Page 615 of 876 FirstFirst ... 115515565605613614615616617625665715 ... LastLast
Results 6,141 to 6,150 of 8752

Thread: Official AMD Zen uarchitecture thread

  1. #6141
    Senior Member
    Join Date
    Dec 2010
    Location
    Switzerland Geneva Cern
    Posts
    2,904
    Quote Originally Posted by Fottemberg View Post
    Reading some forums, I think a lot of Ryzen review will be focused on Intel IPC single thread performance. IMHO will be used even SuperPi ... XD
    There are more modern implementations such as ycruncher that make much better use of the resources of a modern cpu. It is a really bad idea to use superPi to extrapolate a quad or octa core cpu performance.
    Current rig: i5 3570 @3.8GHz, Asus 7750, Dell 24'' 2560X1440, Corsair Vengence 16GB Ram DDR3 @1600MHz, Chieftec 550W PSU, 480GB OCZ SSD + 500GB Seagate + 1TB WD hd, Windows 10 64-bit

  2. #6142
    Banned
    Join Date
    Aug 2013
    Posts
    4,741
    Quote Originally Posted by JeeBee View Post
    Luckily for AMD not much code uses AVX256 because Intel shot themselves in the foot by disabling it on the vast majority of its SKUs.

    AMD have been clear over the past ten years on their path to floating point performance - it's going to be done on the GPU. They've spent a long time building up software libraries to make this efficient.
    I agree that situation of AVX256 on consumer is irrelevant, but I always believed that AVX256 real goal was server and HPC. Same thoughts for AVX512. It is not coming to Skylake consumer, but it is coming to servers and HPC Xeons.

    And I also agree that AMD will use GPUs for HPC. But that was not my point. My point was that even having 256bit SIMD units and datapaths, Intel desings aren't much bigger than Zen. That supposed large advantage that AMD was going to have by limiting the core to 128bit is definitely gone.

    Quote Originally Posted by JeeBee View Post
    10% more IPC?
    The quote was "at least 10% higher IPC".

    Assuming Zen has 40% higher IPC than Excavator

    74.60 * 1.4 = 104.4

    Skylake does 131 at same clocks. The IPC gap is

    131 / 104 = 1.26

    Assuming Zen has 55% higher IPC than Excavator, the IPC gap is

    131 / 116 = 1.13


    Quote Originally Posted by JeeBee View Post
    You can mount MCM chips right next to each other on the package. You may have some duplicated logic across the dies, but the saving on doing multiple designs, tape-outs, verification, validation ...

    Btw, take a look at the 246mm^2 10C Broadwell-E die and tell me what you see: http://hothardware.com/ContentImages...e-die-shot.jpg
    What I see, apart from the massive wasted silicon, is a chip with the same cores/mm^2 as Zen (but it's not a SoC), and it too has a massive uncore at the top. It's just a different way of laying stuff out.
    MCM provides less integration than a single die. Also I don't see your point regarding dies. One die has 2 more cores, and quad-channel controllers, whereas the other die has some SoC stuff instead that.

  3. #6143
    Banned
    Join Date
    Aug 2013
    Posts
    4,741
    Quote Originally Posted by os2wiz View Post
    Your impeccable source for 3.2 GHZ being the highest base clock on 4 core 8 thread Ryzen final product???? Don't tell me Coolaler or WCCFTech.
    I don't use WCCFTEch as source, and everyone knows what I think about them.

    Also say me how many times I have to explain you that Coolaler is not a source, but they are getting the information from another part?

    Quote Originally Posted by os2wiz View Post
    You mention a sample on Canard PC for 4 core Ryzen at 3.2 GHZ you have nothing to prove that is the final engineering sample
    How many times I have to remark what I am not talking about engineering samples? I believe I have already stated about a dozen of times that I am not talking about the clocks of engineering samples. How many times I have to repeat it? 20? 32? 81? Just curious...

