Skip to main content

AMD Ryzen AMA

Enterprise, Best GPUs, CCX Latency, and Scheduling

mikeangs2004: Are the Ryzen CPUs stable enough for use in an enterprise environment? Are APUs a lower R&D priority compared to high core count CPUs?

DON WOLIGROSKI: Personally, I would have no reservations over recommending Ryzen for enterprise clients. It’s a crazy good part. Ryzen-based APUs are well on their way. I think it's great that we could stagger releases to make sure we could give each part the proper focus it deserves.

FireDFL: Which Ryzen CPU would you recommend for a gaming/overclocking/rendering build? I'll be doing light-intermediate rendering before I become an advanced renderer. Also, which GPU do you recommend to pair with any Ryzen CPU if I want to game in 1080p and 1440p at 60FPS?

DON WOLIGROSKI: If you plan to render at all, get the best Ryzen you can afford. It's that simple. The good news is that at every price point you're getting rendering performance that blows away the pre-Ryzen status quo, which was 4 threads below $275 (we deliver 12), and 8 threads under $400 (we deliver 16).

Honestly, you can get great 1440p game performance with a Radeon RX 480 or a GeForce GTX 1060 under $200. Only spend more if you want to crank up detail levels and anti-aliasing.

cryoburner: You mentioned that there are already performance optimizations for certain games such as Ashes of the Singularity. What exactly do some of these software optimizations for Ryzen entail? I've heard that communication between cores on the same CCX might be significantly faster than it is between cores on disparate CCX units. Are these optimizations due to updated code keeping threads that communicate often on the same CCX? Would this avoid the increased latency between different CCX cores?

DON WOLIGROSKI: It’s all over the map, there's no silver bullet, even though that's what people want to hear. The CCX latency is there, but it's not that bad and it's not responsible for the outliers.

I'll give you an example of the kids of things that are holding Ryzen back: a developer found that their game code automatically assumed that AMD CPUs had all-physical cores, because we didn't have SMT before now. Once the game was guided to behave as it does on Intel HyperThreaded CPUs, we saw a notable boost in performance.

It sounds simple, but this is what happens when a new architecture is introduced. It sounds trivial, once you know what's happening it can be easy to attack, but finding it takes work.

wildcard1978: When can we expect a fix for the BIOS, RAM speed bugs, and scheduling issues?

DON WOLIGROSKI: Ryzen RAM speed and compatibility is improving all the time. We have a huge BIOS update enabling 3200 MHz DDR4 that should hit most boards April 11th, and there is another update scheduled for May.

As for scheduling issues, there aren't any “issues” per se. Windows is doing what it's supposed to do. The balanced power plan wasn't working optimally and we fixed that with an updated plan you can download from the AMD website.

That's not to say we can't work with Microsoft to make the schedulers work better in the future, but there's no problem right now. It's working as designed.

Bursar: What is the IPC improvement you are looking for between Ryzen and Ryzen 2?

DON WOLIGROSKI: Personally, I'm looking for as much performance uplift as possible! We haven't disclosed anything yet, but I'm quite optimistic.


MORE: Best CPUs


MORE: Intel & AMD Processor Hierarchy


MORE: All CPU Content

AMD Ryzen 5 1400View Deal
  • BugariaM
    Many people ask clear and technically interesting questions, hoping to get the same answers ..

    And they are answered by a person who is far from the technical plane and the engineering questions.
    He is a manager, he is a salesman.
    His task is more blah-blah, only for the sake of an even greater blah-blah.

    Thanks, of course, but alas, I found nothing interesting for myself here.
    Reply
  • genz
    I intensely disagree Bugariam. All the info he could provide is provided and he asked people actually close to the metal when he did not know. You will not get tech secrets or future insights from ANY AMD or Intel rep on tomshardware; Its far too public and every drop of information here is also given to Intel, Nvidia, and any other competitors hoping to steal AMDs charge. What we did get is a positive outlook on AMD's products.... when you compare that to what we already had from Toms and other publishers who have spent years watching Intel lead and thus don't have faith (or simply got their jobs for their love of Intel) was major.

