Haswell-E Will Have Hidden Cores

29

Comments

+ Add a Comment
avatar

biggiebob12345

I don't know what idiot thinks that Intel is going to be defeaturing a 14 core die down to 8 cores, but it isn't going to happen. If 6 cores are dead the die is getting killed, not defeatured.

More likely than not, the 8 cores are going to be built as 8 cores with the Xeon-bits disabled (ECC memory, etc) and the 6 cores are going to be the defeatured 8 cores.

avatar

Hey.That_Dude

Only slightly true. They have been known to neuter chips before with lasers. My bet is that there will be an 8, 10, 12, 14. The 8 will be a neutered 10 and the 12 will be a neutered 14 core. (That's just my bet.)

avatar

SodaAnt

I think this post is confused. What it should be is that haswell-e will have 14 cores on the die, but the consumer desktop version will have only 8 or 10 enabled. This is like sandy bridge-e where the desktop chips had 8 cores on the die but only six enabled. Only the server versions will have all cores enabled.

avatar

SodaAnt

I think this post is confused. What it should be is that haswell-e will have 14 cores on the die, but the consumer desktop version will have only 8 or 10 enabled. This is like sandy bridge-e where the desktop chips had 8 cores on the die but only six enabled. Only the server versions will have all cores enabled.

avatar

Audiogabby

What socket number are we talking about with this Haswell-E processor? Still 1150?

avatar

riopato

It's most likely an unannounced socket since SandyBridge-E and Ivybridge-E are socket 2011. The fact that Haswell is already on a new socket 1150, Haswell-E will most likely be on an even newer socket (maybe socket 2014?)

avatar

devin3627

doesn't anybody remember that more cores = less performance? fragmentation = latency or basicly... thats the difference between 1 hand vs two, the more you have weighs you down... this design prooves intel couldn't figure out how to have extra cores and maintain performance benefit without latency. no one does physically more than four things at once.. dual-core was revolutionary growing that extra hand for background services. YES! but the more than 2 cores are less necessary in mainstream use. the difference is less and less the more cores you add, the more its a trade-off due to heat and hertz/cache distribution. there's nanotechnologies out there yet to come into practice.

avatar

riopato

You maybe confusing the more cores= less performance with AMD chips. Intel worked around this issue I believe by etching multiple cores on one die along with integrating the memory controller.
Having multiple core is a boon for video processing, Photoshop, 3d and pretty much any kind of heavy duty multitasking for people who need this to do actual work, not to just play games. Even so, I wonder what kind of gaming would it be if games actually took advantage of multiple cores?

avatar

0ly1r3m@1ns

Meh is all I can say.

I'm quite happy with my FX-8350, crazy fast, lots of cores.

Intel really can't beat the FX-8350, in cheap, it goes neck and neck with CPUs in its price range for gaming, just dominates most chips in other categories.

avatar

Viperz

I've just upgraded to an i7 3770k from a 2600k and it's the fastest thing short of a Haswell. I had a friend that actually purchased the FX-8350 and after I told him to do some research he sent it back to Newegg and bought the 3770k as well.
Don't believe me? Google FX-8350 benchmark and look for TomsHardware.

And before you start complaining about price, I picked it up at MicroCenter for $229.99 (about $30 more than the FX-8350) and they knocked $40 off my Gigabyte GA-Z77X-UD5H which was already on sale for $179.99, so that's Mobo & CPU for $369.98.

avatar

limitbreaker

Amd makes some pretty decent chips for the price, you'd have to be crazy to get a i5 instead. It's absolutely insane, specially with all recent games being multi threaded. Crysis 3 runs better with 8 cores :-)

avatar

nadako

The i5 still beats amds 8 core processor its not how many you have but how you utilize each one.

avatar

limitbreaker

thats not true, amd 8 core flat out beats the i5 in almost all test. There are only a few exceptions that work better on the i5 and they are usually instances where fps is so high that it doesn't matter. All new games are well multipthreaded and upcoming games even more so. Another argument is that pc's arent consoles and should not be used like one, the synthetic benchmarks you see never show you how they perform with other tasks running in the background. I personally do many things at once and i need a system that can handle the load without compromise and only amd 8core and i7's are up for the job.

avatar

Danthrax66

http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/5 You have that backwards Intel wins in almost all tests and AMD is the niche.

avatar

limitbreaker

Actually I didn't, go read my post a few more times and maybe you'll eventually get it.