  4. #6144
    Senior Member
    Join Date
    May 2012
    Posts
    4,671
    Quote Originally Posted by Fottemberg View Post
    Reading some forums, I think a lot of Ryzen review will be focused on Intel IPC single thread performance. IMHO will be used even SuperPi ... XD
    Won't make any difference. If Ryzen can go faster in games for less money, nobody will care. Especially if all they have to do is set one slider and then forget it.

    Note- The strapline for the Ryzen reveal event wasn't 'If you're a geek into corner case benchmarks like SuperPi you won't want to miss this !'

    Was it ?
    To find the right answers you must ask the right questions.

  5. #6145
    Banned
    Join Date
    Aug 2013
    Posts
    4,741
    Quote Originally Posted by Nerdmaster View Post
    A year ago you had predicted ~3GHz base clock for zen which was attributed to the process not being suitable for high power cpus. However amd has demonstrated 3.6GHz base sample which is 20% off your estimation.

    Predictions can be either right or wrong. Its pointless to measure who is doing the best predictions and how far off each one is.
    I predicted 3.0GHz base and 3.5GHz turbo for a 95W 8C Zen. And I predicted higher clocks for the low-core models. For instance, I predicted 3.4GHz base for a 4C 65W Zen.

    The current 95W PC/PR chip has 3.4GHz base and 3.8GHz turbo. I was wrong by 400MHz on base and by 300MHz on turbo. The overall error was less than 11%. Not so bad considering that the prediction made by certain Italian guy with sources was off by 6%. And I better don't mention the errors of all those resident experts that predicted base clocks of 4GHz and higher...

    That 3.6/4.0GHz PC/PR chip you mention is rated at 105W. You cannot pretend to compare it with my prediction for 95W. Can you?

    Finally, let me emphasize that my prediction was for average silicon. The 8C PC/PR chips are getting higher clocks than the 6C and 4C PC/PR chips. There are two possible explanations for it: either a problem with the CCX approach or a problem with 14LPP. CPCHardware think it is the latter. They think that AMD is cherry picking the best silicon for the 8C chips, and leaving the silicon that is ustable at higher clocks for the rest of SKUs, and they think this is the reason why the 8C chips have higher clocks.

    If this hypothesis is true then the error in my prediction is even smaller, because the current 3.4/3.8GHz chips would be cherry picked best silicon dies, instead average silicon dies.

  6. #6146
    Senior Member
    Join Date
    Oct 2012
    Posts
    852
    It seems AMD marketing did it again:

    https://videocardz.com/65708/intel-p...-ryzen-7-1800x

    You might have seen this list supposedly showing a full lineup of Ryzen CPU names. As much as I would like to say that the list was fake, there is nothing that would suggest otherwise. In fact, I can already tell you that Ryzen CPUs are already being detected with their ?real? names.

    So far we have seen two Ryzen 7 CPUs which are 1800X and 1700X. The only difference from the leak is the R7 moniker is being shown as Ryzen 7. I?m not sure why ?eight-core processor? is not shown as ?octa-core? though. Of course, we will keep you updated once more CPUs show their full, and hopefully final, names.

    Ryzen CPU Base Frequency Cores Threads
    AMD Ryzen 7 1800X Eight-Core Processor 3 600 MHz 8 16
    AMD Ryzen 7 1700X Eight-Core Processor 3 400 MHz 8 16

  7. #6147
    Senior Member
    Join Date
    Jan 2012
    Posts
    829
    Quote Originally Posted by juanrga View Post
    I predicted 3.0GHz base and 3.5GHz turbo for a 95W 8C Zen. And I predicted higher clocks for the low-core models. For instance, I predicted 3.4GHz base for a 4C 65W Zen.

    The current 95W PC/PR chip has 3.4GHz base and 3.8GHz turbo. I was wrong by 400MHz on base and by 300MHz on turbo. The overall error was less than 11%. Not so bad considering that the prediction made by certain Italian guy with sources was off by 6%. And I better don't mention the errors of all those resident experts that predicted base clocks of 4GHz and higher...

