you can afford the very high price at the moment) you can take many games (on the PC), and if you have high end enough graphics cards, and multiple ones, and enough output screens, with high enough resolution, you can go really wild with parallelism, and many games today.

When enough resources are piled up, and the time and clever engineers are set to the problem (as has already been done in the gaming world, because of the huge market and big $'s that can be made with them). Good results can be achieved.

I fully agree that there are "laws of computing" limits to what can be achieved, but nevertheless, clever solutions (as have been created in the gaming world), can be invented and implemented, which will allow some of the software we use, to scale to very large number of cores/threads, with the resultant performance speed up.

Borrowing Idontcare's terminology, I guess that gaming has been the low hanging fruit of the parallelizable world.
How we move on, and allow more hardware to be highly parallelizable (e.g. 64 core, affordable cpus), and pay for very advanced software to utilize so many cores (where it can overcome the natural limits of the "laws of computing").

Let's put it another way, do you think in 10,000 years time, we will still have mainstream "QUAD" core cpus, at around "3.5 GHz" ?
I think NOT, so sooner or later, these technical problems will be solved.

A century year or 2 ago, when I was much, much younger (without revealing my age), chess computers were pretty hopeless at playing Chess (1970s/1980s). They used slow 8 bit, 1..4 MHz early cpus, such as the 6502/6800/Z80 etc, which were phenomenally slower than today's processors.

Approx 20 years later, they were beating even the World (human) champion at chess.

I doubt we are anywhere near the limits of what cpus can achieve yet, by a very long mile.

 

[ShintaiDK]

It can and is efficiently done, in some cases already.
A good example (except that it is somewhat naturally parallelizable, weakening my arguments) is gaming, and graphics cards.
I.e. (if you can afford the very high price at the moment) you can take many games (on the PC), and if you have high end enough graphics cards, and multiple ones, and enough output screens, with high enough resolution, you can go really wild with parallelism, and many games today.

I dont think I understand your argument. So could you please elaborate?

Gaming is a prime example of poor scaling applications. BF4 that is in the front line if you exclude chess games, only scales due to the multiplayer part. Run it in singleplayer and you are back to the same problems again.

 

[SOFTengCOMPelec]

I dont think I understand your argument. So could you please elaborate?

Because gaming is one of those thigs that scales poorly. BF4 for example only somewhat scale due to the multiplayer part. Aka concurrency. But in singleplayer its back to old fashion.

Sorry, I was mixing together two different concepts at the same time, which has caused confusion, and I was partly wrong! in what I said, as well (due to my mix up in the explanation).

Let's split a typical modern game, into two sections, one section PURELY does the raw, screen graphics.
The other section does everything else, including physics, player interactions, monsters, logic, AI, non-visual aspects of game, etc etc.

The pure/raw graphics processing (i.e. transformations on (various obviously) 10,000,000 pixels (ultra approx), is (in very simple terms, and NOT always as I am about to say), i.e. changing pixels, drawing triangles/polygons, anti-aliasing, etc etc, can (but NOT always) be highly parallelizable with modern graphics hardware.
In other words, a good quality, well written game, will (typically, on a PC), allow going from 1080P, up to 1440P, and maybe more, as long as you have got powerful enough graphic(s) cards, and the particular game allows this.

BUT the other aspects of the game (as I think you were thinking of), MAY NOT allow such multi-core/thread/gpu multitasking to take place. This depends on which game it is (but there is a basic similarity).

---------------------------------------------------

I.e. Some aspects of the graphics (of e.g. games) CAN be speeded up with multiple and very powerful graphics processors, and the necessary cpu power, to keep it running.

But other aspects of the graphics and/or game, are cpu bound, and can only mainly use extra Frequency/IPC of the cpu, and "maybe" some extra cores (in a limited way, in "some" cases).


--------------------------------------------------

I really meant the "graphics" as in updating pixels/triangles/polygons on the screen, NOT other aspects of the graphics and cpu processing of games.

Sorry for my faulty/partly-incorrect, initial explanation.

I should of said raw-graphics-pixel-updates (are usually fully scalable, within limits), rather than saying "games" which was way to broad a term.

 

[SOFTengCOMPelec]

TL;DR; Summary

Low level (raw) graphics processing and some higher level graphics functions, such as fill these 400,000 pixels = Usually parallelizable

Rest of the High level graphics, and other game functions, such as AI, physics etc = Much less or even NOT parallelizable

 

[Makaveli]

What on earth even makes much use of 12 cores? I wouldn't want it unless a good amount of software and games can use all 12.

There is a need for it just not for you.

You need to get out of the gamer/Enthusiast mind set when thinking about xeons.

I would think getting a 8 core chip on the market would be more in the gamer/enthusiast space than a 12 core chip.

