Intel's Haswell: 20x Lower Platform Idle Power than Sandy Bridge, 2x GPU Performance of Ivy Bridge

23 Comments

View All Comments

• lilmoe - Tuesday, September 11, 2012 - link

  Why should they? AMD failed to push Intel forward in the performance race... They won't even be trying anymore. Blame them, not Intel. Why should Intel focus more on something they've already one by a large margin??
  
  Cortex A15 and 64bit ARMv8 are around the corner, it's a huge threat for Intel in the mobile space. Heck, they're literally "begging" carriers to launch Medfield Android smartphones, and trying to push x86 everywhere. Intel was taken by surprise in the mobile space just like Microsoft was. Haswell makes sense bigtime, and NOT focusing on performance, but on power efficiency makes even MORE sense.
  
  I just don't see how anyone is expecting Intel NOT to focus on those sides in their next platform..... But to ease your anger, Intel has designed Haswell to be modular and salable as hell. It can go from ultra low power mobile to ultra high performance. If AMD (somehow) manages to raise the bar in performance, Intel will be ready to beat whatever AMD has to offer by another wide(r) margin. Nothing is motivating Intel in the desktop space ATM.

• tygrus - Tuesday, September 11, 2012 - link

  They have doubled the AVX FP, increased ALU, branch, load/store. TLB, cache bandwidth etc. but 8ops/c vs 6ops/cycle.
  Clock-for-clock it looks like Haswell could do a minimum of 30% more per cycle, 20% to 80% higher benchmark scores, 30% better for avg. user (assuming 10% average clock speed reduction). Judging by the power budget they might have to reduce normal clock speed. They will increase the turbo boost to be similar to current max boost (3.8 to 4GHz) but nominal clock will be 3.1GHz instead of 3.4GHz. They will have to re-layout the execution units to fit additional resources and to spread the heat from critical areas. Additional metal layers will b added for routing signals and power. More local power gated regions to reduce local and global power usage and allow other regions to run hotter than before. Big difference to power consumption when running 2 simple threads doing 2 IPC each compared to 2 higher IPC threads doing 4 IPC each. At least 1 temperature sensors for each core, GPU block, MC etc. and fast throttling to stop area of <5mm^2 silicon from going beyond design limit even thou the CPU case temp and TDP below max.

• Pixelpusher6 - Wednesday, September 12, 2012 - link

  I'm not surprised that Intel isn't really focusing on performance with Haswell considering their push to mobile and no threatening competition from AMD. It seems like Intel is just playing it safe and increasing performance just enough to make it considered an upgrade over Ivy Bridge. Looking at the slides all it seems like they are doing with Haswell is exactly what they did from Nehalem ----> Sandy Bridge, doubling up and adding some new features / instructions. The performance difference between Nehalem and Sandy Bridge wasn't that huge and some of it had to do with process shrink, so I'm expecting the same again. It's hard to say right now without benchmarks but it doesn't seem like a very compelling upgrade from my i5 2500k.
  
  I'm not sure I agree with this new philosophy that chip makers seem to be embracing which seems to be 1 chip to cover all the market segments i.e. mobile, tablet, notebook, desktop, server, etc. I do understand that there are huge cost savings to be had by adopting this model, but I think it is a tradeoff. The processing needs of enterprise are much different than some teenager checking facebook. I think these different market segments developed for a reason and to lump all their needs together as "one size fits all" might be a little short sighted. If AMD management were smart they would see Intel's focus on mobile as an opportunity, unfortunately AMD has the mobile bug too and it seems like they want to undercut Intel on sub $300 laptops. I'm not entirely sold that mobile is the future and like someone else mentioned Intel is already late to the game. They don't really have a product competitive with ARM's offering so they've sort of slotted in to a niche of powerful tablets / ultrabooks in the mobile space for now.
  
  Very soon ARM and Intel will cross paths as ARM strives to increase performance and Intel strives for high power efficiency. With Intel so focused on competing with ARM in the mobile space, it seems like they are not as concerned with advancing the high end. This is where AMD might be able to catch them off guard. Like I mentioned before I'm not completely sold on the idea that mobile is the future of computing. It's entirely possible that mobile hardware is just a stop-gap until everything moves server side. There are still some kinks to be ironed out before tablets, smartphones, etc. can be replaced by thin clients but it can't be ruled out as a possibility. If this is the direction the market goes then enterprise will take center stage. If I were AMD I would be throwing R & D into designing a great server processor first and foremost, and then fill out the rest of their product line based on this. I'm not sure the Bulldozer / Piledriver architecture is gonna get them there, if they can't fix the problems with it they should just cut their losses and start over or even have divergent products for different markets.