Intel Core i7-11700K Review: Blasting Off with Rocket Lake

541 Comments

View All Comments

• blppt - Saturday, March 13, 2021 - link

  Not really---the Bulldozer design had far more problems than just the inadequate Windows scheduler---while "optimized" linux kernels gave better performance, the true issue with Bulldozer was that you had 8 relatively weak cores versus 4 strong cores from Intel, and at the time, coders were still struggling to optimize for anything over 2 cores.
  
  In a sense, the designers of Bulldozer REALLY misread the timeline of highly multithreaded coding taking over the market. Heck, even if it was released now alongside the equivalent Core 2 quads, it would still stink for the majority of users, because no game gets any significant advantage of 4+ physical cores even today---and most games still value high single thread/core performance.

• Oxford Guy - Sunday, March 28, 2021 - link

  'the true issue with Bulldozer was that you had 8 relatively weak cores versus 4 strong cores from Intel, and at the time, coders were still struggling to optimize for anything over 2 cores. In a sense, the designers of Bulldozer REALLY misread the timeline of highly multithreaded coding taking over the market.'
  
  My guess is that AMD designed Bulldozer for the enterprise market and didn't want to invest in an additional design more suited to the consumer desktop space. Instead, its additional design priority was the console scam (Jaguar). While that was a good move for AMD it wasn't beneficial for consumers, as consoles are a parasitic redundancy.
  
  One thing many ignore is that Piledriver supercomputers occupied quite high spots in the world performance lists. Mostly that was due to the majority of their work being done by the GPUs, though. Even the original Bulldozer, in Opteron branding, was used in some.
  
  The cheapness of Piledriver chips was also probably a factor in the adoption of the design for supercomputers. Turn down the voltage/wattage so that you're in the efficient part of the improved 32nm SOI node and rely almost completely on heavily threaded code when not running GPU-specific code... and voila — you have an alternative to the monopoly-priced Intel stuff.
  
  But, on the desktop, Piledriver was a bad joke. That's because of its very poor single-thread performance mainly. Not everything can be multi-threaded and even if it is that can mean a speed regression sometimes. The slowness of the L3, the lack of enough operations caching... the design wasn't even all that optimized for multi-thread performance — especially FPU stuff. The cores were very deeply pipelined, designed to use very high clocks. They were not efficient with avoiding bubbles and such. I read that AMD relied too heavily on automated tools due to cost sensitivity.
  
  My vague understanding of the design is that it was narrow and deep like the Pentium 4. Why AMD tried NetBurst 2.0 is beyond me. Even for the enterprise market it's a bad move because power efficiency is important there, especially with servers (rather than supercomputers which, I think, were more tolerant of high power usage – in terms of acceptable design requirements). Even turning down the clocks/voltage to get the best efficiency from the node doesn't fix the issue of the pipelining inefficiency (although hand-tuned code used for some enterprise/scientific stuff would mask that weakness more than general-purpose consumer-grade apps would).

• usiname - Monday, March 8, 2021 - link

  Intel know very well that is mistake, but alder lake's big cores are not much better in term of power consumption and 10nm quality wafers so their only choice is to keep the production of max 8 core mainstream. This is bad for them, because even 3 years after AMD show 16 core mainstream they can't and as our very well known Intel they will cheat by introduce their 16 core cpu with 8 fake cores. Even more, when they introduce their "super duper" 12900k with 16 cores they will set price higher than every amd main stream and this is triple win, cheap 8 core cpu for manufacturing, better binned with higher clocks and on price of $800-1000. If you think intel trying to do something new and innovative you are wrong.

• GeoffreyA - Monday, March 8, 2021 - link

  You may well turn out to be right about 16 cores having 8 junk ones, and knowing Intel, that's how they operate, with smoke and mirrors when they can't compete properly.

• Hifihedgehog - Tuesday, March 9, 2021 - link

  The irony here is how Intel used to give ARM smack for having inferior single core performance while they were surpassed in multicore by the likes of Qualcomm. I believe—paraphrasing—what they would say is not all cores are created equal. Well, it looks like Intel is trying to look like they are maintaining parity when they are really just giving us mostly crappy cores that can’t perform well at all.

• GeoffreyA - Tuesday, March 9, 2021 - link

  Also, as others have pointed out before, the nomenclature is just there to obfuscate the whole picture. Not knowing anything much about Alder Lake, I did some searching and saw that it's Golden Cove + Gracemont. Wondering what exactly GC was, I searched a little but couldn't find the answer, so I'll guess it's just Sunny Cove with a new name stuck on.

• Bagheera - Monday, March 8, 2021 - link

  the more I read about Intel's 10nm (check relevant articles on SemiWiki and Semi accurate), the more I feel ADL is designed as a workaround for the power/thermal limitations of their 10nm process, instead of being some sort of revelation for next gen performance.
  
  Note how TGL remains Intel's only viable product on 10nm at the moment, with Ice Lake SP now two years late. I think Intel knows their 10nm may never be ready for desktop parts, so ADL is a way to have a desktop product on 10nm except not really (it's more akin to a mobile part).
  
  It will probably do fine for gaming, but highly doubtful it will be a meaningful competition to Zen 4 for the prosumer space.

• blppt - Monday, March 8, 2021 - link

  Bulldozer/Piledriver were a dumb design because they relied HEAVILY on highly-threaded applications to achieve their performance. Almost none of which existed in the late 2000s when they launched. Single thread/core performance was absolutely pathetic compared to Intel's offerings at the time (Sandy Bridge and on).

• Zan Lynx - Sunday, March 14, 2021 - link

  And yet here we are today with 8 cores in game consoles and AMD's Mantle API being the basis of both DirectX 12 and Vulkan. Multicore was always the future and it was obvious even in 2003.

• blppt - Monday, March 15, 2021 - link

  "And yet here we are today with 8 cores in game consoles and AMD's Mantle API being the basis of both DirectX 12 and Vulkan. Multicore was always the future and it was obvious even in 200"
  
  You're missing the point. Even TODAY, few (if any) games gain an advantage of more than 4 physical cores + 4 Virtual. It is still, to this day, far more advantageous for games to have 4 strong physical cores than 8 weak ones. The latter was Bulldozer.
  
  Never mind back in 2011/2012.