Sunday, June 16th 2013

Haswell-E - Intel's First 8 Core Desktop Processor Exposed

Another day, another Intel leak and a few surprises as well. During the last few days we covered Intel's desktop roadmap for the next twelve months, bringing you news and insights on Intel's plans for the aforementioned time interval. Today we bring you news on what's to follow in the second half of 2014, specifically, on Intel's Premium Desktop plans for the interval, namely Haswell-E, DDR4 and the X99 PCH.

Haswell-E will be Intel's last and best offering using the 22 nm fabrication process, it will come in two versions, core count wise, 8 core part(s) as well as 6 core part(s) with hyper-threading enabled, therefore, boasting no less that 16 execution threads for the 8 core chips and 12 execution threads for the 6 core version(s). Judging by that alone, Haswell-E should constitute a far superior upgrade over Ivy Bridge-E, compared to what the latter will be in relation to Sandy Bridge-E, Haswell-E offering two additional physical cores that translate into four additional execution threads. The new chips will boast 2.5 MB of L3 Cache per core, summing up to 20 MB total L3 cache for the 8 core parts. TDP will remain in the same neighborhood it was in the case of its predecessors, around 130-140 W.

Haswell-E will of course be accompanied by a new platform, dubbed Wellsburg, the X99 chipset will bring a host of new features, the most important one being quad channel DDR4 support. With four basic frequency settings, starting at 1333 MHz and moving up in increments of 266 MHz to a maximum of 2133 MHz, at which point overclocking should be employed to attain superior clocks. However we sincerely doubt that any DDR4 modules/kits, clocked below 2133 MHz, will be made available for this platform. Modestly clocked (1333 MHz), low voltage (1.2 V) kits are supported by the new platform as well. The DIMM connector was also modified to support Non Volatile DIMMs, receiving four more pins for the purpose (288 instead of 284), modification that will not negatively impact compatibility with 284 pin modules in any way.

Other points of interest regarding the X99 chipset are:
  • Up to six USB 3.0 ports
  • Up to eight USB 2.0 ports
  • Up to ten SATA 6 Gbps ports
  • Integrated Clock support
  • TDP of 6.5W
The LGA 2011 socket will be updated too, dubbed LGA 2011-3, the socket will see the pin layout changing while remaining numerically the same. The change, going by Intel's own slides, claiming superior efficiency. The chip's IHS received a makeover as well, looking very different from current Intel offerings.

Given the information at hand, trying to quantify performance gains, speculate on overclocking potential and other such conclusory attempts to wrap up the above presented information, I admit is quite enticing and intriguing, but I'll end here and outsource the pleasure of doing that to you.

Post Scriptum
A big hand for radrok, for bringing this to my attention.Source: VR Zone
Add your own comment

77 Comments on Haswell-E - Intel's First 8 Core Desktop Processor Exposed

#1
HammerON
The Watchful Moderator
My next upgrade:)
Posted on Reply
#2
Aquinus
Resident Wat-man
by: buggalugs
Anyway the current X79 platform sucks, the sooner they kill that off the better. Maybe Haswell-e will save the highend.
You just don't learn do you? 40 PCI-E slots are there to be used and last time I checked the 4770k and 3770k aren't all that much faster than the 3820 so I don't see what your beef is. X79 is just like z68 and z77; it's an over-glorified PCH. Anyone who really wants good performance out of IO and really needs more than what X79 is offering should be looking at getting a better SATA controller and if you're going to complain about price, I'll say once again, you're investing in enthusiast level hardware, you do the math.

As an owner of a SB-E rig I can say, once again, that your hate for skt2011 and X79 is really pretty ridiculous. Even more so for someone who has never owned a skt2011 machine, so you really have no idea what you're talking about because it feels just as responsive as a skt1155 rig. I don't know what you're beef is, but skt2011 does not suck. Just because it doesn't have everything you ever wanted doesn't mean it is bad and unlike you, I actually have a ton of devices plugged into the PCH and I do a an X79 board and I can say that it performs admirably, so don't talk about what you don't know.

