Stories
Slash Boxes
Comments

SoylentNews is people

posted by mrpg on Thursday July 26 2018, @01:33AM   Printer-friendly
from the i-see dept.

Leaked benchmarks show Intel is dropping hyperthreading from i7 chips

Benchmarks found in the SiSoft Sandra database list a Core i7-9700K processor. This increases the core count from the current six cores in the 8th generation Coffee Lake parts to eight cores, but, even though it's an i7 chip, it doesn't appear to have hyperthreading available. Its base clock speed is 3.6GHz, peak turbo is 4.9GHz, and it has 12MB cache. The price is expected to be around the same $350 level as the current top-end i7s.

For the chip that will sit above the i7-9700K in the product lineup, Intel is extending the use of its i9 branding, initially reserved for the X-series High-End Desktop Platform. The i9-9900K will be an eight-core, 16-thread processor. This bumps the cache up to 16MB and the peak turbo up to 5GHz—and the price up to an expected $450.

Below the i7s will be i5s with six cores and six threads and below them, i3s with four cores and four threads.

Meanwhile, AMD's 7nm Ryzen 2 is rumored to boost instructions per clock (IPC) by 10-15% and increase the number of cores per core complex (CCX) from 4 to 8, potentially resulting in mainstream 16 core, 32 thread CPUs.


Original Submission

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 4, Interesting) by FatPhil on Thursday July 26 2018, @07:10AM (4 children)

    by FatPhil (863) <pc-soylentNO@SPAMasdf.fi> on Thursday July 26 2018, @07:10AM (#712947) Homepage
    I don't know - does it? https://www.mersenne.org/report_benchmarks/ I wouldn't know which of those chips are HT, and which aren't, as I stopped following the details of intel/amd tech over a decade ago, but the data's almost certainly there.

    However, compare SpecFP for the two most similar systems I could find with a brief scan down the list such that the chip generation was the same, #cores was the same, Hz was as similar as possible, but HT was on vs. off, according to https://spec.org/cpu2017/results/rfp2017.html :
    PowerEdge R740 (Intel Xeon Platinum 8164, 2.00 GHz) 52 threads 52 cores 1 thread/core 215 SpecFP
    PowerEdge R740xd (Intel Xeon Platinum 8170, 2.10 GHz) 104 threads 52 cores 2 thread/core 219 SpecFP

    Congratulations, HT, with #threads doubled, and 5% more clock cycles, you managed 4% more processing. I think that means you made things worse.

    If HT was really such a great thing, why on the massively multi-core chips do they still only use 2-way HT?
    If 26 cores of 2-way HT is supposedly better than 52 cores of 1-way (i.e. no) HT, then why isn't 17 cores of 3-way HT better than 26 cores of 2-way HT? One would need to see a breakdown of the number of each different fundamental block on each core to be able to take a stab at answering that, but if it's because in order to have another hyperthread you need the vast proportion of what you'd need to have another actual core, plus a bit more overhead, then the question remains - why do you even have 2-way HT in the first place? Perhaps this new i7 answers that question...

    CPU architecture tech is like internal combustion engine tech. There really isn't much variation at all, all we're now doing is tweaking small things, and then giving them more fuel. It needs a revolution. Alas the Mill Architecture seems to be a bit moribund, that certainly looked like it had potential. In these days of altera and xilinx, and an environment where there are plenty of free and open implementations of various cores, there's no excuse for them to not have got out prototypes, apart from perhaps IP reasons slowing them down. And if they never sell a unit because they never get even one chip out of a fab, how useful were such IP protections really? Come on, Ivan, get a bloody proto out please!!!
    --
    Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
    Starting Score:    1  point
    Moderation   +2  
       Interesting=2, Total=2
    Extra 'Interesting' Modifier   0  
    Karma-Bonus Modifier   +1  

    Total Score:   4  
  • (Score: 4, Insightful) by Rich on Thursday July 26 2018, @09:24AM

    by Rich (945) on Thursday July 26 2018, @09:24AM (#712985) Journal

    ... why isn't 17 cores of 3-way HT better than 26 cores of 2-way HT? One would need to see a breakdown of the number of each different fundamental block on each core

    I guess HT was introduced once they saw how much of their valuable silicon sat idle on average on the P4, which had a very long pipeline and wasn't too good at predicting avoidable stalls. Take a CPU with a shorter pipeline and as the predictors get better, the amount of unused logic drops. There must be a point where the gains from using the stall gaps with HT become smaller than the losses from the extra overhead for the HT logic with its second register set.

    Neglecting bottleneck effects, the ballpark rule for an upper bound of gains would be: halving the fundamental blocks in stalled state means half the processing power for HT. The stalled state percentage could be estimated by theoretical upper bound of IPC (clock * execution units) and actual benchmark IPC throughput.

  • (Score: 3, Interesting) by fyngyrz on Thursday July 26 2018, @12:27PM (2 children)

    by fyngyrz (6567) on Thursday July 26 2018, @12:27PM (#713044) Journal

    Congratulations, HT, with #threads doubled, and 5% more clock cycles, you managed 4% more processing. I think that means you made things worse.

    It's not that simple.

    It depends - a lot - on the instruction mix, how much memory outside the caches is being hit, and for many CPUs, whether there are FPU operations going on.

    Knowing what the various issues are, one can tailor c/c++ code to take considerable advantage of hyperthreads in image processing, for instance.

    For "just code" where the author(s) of the code are not paying attention to these issues, threading can certainly bog down. But that's not a hardware problem, per se. It's a programmer problem. The world has a lot of those problems. :)

    • (Score: 2, Funny) by Anonymous Coward on Thursday July 26 2018, @04:05PM

      by Anonymous Coward on Thursday July 26 2018, @04:05PM (#713180)

      Yes, if not for those pesky people, then great things like IA-64 and Soviet communism would have been a success!

    • (Score: 2) by sjames on Thursday July 26 2018, @06:37PM

      by sjames (2882) on Thursday July 26 2018, @06:37PM (#713296) Journal

      For a simple rule of thumb, if you're doing heavy floating point, turn HT off. If it's a network server or desktop, turn HT on.