Thursday, November 1st 2012
Top Intel Ivy Bridge-E Core Processors To Still Pack Six Cores
Intel's 2011-launched Core i7 "Sandy Bridge-E" HEDT platform is based on a 32 nm silicon that's common with Xeon E5 series processors. While the silicon physically packs eight CPU cores and 20 MB last-level cache (LLC, or L3 cache), client Core i7 processors are configured with only a maximum of six cores, and up to 15 MB L3 cache. According to a MyDrivers.com report, the maximum core count won't change with next-generation 22 nm Ivy Bridge-E Core i7 processors.
Ivy Bridge-E will be an upscale of Ivy Bridge. Similar to Sandy Bridge-E, the silicon will feature up to eight cores and 20 MB L3 cache. In its Core i7 avatar, however, the chip will be configured with no more than six cores, and no more than 15 MB L3 cache. The new chip will introduce IPC improvements, PCI-Express Gen 3.0 certified root complex (one which NVIDIA will approve of), higher CPU core clock speeds, and support for faster memory.
TDP could be the only reason Intel isn't willing to unlock cores 7 and 8 on client processors. Eight core, 20 MB LLC-laden Xeon E5 models based on today's 32 nm silicon, with 130W TDP, barely manage to scrape the 3.00 GHz mark. Given that, the prospects for Ivy Bridge-E client CPUs to run with all cores and LLC enabled, and yet deliver higher clock speeds than predecessors were always going to be low.
Intel Core i7 "Ivy Bridge-E" HEDT processors are compatible with existing socket LGA2011 motherboards (subject to BIOS update), and are slated for Q3-2013.
Source:
MyDrivers
Ivy Bridge-E will be an upscale of Ivy Bridge. Similar to Sandy Bridge-E, the silicon will feature up to eight cores and 20 MB L3 cache. In its Core i7 avatar, however, the chip will be configured with no more than six cores, and no more than 15 MB L3 cache. The new chip will introduce IPC improvements, PCI-Express Gen 3.0 certified root complex (one which NVIDIA will approve of), higher CPU core clock speeds, and support for faster memory.
TDP could be the only reason Intel isn't willing to unlock cores 7 and 8 on client processors. Eight core, 20 MB LLC-laden Xeon E5 models based on today's 32 nm silicon, with 130W TDP, barely manage to scrape the 3.00 GHz mark. Given that, the prospects for Ivy Bridge-E client CPUs to run with all cores and LLC enabled, and yet deliver higher clock speeds than predecessors were always going to be low.
Intel Core i7 "Ivy Bridge-E" HEDT processors are compatible with existing socket LGA2011 motherboards (subject to BIOS update), and are slated for Q3-2013.
111 Comments on Top Intel Ivy Bridge-E Core Processors To Still Pack Six Cores
More cores doesn't = faster ;)
I am so disappointed Intel, really, it seems we are going backwards instead of progressing.
And while they're at it, they should use one of those spare QPI links to put another 8-core die on die for single-socket 16C/32T!
If we remove the thermal paste issue from IB, you're still stuck with what amounts to an incremental improvement of SB. With an incremental improvement you've got enough room for either higher frequencies, or more cores. If you try to do both you break the TDP limits you set for yourself.
As an owner of a 2011 rig, I'm sad but not surprised. "Everyone" is predicting the death of the desktop PC. The payback Intel get for putting their best foot forward (read: enthusiast aimed silicon) is, on paper, getting smaller and smaller. They have little to gain by investing in more substantial platforms.
SB-e was supposed to be a revelation. If you cut the two unused cores off the die, and actually had the connectivity they've just started to deliver on at launch, it might have been. As it stands, SB-e was just a slightly bigger 1155 offering. IB-e is likely to be the same, but so late that IB in the 1155 socket will have been replaced by 1150 offerings. Too little, far too late.
Back to the topic at hand. Intel having two "defective" cores is unlikely. They have 8 core offerings on the server side already. The 130 watt TDP is the wall they work against. Their initial planning was for 8 cores. To cut those 2 cores out, after the initial designs hinged upon them, is an unreasonable expectation. It's easier to have the two cores removed after production, than to completely restructure the design. IB is inheriting this issue, because they chose to ramp up frequency rather than having more cores. Given that very little on the market now prefers cores over frequency, the choice is understandable. I can't support it myself, but it is understandable...
Edit:
Screwed up. Socket 1150 is Haswell, not socket 1156. Fixed.
It is very very unlikely they use a completely different die since it would drive the costs up through the roof. I agree with you on the first part, but think about this: where will haswell fit in here?
Because it could end up being almost as fast as the 6 core SB-e (or even Ivy-e), so what's the point of socket 2011 on consumer side?
And I don't see why TDP is a problem, these chips are for people who want the best and will use the best cooling, they already announced a 150W TDP 6 core SB-E.
The 6 cores version probably are defective chips (2 locked cores), it's cheaper to produce only one die and then lock the cores (the same way AMD did it).
Same thing happened on socket 1366 with Xeons, they had 6 core and 4 core versions built on 32nm and both were from the same die.
Something happened around the Core technology period that shot them forward, and AMD has never since been able to catch up. Its like they found something that helped them drastically, and AMD can't reproduce those results.
Ivy Bridge-E is unlikely to change any of this.
Therefore the LGA2011 platform is basically beyond consideration for many regardless of the number of cores / threads.
Of course, this is assuming Haswell is worth it. A preliminary assumption, that Haswell will be a substantial advance, may be incorrect. Intel will hopefully make a Haswell that can compete against ARM, while scaling for Desktop applications. The problem is that I can't make that assumption. Intel sees mobile computing as the future, and is already showing a willingness to throw traditional platforms under the bus. Hopefully this is paranoia, but the prophesy of a disappearing PC may become self fulfilling if Intel and AMD don't maintain the PC market.
The "danger" here is that they don't make unlocked chips, why? Well I guess it's the famous "Because they can" (if AMD doesn't wake up).
It's clear from SB/IB on 1155, they could ramp up the clock and rate them as 125W TDP chips, but they have no need for that.
I'm just waiting to see what kind of stupid limitations Haswell will have
Other than the workload, everything else comes down to supply and demand. Pretty much. ROI is obviously going to depend on harvesting salvage parts. That has always been the case with processors, although defective cores argument might also include cores that are out of voltage spec in regard to those adjacent but fully functional.
FWIW: 32nm Gulftown/Westmere also had a 2-core variant ( 4 inactive cores), the X5698, that was built primarily for high clockspeed and lightly threaded application. From what I understand, the inactive cores weren't defective- just fused off to allow higher frequency of the two active cores. It also retained all other aspects of the 6-core variants (full cache, max QPI etc.)