Intel Confirms Devil's Canyon, Haswell-E, Broadwell Rumors
Intel announces a 2014 Haswell refresh that will usher in a plethora of new enthusiast-class CPU models to be released before 2015’s Broadwell.
What’s a CPU company to do when its new motherboards are launching months ahead of its new processors? Well, when the socket’s the same, just refresh the old processors! Intel quoted a mid-year launch during its Monday press briefing, but mid-year will need to come early if Intel wants this launch to coincide with Z97. Motherboard makers tell us to expect the new Haswell models three months after Z97’s mid-April launch, which sounds to be a few months shy of Intel’s mid-year launch claim. Oops?
Devil's Canyon

An improved Thermal Interface Material (TIM) between the CPU core and heat spreader is set to improve thermal transfer in a new, enthusiast-class Haswell-based processor codenamed “Devil’s Canyon”. Intel didn’t name the processor model, but industry insiders tell us to expect this change will apply at least to the firm’s upcoming Core i7-4790. Without hazarding further guesses on model numbers, we’re trusting Intel’s stated intention of a mid-year launch.
Pentium Anniversary Edition

In yet another high-profile Haswell refresh, Intel unveiled an unlocked Pentium (20th) Anniversary Edition, which is also slated to launch mid-year.
Haswell-E Gets DDR4

Pushing beyond the limits of quad-channel DDR3, Intel smashes all hope for Haswell-based X79 processor upgrades by announcing that Haswell-E will utilize DDR4. While Intel tells us to expect a launch sometime in the second half of 2014, motherboard manufacturers whisper about August. Turning to official documents that we weren’t supposed to see, the associated X99 PCH adds integrated clock generation to the X79 that it’s set to replace. The four disabled SAS ports of X79 become enabled SATA ports in the X99, bringing the total SATA count to ten.
And Broadwell?

Though Intel occasionally discussed Broadwell’s launch in the same breath as Haswell-E, the firm stipulated that it wasn’t ready to announce a launch date. Motherboard partners estimate that the socketed version will launch no earlier than CES 2015.
Story updated with photos.


Haswell is a new architecture (two extra execution ports, 24 extra re-order buffer entries, re-arranged cache, re-arranged branch prediction and a handful of other enhancements compared to SB/IB) and is not even one year old yet. Broadwell is little more than a die-shrink so there is not much reason to expect major improvements there apart from power.
Skylake is the next architecture update but considering how little of a step SB/IB to Haswell was, I would not expect Skylake to yield more than the usual ~10% that has become the norm. As I have been saying for years, most of the easy performance gains have been tapped out; there aren't any major cost-effective IPC breakthrough left so any future major desktop performance gains will have to be achieve through parallelism but such parallel CPUs are pointless until a whole lot more mainstream software becomes finely threaded, which is easier said than done.
The slide does say "unlocked desktop processors" and that does usually mean the highest-end of the range.
IIRC, there was a leaked slide a few months ago that showed the eDRAM "L4 cache" being standard across the Broadwell lineup. I would not be too surprised if Intel's hyped air gestures used Iris Pro GPGPU.
Haswell is a new architecture (two extra execution ports, 24 extra re-order buffer entries, re-arranged cache, re-arranged branch prediction and a handful of other enhancements compared to SB/IB) and is not even one year old yet. Broadwell is little more than a die-shrink so there is not much reason to expect major improvements there apart from power.
Skylake is the next architecture update but considering how little of a step SB/IB to Haswell was, I would not expect Skylake to yield more than the usual ~10% that has become the norm. As I have been saying for years, most of the easy performance gains have been tapped out; there aren't any major cost-effective IPC breakthrough left so any future major desktop performance gains will have to be achieve through parallelism but such parallel CPUs are pointless until a whole lot more mainstream software becomes finely threaded, which is easier said than done.
When Iris Pro got announced, at least one chip supply analyst firm estimated the cost of that 128MB chip at around $80. Since Intel has ramped the prices for a fair chunk of their CPU lines by ~$40 since Sandy Bridge, I would not be too surprised if Intel ended up absorbing half the cost in their inflated margins and tacking the other half on MSRPs... so I'm expecting Iris Pro Haswell/Broadwell desktop chips to cost ~$40 extra compared to their Haswell-v1 counterparts.
BTW, if you buy an 8-cores Haswell-E, that would be on whatever comes after LGA2011 and instead of 2x4GB, you would need 4x4GB to fill all four channels. I doubt many people going for -E chips would be comfortable with only 8GB total RAM to go with their $1000+ CPU... and with eight cores on the die, those two extra memory channels may become a fair bit more important.
@feroox: the performance doubling was more like ~18 months for CPUs. 6-12 months was for the early years of GPUs where everyone was doing radical re-designs with nearly every product generation and 2-3 concurrent design teams on staggered product cycles. Now that most efficient ways of doing stuff have been discovered and that the DX/OGL feature sets have stabilized, further performance scaling now mostly depends on process technology, TDP and surface area just like it does for CPUs and the genuinely new product cycle dropped to 18-24 months.
Sockets are the size they are due to the time it takes electric to get from one side to the other, or something. I'd like to see a PCI-E CPU so we can finaly stop it with the mobo switches. Why can't the chipset be on the equivalent of a GPU's PCB, and slide the whole thing into a universal slot on the mobo?
The reason they do 10% upgrades every year is because there is no one to compete. Even when AMD tried to match Intel they were not a real threat. The GPU market is VERY competitive at the moment, look at rebranded $450 GTX 680's (770's) selling for $330. Its incredible. If we had that in the CPU market, things might actually go DOWN in price every year.
Sockets are the size they are due to the time it takes electric to get from one side to the other, or something. I'd like to see a PCI-E CPU so we can finaly stop it with the mobo switches. Why can't the chipset be on the equivalent of a GPU's PCB, and slide the whole thing into a universal slot on the mobo?
The reason they do 10% upgrades every year is because there is no one to compete. Even when AMD tried to match Intel they were not a real threat. The GPU market is VERY competitive at the moment, look at rebranded $450 GTX 680's (770's) selling for $330. Its incredible. If we had that in the CPU market, things might actually go DOWN in price every year.
I can't speak authoritatively on this, but wouldn't having a lot-in CPU, especially with PCIe, considerably limit its bandwidth and the number of lanes to various other devices? Having everything integrated allows much more bandwidth and specialization with how it all works together, especially when talking about communicating with memory, southbridge devices, and a GPU taking its own set of PCIe lanes.
Slot style intel CPU's used to exist in super old computers [probably 15+ years ago]. I imagine they were killed off for a reason.