Announcement

Collapse
No announcement yet.

AMD Launches The Ryzen Threadripper 7000 Series: Up To 96 Cores, DDR5 RDIMMs, PRO & HEDT CPUs

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • guglovich
    replied
    Originally posted by bridgman View Post

    In general as the core count goes down the clock speed goes up, so total power doesn't vary much. The 32 core part has a 4.0 GHz base clock while 24 core is 4.2 GHz. I'm sure there are minor differences in power between the SKUs but the overall approach is to have a range of parts that can all function within the same power envelope.

    If you look at Zen4 consumer parts you will see a similar pattern - 12 and 16 core parts are both 170W, while 6 and 8 core are both 105W. Remember that we still have 4 CCDs on the 24 core TR parts - we disable 2 cores on each CCD but the rest of the chip including data fabric and the big L3 cache remains fully active.

    You see a similar pattern with EPYC parts - the high clock 32, 24 and 16 core parts all have the same TDP, but they also have roughly the same cache. We offer lower TDP EPYC parts with reduced cache and lower clocks, but TR parts are generally max everything.
    Now that's clarity, that's what I thought, physically the junior and mid-range models have the same number of cores, but they are disabled. So it's the structure of the processor itself under the cover.

    Leave a comment:


  • skierpage
    replied
    Originally posted by pong View Post
    So my point is that one could probably in many relevant use cases refactor the system architecture so that a "CPU" could almost be a peripheral of the GPU if not handled BY the GPU (assuming a future modified SW / system design permitting that holistically)...
    I look forward to your re-thought system and chip design. Seriously. CPU should become a lightweight I/O processor that gives GPU cores access to different levels of memory and direct access storage.
    WS-HEDT
    "Workstation - high-end desktop"? I've never come across the HEDT acronym before, I guessed "hybrid electronic design tools."

    Leave a comment:


  • F.Ultra
    replied
    Originally posted by guglovich View Post

    That's what I'm writing about. Maybe my translator got it wrong.
    It's less translation and more that TDP is an often misleading term. To make it even worse, a CPU from AMD and a CPU from Intel with the exact same TDP or even CPU:s from the same vendor but from different generations will both draw different amount of power and dissipate different amounts of heat.

    One have to benchmark the actual power usage to make any determination regarding consumption.

    E.g the 5800x3d have a TDP of 105W while the 7800x3d have a TDP of 120W. In actual measurement with all cores on max usage for Blender the 5800x3d draws 108W of power but the 7800x3d draws only 86.4W:



    edit: not sure how to get the timestamp feauture in YT to work with Phoronix, the chart is at 07:03

    Leave a comment:


  • bridgman
    replied
    Originally posted by guglovich View Post
    32 and 24 I mean average. But anyway, this is the first time I've seen such a thing, that in one lineup the same amount of TDP between different models. If anyone has an example from the past, I'd be interested.
    In general as the core count goes down the clock speed goes up, so total power doesn't vary much. The 32 core part has a 4.0 GHz base clock while 24 core is 4.2 GHz. I'm sure there are minor differences in power between the SKUs but the overall approach is to have a range of parts that can all function within the same power envelope.

    If you look at Zen4 consumer parts you will see a similar pattern - 12 and 16 core parts are both 170W, while 6 and 8 core are both 105W. Remember that we still have 4 CCDs on the 24 core TR parts - we disable 2 cores on each CCD but the rest of the chip including data fabric and the big L3 cache remains fully active.

    You see a similar pattern with EPYC parts - the high clock 32, 24 and 16 core parts all have the same TDP, but they also have roughly the same cache. We offer lower TDP EPYC parts with reduced cache and lower clocks, but TR parts are generally max everything.
    Last edited by bridgman; 20 October 2023, 06:26 PM.

    Leave a comment:


  • guglovich
    replied
    Originally posted by bridgman View Post

    Sorry, what is "the middle model" ? Maybe we are looking at different slides, or maybe I am taking non-boost clocks into consideration while you are not. I don't think TDP is just "base or average"... power-limiting factors into boost as well (with the caveat that I'm on the GPU side and not up on the nuances of CPU power management).

    If you are talking about 32 vs 24 core it's probably a mix of "yeah 24 core probably draws less power than 32 core but not enough difference to go through the exercise of qualifying a different TDP" and "the more efficient bins tend to get used in the higher core count SKUs".
    32 and 24 I mean average. But anyway, this is the first time I've seen such a thing, that in one lineup the same amount of TDP between different models. If anyone has an example from the past, I'd be interested.

    Leave a comment:


  • guglovich
    replied
    Originally posted by stesmi View Post

    However, the Max Boost is not Max All-Core Boost, so if you're idling 22 out of 24 or 30 out of 32 and full enchilada on 2 of them, the 24- and 32-core versions aren't going to behave any differently and boost the same. I expect, if we would look at the boost tables, that they will start differing the more cores are used.
    Yes, it is. All the more reason to have better and worse cores, under the same cover.

    Leave a comment:


  • guglovich
    replied
    Originally posted by F.Ultra View Post

    TDP is not consumption
    That's what I'm writing about. Maybe my translator got it wrong.

    Leave a comment:


  • F.Ultra
    replied
    Originally posted by NM64 View Post

    Oh happy day! I mean, it makes sense since I'm pretty sure it's the same I/O die, hence why it sounds like RDIMM is now required.

    ...actually, is it absolutely confirmed that UDIMM is not supported at all?

    Now the question is, why doesn't Epyc and/or Threadripper integrate a small iGPU into its giant I/O die since servers and stuff commonly have a motherboard-integrated iGPU for various management things...

    (also, you didn't need to quote my whole post, especially the ARM part where my line-break got accidentally stripped out because I had javascript disabled :P)
    simply because the motherboard vendors use "their own" BMC cards for such features so a iGPU would just be disable anyway.

    Leave a comment:


  • skeevy420
    replied
    Originally posted by NeoMorpheus View Post

    Theres an interesting section in this video that points to price per core:

    https://youtu.be/LgUee00qXIY?si=TXSW4lsnc8cqTXhL
    And here I am doing the math on the best valued one and thinking, "Drew, That's Too Much!"

    Leave a comment:


  • NeoMorpheus
    replied
    Originally posted by DavidBrown View Post

    These are very much aimed at high-end professional users, not personal use.

    For example, companies that work on chip design can easily be paying $100,000 per year per license for the software they use. These will be floating licenses, shared amongst engineers in a team, that check out licenses as needed for long-running tasks such as simulation, thermal analysis, placing and routing, etc. If $10K processors for 8 users on the team mean they get these tasks done fast enough that the company can get away with one shared license fewer, then those $10K processors save them money.

    Or if an engineer costs the company $120K per year and the $10K processor makes them 10% more efficient, it saves them money.

    The pricing set by AMD is probably influenced by how much their own engineering and development departments would pay for them.
    Theres an interesting section in this video that points to price per core:

    Leave a comment:

Working...
X