 

[ShintaiDK]

Sorry, I was mixing together two different concepts at the same time, which has caused confusion, and I was partly wrong! in what I said, as well (due to my mix up in the explanation).

Let's split a typical modern game, into two sections, one section PURELY does the raw, screen graphics.
The other section does everything else, including physics, player interactions, monsters, logic, AI, non-visual aspects of game, etc etc.

The pure/raw graphics processing (i.e. transformations on (various obviously) 10,000,000 pixels (ultra approx), is (in very simple terms, and NOT always as I am about to say), i.e. changing pixels, drawing triangles/polygons, anti-aliasing, etc etc, can (but NOT always) be highly parallelizable with modern graphics hardware.
In other words, a good quality, well written game, will (typically, on a PC), allow going from 1080P, up to 1440P, and maybe more, as long as you have got powerful enough graphic(s) cards, and the particular game allows this.

BUT the other aspects of the game (as I think you were thinking of), MAY NOT allow such multi-core/thread/gpu multitasking to take place. This depends on which game it is (but there is a basic similarity).

---------------------------------------------------

I.e. Some aspects of the graphics (of e.g. games) CAN be speeded up with multiple and very powerful graphics processors, and the necessary cpu power, to keep it running.

But other aspects of the graphics and/or game, are cpu bound, and can only mainly use extra Frequency/IPC of the cpu, and "maybe" some extra cores (in a limited way, in "some" cases).


--------------------------------------------------

I really meant the "graphics" as in updating pixels/triangles/polygons on the screen, NOT other aspects of the graphics and cpu processing of games.

Sorry for my faulty/partly-incorrect, initial explanation.

I should of said raw-graphics-pixel-updates (are usually fully scalable, within limits), rather than saying "games" which was way to broad a term.

But CPU load barely if at all changes between running 480p and 2160p. And all the settings is just GPU load.

 

[SOFTengCOMPelec]

But CPU load barely if at all changes between running 480p and 2160p. And all the settings is just GPU load.

Exactly!. The point I was trying to illustrate is that parallelism CAN be usefully done, as it is in the GPU.

When the need arises, which it may well, in the coming years, as games get better and better, and later cpus continue to be frequency/IPC bound, on a per core basis. Then solutions to the "Amdahls law" etc issues will hopefully be developed, allowing future games to become increasingly more powerful (in the multi-core cpu utilization sense).

Example
Microsoft/Sony who hopefully know a little bit about game software development, chose an 8-core next generation console solution, even though, if they had really wanted, a (probably) similar performance dual core cpu (at a much higher frequency and IPC), could have been fitted (probably at greater expense).
This could mean that Microsoft/Sony are very confident that 8 cores (I know it is rumoured that less than 8 cores are to be used for gaming) can be effectively used to make modern/competitive video games.

Assuming they are right (and ignoring that NOT all 8 cores are necessarily used), then 8 core Haswell-E's, should work very well, and give huge gaming performance gains.
If the game is written flexibly enough, it may even be able to use all 12 cores (or even 24 threads) of the Xeon cpu, we are suppose to be discussing in this thread.

Sadly the tiny market share of gamers with 12C/24T cpus, means that such a market would be largely ignored/forgotten by the games manufacturer's.

Anyway, you have raised a very good point, which is that stuff such as "Amdahls law" may mean that even if the game tries to use 12C/24T, it may need some very big technical issues to be solved first, which may take a long time (years+) to solve, or even be partially/fully insolvable.
E.g. Promised around 1950, in 50 years time, We were suppose to have cheap Fusion energy by now, and there isn't any (I'll just go and sit in the sun tomorrow joke).

 

[videogames101]

I just want an affordable 8 core mainstream K processor. Too much to ask?

4770k = $300 (close enough for this)

So price per core for 4770k = $75

Lets say yield is 80% for the 4770k die. (change this to whatever you want, just a wild stab at it)

Hypothetical 8 core cpu = 8 cores * $75 per core * (1 / .8) for decreased yield = $750

How would this ever be "affordable"?

I imagine this is why there isn't one on the market.

Obviously simplified, but you get the picture I think. You guys act like releasing a X-core CPU at a reasonable price is just a matter of slowing the clocks down, which in reality has almost nothing to do with it. Extra die size costs money. It reduces yield and reduces the number of dies per wafer. It will not end up being affordable without stripping down the cores' functionality to save die area. There are real examples of this, see AMD's 8 "core" CPUs.

Forget software, this is economics.

An affordable 8-core CPU is not going to give Intel 60% gross margin, so they aren't going to release one. Not with Haswell cores anyways.

 

[Kougar]

The more choices the better. A 4770K OC'd just wasn't fast enough to run my VM workloads.

Pricing would probably be too high for me unless I won the lottery, and I think a dual-socket board be way cheaper. But at least people would finally have the option... they'd better not disable VT-d on it though!