--

As far as the new Haswell-E revision of skt2011 is concerned, that would have been expected with the introduction of DDR4 so I'm not surprised in the slightest. I would also imagine that Haswell-E is probably, at bare minimum, at least a year away.

by: Melvis
Ok yes for DDR4 but that's just silly.
Why? DDR4 works very differently than DDR3. They're ditching channels in favor of point-to-point links. That requires redoing the circuitry.
Posted on Reply
#3
jihadjoe
by: Melvis
I can under stand I guess with this new 8 core as its intels first and all, but i just don't get it with socket 1150, that socket is going to be very short lived?
Broadwell will be the drop-in upgrade to socket 1150, just as Ivy was the drop-in upgrade to 1155.

2011 is a completely different platform.
Posted on Reply
#4
adulaamin
"All Processors Unlocked" <<< :toast:
Posted on Reply
#5
progste
i thought there were already 8 core Xeon CPUs?
Posted on Reply
#6
Aquinus
Resident Wat-man
by: adulaamin
"All Processors Unlocked" <<< :toast:
Does it really matter? The 3820 showed that even locked SB-E chips can overclock plenty well if you kick the bclk strap up to 125Mhz. I would much rather have an unlocked i3 CPU as opposed to having all skt2011 chips being K edition CPUs. Heck, I run 4.5Ghz on my 3820 and I don't even touch Turbo to do it!

by: progste
i thought there were already 8 core Xeon CPUs?
There are, but they're not overclockable at all afaict. I don't even think they will let you adjust the bclk strap. :(
http://ark.intel.com/products/64597/Intel-Xeon-Processor-E5-2665-20M-Cache-2_40-GHz-8_00-GTs-Intel-QPI
Posted on Reply
#7
Melvis
by: Aquinus
Why? DDR4 works very differently than DDR3. They're ditching channels in favor of point-to-point links. That requires redoing the circuitry.
Can you give me a link to this? the difference etc? Id like to learn more about it. All ive seen since DDR1 is just speed increases, nothing more :ohwell:

by: jihadjoe
Broadwell will be the drop-in upgrade to socket 1150, just as Ivy was the drop-in upgrade to 1155.

2011 is a completely different platform.
Ok yep I get ya and i can see that now, I don't keep up with Intel that much.

They realy should of just kept skt 1155 for Haswell and Broadwell as well, that would of been a realy good socket to keep then? Not like there is a huge difference between them all is there?
Posted on Reply
#8
Rayz
What

No SATA Express?! The next extreme generation won't support it while the new mainstream one (Z97) will?! someone there in Intel got crazy. The next SSD's will be capable to exceed SATA36G speed, and the most common owners ones (the Extreme platforms users) won't be able to make a use of them?

ABSURD.
Posted on Reply
#9
Aquinus
Resident Wat-man
by: Melvis
Can you give me a link to this? the difference etc? Id like to learn more about it. All ive seen since DDR1 is just speed increases, nothing more
How much depth? You can get the JEDEC spec here but it's rather verbose.
Posted on Reply
#11
Aquinus
Resident Wat-man
by: blibba
There's really no reason Intel couldn't release a 12-core desktop part today, other than profit margins on Xeons.

Check out the Xeon E5-2697v2 when it's officially announced in the near future.
Or maybe because it will perform worse than a 4 core part for about 99% of most people's needs because of the low clock speed. If you really need a 12c rig, then go buy a Xeon but your regular consumer, even your regular enthusiast, doesn't need a 12c rig, nor would it perform better in most cases. There is zero motivation for Intel to release a 12c for your typical consumer when a quad-core does just as well.
Posted on Reply
#12
Doc41
I don't understand why haven't they done that a long time ago, i looked around their site and found a 10 core Xeon E7 that was launched in Q2'11 but hey thats intel.

Anyway still good news and i might upgrade to this if my wallet allows at the time.
Posted on Reply
#13
Over_Lord
News Editor
EDIT:

Thanks. Appreciate the quick reply.
Posted on Reply
#14
D4S4
by: de.das.dude
doensnt intel house the memory controller in the chipset?
not since nehalem
Posted on Reply
#15
blibba
by: thunderising
It's a SHAME that the AUTHOR didn't even THINK of PUTTING A SOURCE LINK to this article.

