Page 21 of 34 FirstFirst ... 11192021222331 ... LastLast
Results 201 to 210 of 336

Thread: Ryzen mobile- AKA Raven Ridge

  1. #201
    Senior Member
    Join Date
    Aug 2017
    Location
    GA/GB
    Posts
    153
    Cinebench is a floating-point benchmark. Bulldozer and its relatives are renowned for lacking in that department.
    CPU Grade [WiP]

  2. #202
    Senior Member
    Join Date
    Jun 2010
    Posts
    1,133
    Quote Originally Posted by NTMBK View Post
    The gap widened as Intel shrunk their CPUs further and AMD didn't, but even when they were both on 32nm Bulldozer was a failure.
    I think it wasn't just Bulldozer. If they had the Steamroller core from the beginning on a better 32nm process they could have beaten Sandy Bridge. The process was indeed a major problem for Bulldozer. It was hardly an improvement over the 45nm process.



    I had the opportunity to test it myself back then. The efficiency per core advantage of my Llano based Athlon II X4 651 was very small compared to my Regor based Athlon II X2 240. Intel's 32nm process improved a lot more over their 45nm process.

  3. #203
    Are any of the following genuinely debateable?

    1. Coffee lake isn't coming to 15w during 8th gen.
    2. 15w intel = kabylake refresh = gt2 graphics; same as skylake and broadwell w/ 24 execution units.
    3. Br limiting factor in graphics performance is CPU and ram speeds.
    4. Rr = process shrink = increased perf/watt.
    5. Rr = zen = increased ipc vs excavator.
    6. Rr = zen = shorter pipeline = lower frequency requirements = lower power requirements = increased perf/watt.
    7. Lower power requirements = more headroom for gpu
    8. Rr = higher DDR4 frequency compatibility vs br.
    9. Rr = 4c/8t > 2cu/4c/4t of br
    10. Rr = single CCX = no infinity fabric latencies
    11. Process shrink + increased cpu ipc + increased perf/watt from lower frequency requirements + no CCX + 4 cores vs 2 compute units + higher DDR4 frequency + more power headroom vs cpu = graphics improvements without yet touching graphics compute cores.

    What about on the igpu side?

    12. Process shrink
    13. There has been a leak of 11 compute units
    14. What else? RPM? Vega 56/64 crazy overvolt creating vega is a power hog mentality? Is 56/64 running above its ideal frequency?
    Last edited by VainT; 1st September 2017 at 09:12 AM. Reason: mistake on 2 with skylake instead of kaby

  4. #204
    Senior Member
    Join Date
    May 2012
    Posts
    4,744
    Quote Originally Posted by VainT View Post
    Are any of the following genuinely debateable?

    1. Coffee lake isn't coming to 15w during 8th gen.
    I'm pretty sure it is ? Four 15W 4C/8T i5s and i7s I believe have been announced so far. Which will probably form the vast bulk of their 'mainstream' mobile effort. And they're not actually that bad as far as iGPU goes, the UHD 620 graphics they come with is ~6% faster in 3Dmark 11 than even the fastest 15W Pro A12-9800B.

    But then that in itself is a bit of a problem. Because the Pro A12-9800B is a 28nm Bristol Ridge with 512SPs, DDDR4-1866 and Excavator cores.

    And Raven Ridge (for which mobile Coffeelake was supposed to be the answer) is presumably going to be much faster ? That's just in DX11 btw. In DX12/Vulkan, it will probably be even uglier for Intel. Because with a 4C/8T Zen, there won't be any CPU bottleneck issues either...

    Call me biased, because when it comes to APUs I freely admit that I am, but I personally think that Ryzen Mobile is going to be both the most interesting, and most successful, of all the Zen based products. Because both Zen and Vega do seem to be rather good at lower voltage/frequency ? Which in mobile, isn't going to hurt them one bit.
    To find the right answers you must ask the right questions.

  5. #205
    Quote Originally Posted by Stuckey View Post
    I'm pretty sure it is ? Four 15W 4C/8T i5s and i7s I believe have been announced so far.
    That is not coffee lake.
    There is an article about it on the front page of semiaccurate.