http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/3

avatar

AFDozerman

I hate to keep commenting on this article, but I feel like I have a lot to input. That being said, not only is pretty much every piece of software out there optimized for multithreading of some sort, but AMD's floating point units in the FX series are real monsters, usually keeping up with the i-whatever's in single threaded tasks, and that's not even taking into account the work AMD did in implementing AVX and it's own XOP instruction sets.

avatar

PCLinuxguy

Agreed. My upcoming rig will be running piledriver unless the next gen comes out before I'm at that stage. I'm just going between the 6300 or 8350. mainly I've just been comparing the TDP wattage on them to see if I want to go for an 8 or a 6 core. overall though I find that AMD is still the underdog and underrated. As I've said before, benchmarks are nice but in real world 'driving' the end user won't know the difference in performance. I've test driven a SB based i7, and the first gen FX 8 cores on a pair of rigs and both were blazing fast but neither one had a leg up over the other. This is why I go with AMD. great chip for less cash that can do more than it's given credit for.

avatar

ThomFrost

sounds like might be 8 for general processing and the other 6 for maybe graphics processing.
thats what i would do.
so use the 6 extra cores for improved graphics processing.
just a thought.

avatar

AFDozerman

I highly doubt it. If they were execution units, only six EUs would provide abysmal graphics performance, and if they were X86, things would be even worse. The only API that I know of that can even render through X86 is openGL and only after drivers and specific code are written. Even then, the performance would be terrible. No, this is a case like sandy bridge-E where the Xeon variants had 8 cores followed by laser cutting and clock goosing to create SBE with six.

avatar

limitbreaker

That's why they have socket 1155/1150 chips... Mainstream cpu with integrated gpu. Leave the high end stuff unpolluted.

avatar

Caboose

Intel Shit IGP is still shit IGP. Even if it has "6-cores"

avatar

PCLinuxguy

though normally the -E chips don't have graphics cores. but even with Haswell's updated IGP, the IGP in AMD's Richland APUs (such as the A8 and A10) have far superior graphics despite haswell being slightly newer. Sure it might not be a "great cpu" to some people but it's still dominates Intel when it comes to graphics.

avatar

Julian Reiche

Haswell currently maxes out at 40 EU or "cores" in its iris pro variants, the rumour is for actually CPU cores, looking at the die size of current haswell chips with the 40 execution units you can see how you could simple use that space for more cpu cores

I also assume these cores will be laser etched to ensure no software could re-enable them :)

avatar

Hey.That_Dude

Sounds like that means the Xeon Haswell-E chips will pump out 14 cores at the maximum. That's pretty sweet. Even if the chip cost twice as much as the 8 core part it's almost a 50% increase in cores, so it might be worth shelling out for it.

avatar

AFDozerman

Depends on clock throttling. Could be that flops don't go up all that much even with 50 percent more cores.

avatar

Hey.That_Dude

My guess it that the 22nm 3d transistor tech they have will have matured by the time they put this out there. I bet there will be about a 5-10% increase in efficiency. As for throttling, it might be smart enough to change its clock depending on the load it's fed. After all, no CPU is perfect.

avatar

devin3627

you are on a role. new technologies are due.

avatar

AFDozerman

I can really imagine that it's dependent on a couple of factors carrying on from regular Haswell. The first, and most likely to carry over, is the integrated voltage controller, which actually adds to the thermals significantly. It's being blamed for the poor overclockability of regular Haswell, which is even worse than IVB. The second that I can think of, and most likely to not carry over, is the poor thermal paste that Intel keeps putting on their chips. I'm pretty sure the E-series won't have this problem, but you never know. I guess the only way to be sure is to wait and see, though.

avatar

Hey.That_Dude

That sounds more like a "No one's ever done this before so lets try these gates set up like this and hope that the heat isn't TOO bad, after all the computer model says this should work great"
Trust me, there are a bunch of ItoF and FtoI converters on your chip that don't do that much in the grand scheme of things and they take up a bunch of room, but when you need them because some programer can't figure out how to get around converting they can save you HUNDREDS of cycles in compute time. Just like these voltage regulators will keep these chips a lot safer from the increasingly crappy power signals we're getting (after all your power supplies aren't perfect). Not to mention that as the voltage requirements continue to plummet, on-board regulators will be WAY better than an external device (seriously, 1.2V is fairly far from 0.7V... but how about when we get down to 0.9V or 0.8V then a 0.2V change is the difference between a working processor and a failed one.