Really, a SHAME. Considering I'm the AUTHOR of the ORIGINAL ARTICLE.
It's a shame that the author of this comment formatted it in such a juvenile way. Really, a shame. Considering I'd have been sympathetic to his remarks otherwise.
Posted on Reply
#16
Naito
New socket revision most likely for DDR4 support. Currently, most Intel sockets seem to survive one tick-tock cycle, that should be long enough between upgrades. Throwing a newer CPU into an older board, or putting an old CPU into a new board would likely bring minimal increase in performance and/or features. This is especially true when newer technologies such as USB 3.0 and SATA 6GB/s are being added and the CPUs (within one tick-tock cycle) usually bring with them expanded connectivity.
Posted on Reply
#17
Am*
by: de.das.dude
i call BS on this. its just to make you all buy a new motherboard XD
long live AMD
i went from an am2 motherboard to am2+ to AM3+ with the SAME CPU XD
First of all, you're talking nonsense. There is no way on earth that you used an AM2 processor on an AM3+ board, since your Phenom is an AM3 processor -- some AM2+ motherboards were compatible with AM3 but required a NEWER PHYSICAL SOCKET to support them, most were already shipped with older sockets which were NOT forwards compatible, which means AM2+ is no better or more forward compatible than 1155 or 775 or any other common Intel socket that supports at least 2 generations of processors. You're completely confusing your processor with the Phenom II 940, which was specifically AM2/AM2+ socket ONLY. Oh and the only reason AM3 processors are backwards/forwards compatible is because they have BOTH an DDR2 and a DDR3 controller on the CPU.

Secondly, can you even read? People are complaining that Haswell-E CPUs will not work with current LGA 2011 boards, not the fact that their 3930/3820Ks are not forwards compatible with HW-E socket mobos. That's the equivalent of trying to stick a AM3 CPU in an AM2 motherboard -- not physically possible due to socket layout & feature changes. It is one thing to complain about occasional socket changes from Intel (I complained about it a fair bit back when 1156 was EOL'd & 1155 was announced, and for good reason) but for you to troll with your moronic "AMD is teh bestt cuz my CPU wurks on 2 socketz!!!1!!" is beyond retarded -- you swapped 3 motherboards to use the same slow-end CPU, and AM3+ is pretty much a fake socket, since 900 series chipset motherboards are merely re-branded 800 series anyway (with a slightly overclocked HyperTransport bus). Nothing there warrants a new socket anyway.

by: de.das.dude
and its already stated that the new socket is just a restructuring of pins. its numerically the same. this is just so that people have to buy new boards. nothing esle.
How can you possibly say something so stupid? Did you even see that this new socket will use DDR4? Or the fact that Haswell houses certain voltage regulators on-die, which easily justifies a new socket on its own? Who are you trying to kid, other than yourself?

by: de.das.dude
doensnt intel house the memory controller in the chipset?


Welcome to the 21st century. Memory controllers have been on CPUs since around 2008 for Intel, and all the way since at least 2003/04 from AMD.
Posted on Reply
#18
de.das.dude
Pro Indian Modder
yeah, looked like i had some reading to catch up to. now that i did, its clearer :banghead:
Posted on Reply
#19
happita
I love jumping ship to new hardware like the next person, but I feel like Haswell-E won't be bringing anything new that Haswell hasn't already brought to the table. Well, except 8-core variants. But then again, very, very few applications utilize 8-core CPUs. I knew holding onto my SB was a good idea. Broadwell, here I come :toast:
Posted on Reply
#20
Am*
by: happita
I love jumping ship to new hardware like the next person, but I feel like Haswell-E won't be bringing anything new that Haswell hasn't already brought to the table. Well, except 8-core variants. But then again, very, very few applications utilize 8-core CPUs. I knew holding onto my SB was a good idea. Broadwell, here I come :toast:
Forget Broadwell, Skylake/Skymont is where it's at. We may even get 8 core processors & DDR4/PCI-E 4.0 on the mainstream socket by then. Also Intel said they will be hitting the limits of silicon with that generation, which will be a very interesting time for us enthusiasts to find out what the new material/manufacturing & binning processes will be.
Posted on Reply
#21
lilhasselhoffer
So, let's take a look at this socket thing critically.