    It also is GT2 and not GT3e graphics. (620 is GT2 and this line isn't directed at your post per-say)
    Last edited by VainT; 1st September 2017 at 08:46 AM.

  6. #206
    Senior Member
    Join Date
    May 2012
    Posts
    4,744
    Quote Originally Posted by VainT View Post
    That is not coffee lake.
    It also is GT2 and not GT3e graphics.

    There is an article about it on the front page of semiaccurate.
    Sorry, yes, the mobile version is Kabylake-R. But they still call it 8th generation, which is confusing. But the point still stands about its chances vs Mobile Ryzen. Because they probably aren't going to have GT3/4e, not in 15W anyway. Because they already doubled the CPU cores/threads, and I doubt 15W leaves them with much budget to add eDRAM.
    To find the right answers you must ask the right questions.

  7. #207
    Senior Member
    Join Date
    Aug 2017
    Location
    GA/GB
    Posts
    153
    Quote Originally Posted by VainT View Post
    Are any of the following genuinely debateable?

    1. Coffee lake isn't coming to 15w during 8th gen.
    2. 15w intel = kabylake refresh = gt2 graphics; same as skylake and broadwell w/ 24 execution units.
    3. Br limiting factor in graphics performance is CPU and ram speeds.
    4. Rr = process shrink = increased perf/watt.
    5. Rr = zen = increased ipc vs excavator.
    6. Rr = zen = shorter pipeline = lower frequency requirements = lower power requirements = increased perf/watt.
    7. Lower power requirements = more headroom for gpu
    8. Rr = higher DDR4 frequency compatibility vs br.
    9. Rr = 4c/8t > 2cu/4c/4t of br
    10. Rr = single CCX = no infinity fabric latencies
    11. Process shrink + increased cpu ipc + increased perf/watt from lower frequency requirements + no CCX + 4 cores vs 2 compute units + higher DDR4 frequency + more power headroom vs cpu = graphics improvements without yet touching graphics compute cores.

    What about on the igpu side?

    12. Process shrink
    13. There has been a leak of 11 compute units
    14. What else? RPM? Vega 56/64 crazy overvolt creating vega is a power hog mentality? Is 56/64 running above its ideal frequency?
    1) Coffee Lake is desktop-only. Cannon Lake is the mobile equivalent and we don't know much about that right now since Kaby Lake-R has apparently joined in on the eighth-generation goodness.

    2) Yup, Kaby Lake-R is identical to Kaby Lake-H where the iGPU is not Iris Pro. There are two minor differences in features (both on the iGPU), but the silicon of the CPU and iGPU are exactly the same.

    3) Going by prior APUs, this seems like a good guess and I wouldn't be surprised. Excavator is a stepping stone over Steamroller, but that won't excuse it from holding back such a powerful iGPU, which is what Steamroller did.

    4/5/6) Pretty much a given.

    7) This is true. And you can bet the iGPU will be more power efficient too.

    8) I am expecting DDR4-2667, in-line with Summit Ridge. However, if Vega struggles with memory that low, it might be entirely possible that Raven Ridge will support higher frequencies as standard, which will also improve Infinity Fabric performance.

    9) A quad-core, quad-thread Ryzen 3 chip can comfortably trade blows with an FX-8320, let alone the APUs. The biggest jump by far, will be floating-point performance (which we already know).

    10) There might still be latencies between CPU and iGPU, though I don't have much insight into this.

    11) Yup, sounds believable. Even with the same 512 USP configuration as Bristol Ridge, the die shrink will allow higher frequencies. Faster RAM support out of the box will also improve memory bandwidth, which, ever since Kaveri, has been holding back this configuration. Vega's cores will also bring some improvements. Now, we add 50% more of those cores on-top, and I think AMD will have the iGPU performance crown.

    12) Pretty much a given.

    13) I expect this leak to be while AMD gathers yields of complete 12-CU configurations. Unless there's a specific reason for 11 compute units, I'd expect there will be 12 in the halo product.