    That 3.6/4.0GHz PC/PR chip you mention is rated at 105W. You cannot pretend to compare it with my prediction for 95W. Can you?

    Finally, let me emphasize that my prediction was for average silicon. The 8C PC/PR chips are getting higher clocks than the 6C and 4C PC/PR chips. There are two possible explanations for it: either a problem with the CCX approach or a problem with 14LPP. CPCHardware think it is the latter. They think that AMD is cherry picking the best silicon for the 8C chips, and leaving the silicon that is ustable at higher clocks for the rest of SKUs, and they think this is the reason why the 8C chips have higher clocks.

    If this hypothesis is true then the error in my prediction is even smaller, because the current 3.4/3.8GHz chips would be cherry picked best silicon dies, instead average silicon dies.
    I already wrote a comment about this but ill repeat.
    What you are describing is the general binning process all companies use, they keep the best silicon for the top products and use binning to segment the chips lower down the scale, with lower skus achieving lower frequencies whilst have cache/cores/threads fused off based on yields and silcon performance.

    This is exactly the same thing that intel does with the 7700k and lower, 7700k is a 4/8 die, similar to summit ridge being a 8/16 die, it looks like they will both bin the full die to the highest frequencies, are you saying it is not fair to say kabylake is not really a typical 4.5ghz die because it is only used on 1 top sku? Of course it is else by that logic pentium g4630 @ 3.5 ghz is the real typical frequency for kabylake and intel must be 'cherry picking' the best dies for 7700k!,
    IF AMD gets 8/16 to 3.6/4.0 @95w and can be produced a volume then we can say the same about summit ridge.
    Just a thought if it does achieve those clocks and gets around broadwell ipc (instead of sandy) what would your margin of error be then? 20-25%?

  8. #6148
    Quote Originally Posted by juanrga View Post
    I don't use WCCFTEch as source, and everyone knows what I think about them.

    Also say me how many times I have to explain you that Coolaler is not a source, but they are getting the information from another part?



    How many times I have to remark what I am not talking about engineering samples? I believe I have already stated about a dozen of times that I am not talking about the clocks of engineering samples. How many times I have to repeat it? 20? 32? 81? Just curious...
    If you are not talking abot quad core engineering samples there is nothing you are basing your 3.3 GHZ claim on. AMD can create a a special Black Editiin quad core 4/8 with 85 or 90 watt TDP and have a base clock of 4.0 GHZ and 4.3 Boost. Like you said the 8 core with 4.0 boost enabled has a 105 watt TDP. So your arguement is vacuous. They can and most probably will have a quad core in one iteration that will compete at least on a Skylake level at an elevated TDP.

  9. #6149
    Quote Originally Posted by juanrga View Post
    I don't use WCCFTEch as source, and everyone knows what I think about them.

    Also say me how many times I have to explain you that Coolaler is not a source, but they are getting the information from another part?



    How many times I have to remark what I am not talking about engineering samples? I believe I have already stated about a dozen of times that I am not talking about the clocks of engineering samples. How many times I have to repeat it? 20? 32? 81? Just curious...
    If you are not talking abot quad core engineering samples there is nothing you are basing your 3.3 GHZ claim on. AMD can create a a special Black Editiin quad core 4/8 with 85 or 90 watt TDP and have a base clock of 4.0 GHZ and 4.3 Boost. Like you said the 8 core with 4.0 boost enabled has a 105 watt TDP. So your arguement is vacuous. They can and most probably will have a quad core in one iteration that will compete at least on a Skylake level at an elevated TDP.

  10. #6150
    Just a rumor. Intel could release a special i7 Skylake X @ 165W.
    Do you remember the Pentium 4 EE?
    Everything I post in these or any other forums, or anywhere else online for that matter, is strictly my personal opinion.

    Bits And Chips Site
    Bits And Chips - ENG Twitter

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
WordPress Appliance - Powered by TurnKey Linux