LGA 1366/1156 - Nehalem and Westmere
LGA 2011/1155 - Sandy Bridge and Ivy Bridge
LGA 2011-3/1150 - Haswell and presumably Broadwell

One socket gets two architectures (a new one, then die shrink). The grand old days of LGA 775 don't exist any more. This isn't a great thing for purchasing, but it is a very good thing for performance. If you'd like to go the "AMD has better backward compatibility" route I'd like to ask you how you expect new features, when a chip is locked into a six year old socket. If you don't subscribe to this theory then you can purchase an AMD setup.


Now, connectivity. Aquinus makes the same point time and again, which I happen to disagree with. Rather than rehash this, I'll just say X79 was intentionally crippled by Intel, due to heat. This is the same reason all SB-e 6 cores are 8 cores with two cores lasered off. That's a disappointment to me, but seems to be justified as "an expense inherent in the enthusiast level market." I cannot abide this.


No Thunderbolt, no problem. Intel requires an IGP to get Thunderbolt to work. It produces insane amounts of heat, and still costs a small fortune to get a device. Firewire is a very apt example of this crap, and Intel seems to be putting it into the same failed experiment shoe box. It's cool, it'll still see some incorporation, but it isn't likely to be adopted by the mainstream. I'm not seeing a huge problem here.


DDR4 is interesting. It would offer a very good reason to restructure the socket. I can't say I'm looking forward to its adoption, but it'll be nice to get DDR3 out of here. Less power consumption, better connection to the memory controller, and higher bandwidth are always nice. Wikipedia lays it out in pretty plain terms for those people who don't know about it already: http://en.wikipedia.org/wiki/DDR4



Finally, and the biggest caveat, late 2014. Let's assume that the date is, for the sake of calculation ease, November 16 2014. SB-e saw release November 16 2011. You're going to tell me that three years on a platform is insufficient? Socket 1155 ran from January 2011 to June 2013. 29 months as opposed to 36 from LGA 2011. I don't see half as much vitriol for LGA 1150 as LGA 2011. Why? LGA 2011 has its problems, I do feel disappointed, but I don't feel cheated. LGA 2011 is not a home run, but it is a solid base hit for people who want something more than a gaming rig.



Edit:
Changed spelling.
Posted on Reply
#22
Jstn7477
by: Melvis
They realy should of just kept skt 1155 for Haswell and Broadwell as well, that would of been a realy good socket to keep then? Not like there is a huge difference between them all is there?
Intel integrated voltage regulators into Haswell's die, do instead of the motherboard feeding 6 voltages to the CPU, it only sends 2 now. LGA 1150 boards essentially step down 12v to ~1.8v (user configurable) and the processor then takes that and makes your vcore, system agent voltage, etc. Essentially, this further reduces motherboard complexity.
Posted on Reply
#23
radrok
I've been crying for a consumer 8 core since the first time I saw that the 3930K/3960X were 8 core die with 2 cores fused off.

All I can say is about damn time Intel.


by: HammerON
My next upgrade:)
Agreed!

:toast:
Posted on Reply
#24
Jstn7477
I'd love to have at least one of these to replace a pair of LGA 1155 systems, not to mention that it will probably use fluxless solder for the IHS TIM due to the power dissipation of the chip. 1st gen LGA 2011 really didn't appeal to me at all considering the current chips are now two generations behind the mainstream ones.
Posted on Reply
#25
Naito
by: lilhasselhoffer
You're going to tell me that three years on a platform is insufficient? Socket 1155 ran from January 2011 to June 2013. 29 months as opposed to 36 from LGA 2011.
^This. A majority of people probably upgrade every two to three years anyway, which is the lifespan of the current sockets.
Posted on Reply
Add your own comment