    14) If you can undervolt Vega without touching the clocks, you'll get a pretty power efficient chip. Now, we already have 1,108 MHz in a desktop Bristol Ridge iGPU, with 512 USPs. That drops to 900 MHz for the laptop version. With the increase in USPs, there's really no need for the frequency to increase. I actually expect it to drop a little. But, as with most of the APUs, the iGPU overclocking headroom will exist. I get 1,000 MHz out of my 844 MHz HD 8570D.
    Last edited by DylRicho; 1st September 2017 at 12:49 PM.
    CPU Grade [WiP]

  8. #208
    Senior Member
    Join Date
    Sep 2011
    Posts
    924
    Vega IGP could be very good for overclocking, but the question is at what clock It will hit a BW limit.

  9. #209
    Senior Member
    Join Date
    Aug 2009
    Posts
    1,383
    Quote Originally Posted by NTMBK View Post
    It was a ground up redesign of the CPU which barely improved on the 32nm Llano CPU it replaced:



    And of course it was woefully uncompetitive against the 32nm Sandy Bridge CPUs it was competing against.

    The gap widened as Intel shrunk their CPUs further and AMD didn't, but even when they were both on 32nm Bulldozer was a failure.
    Quote Originally Posted by Moral Hazard View Post
    The "Bulldozer catastrophe" was the process difference. For its core size and process Bulldozer performed fine.
    The Bulldozer cores, or at least the ones I could readily find numbers for, had ~150 million transistors (assuming you split the common resources of a "compute unit" evenly between the two cores). The Intel "Core" chips have had ~300 million, with the first ones being a little below that number and the more recent ones having a little above. These rough estimates divide the "uncore" transistors between the cores.

    Assuming Polack's rule holds, you would expect Intel to get about 40% more performance per-core, holding process constant. Of course to get this they have to make the chips twice as big.

    You can argue that going with the smaller chips was a mistake on AMD's part, but even so that would be a strategic error in expecting code parallelization to advance faster than it ultimately did, and not a flaw in the design of Bulldozer.

    To see the effect of the core size in action, look at Zen. Zen cores were designed to be the size of a Bulldozer "compute unit" (iow twice as big). You would expect that to earn them a ~40% IPC increase, which is what they got. That also makes them roughly the same size as Intel's chips and, not surprisingly, they perform about the same.
    Fear leads to uncertainty.
    Uncertainty leads to doubt.
    Doubt leads to Intel.

    I sense much fear in Intel.

  10. #210
    Senior Member
    Join Date
    Dec 2012
    Posts
    5,011
    Quote Originally Posted by Stuckey View Post
    I'm pretty sure it is ? Four 15W 4C/8T i5s and i7s I believe have been announced so far. Which will probably form the vast bulk of their 'mainstream' mobile effort. And they're not actually that bad as far as iGPU goes, the UHD 620 graphics they come with is ~6% faster in 3Dmark 11 than even the fastest 15W Pro A12-9800B.

    But then that in itself is a bit of a problem. Because the Pro A12-9800B is a 28nm Bristol Ridge with 512SPs, DDDR4-1866 and Excavator cores.

    And Raven Ridge (for which mobile Coffeelake was supposed to be the answer) is presumably going to be much faster ? That's just in DX11 btw. In DX12/Vulkan, it will probably be even uglier for Intel. Because with a 4C/8T Zen, there won't be any CPU bottleneck issues either...

    Call me biased, because when it comes to APUs I freely admit that I am, but I personally think that Ryzen Mobile is going to be both the most interesting, and most successful, of all the Zen based products. Because both Zen and Vega do seem to be rather good at lower voltage/frequency ? Which in mobile, isn't going to hurt them one bit.
    GTA5 720p 24-30+fps

    15W UHD 620:
    https://www.youtube.com/watch?v=kTkJyC04hx0

    15W AMD A10-9600P
    https://www.youtube.com/watch?v=XcsGI3dOTo4

    Even though I may share your opinion, I refuse to give AMD the benefit of doubt on unreleased products anymore.

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