    I personally think he did not remind us that the current crop of 8 core consoles will inevitably force AMD's core advantage to eat all the competition Intel currently has. In 5 years every single Ryzen 1 processor will terrorize the Intel processors they competed with.... Ryzen 5s will have 50% performance gains over Kaby i7 etc etc.

    Intel knew this was the future, that is why all Intel consumer processors have stuck to 4 cores to try and keep the programming focus on their IPC lead. Now that that lead is only 6% and the competition has more cores, we will see the shift toward 6+ cores that we saw when Core 2 Duo came and made dual FX and Dual Pentiums viable mainstream gaming chips, and when Core Quad and Nehalem made quad cores viable gaming chips.

    As the owner of a 3930k, you can read my past posts and see I have always said this is going to happen. Now, a month after you are seeing the updates come out already. Wait till there are 12 threaded games on the market (this year I expect) and you will see just how much the limitation of the CPU industry's progress was actually created by Intel's refusal to go over 4 cores in the mainstream.

    For all the talk of expense creating 6 and 12 core processors, Intel could have had consumer 8 core low clock chips in mainstream for prosumers and home rendering types years ago and they didn't. My theory is that they are scared of heavily threaded applications in the mainstream creating opportunity for competition to outmanouvre their new chips based on slower, more numerous cores. It's not like a 2ghz 6 or 8 core in the mainstream was never an option.
    Reply
  • Calculatron
    I remember being really excited for the AMD AMA, but could not think of anything different from what everyone else was already asking.

    In retrospect, because hindsight is always 20/20, I wish I would have asked some questions about Excavator, since they still have some Bristol Ridge products coming out for the AM4 platform. Even though Zen is a new architecture, there were still some positive things that carried over from the Bulldozer family that had been learned through-out its process of evolution.
    Reply
  • Ernst01
    As a long time AMD Fan it is so cool AMD has more in the future for us.
    Reply
  • TJ Hooker
    "TDP is not electrical watts (power draw), it's thermal watts."Argh, this kind of annoys me. "Electrical watts" and "thermal watts" are the same thing here, power draw = heat generated for a CPU. There are reasons why TDP is not necessarily an accurate measure of power draw, but this isn't one of them.
    Reply
  • alextheblue
    Thank you Don!
    Reply
  • Tech_TTT
    19562297 said:
    Many people ask clear and technically interesting questions, hoping to get the same answers ..

    And they are answered by a person who is far from the technical plane and the engineering questions.
    He is a manager, he is a salesman.
    His task is more blah-blah, only for the sake of an even greater blah-blah.

    Thanks, of course, but alas, I found nothing interesting for myself here.

    I agree with you 100% ... Ask me anything should include people from the R&D department and not only sales person. or maybe a team of 2 people , Sales and Research. or even better? the CEO him/herself included.

    Reply
  • Tech_TTT
    @Tomshardware : WE DEMAND APPLE AMA !!!
    Reply
  • genz
    19566458 said:
    "TDP is not electrical watts (power draw), it's thermal watts."Argh, this kind of annoys me. "Electrical watts" and "thermal watts" are the same thing here, power draw = heat generated for a CPU. There are reasons why TDP is not necessarily an accurate measure of power draw, but this isn't one of them.

    That is simply not true.

    Here's an example. 22nm and 18nm TDP is usually far higher than actual draw because the chip is so small any cooling solution has a much smaller surface area to work with. Another example: When Intel brought over onboard memory controllers from the bridge to the CPU socket, the TDP of their chips went unchanged because (thermally speaking) the controller was far away enough from the chip to never contribute to thermal limitations... despite the temperature of the chip rising much faster under OC because of the additional bits, and the chips themselves drawing more power due to more components. A final example: I have a 130W TDP chip that without overvolting simply cannot reach a watt over 90 even when running a power virus (which draws the max power the chip can draw - more than burn-in or SuperPi). The TDP rating is directly connected to the specific parts of the chip that run hot and how big they are, not their true power draw. This is why so many chips of the same binning have the same TDP despite running at lower clocks and voltages than each other.

    Further to that, TDP is rounded up to fixed numbers to make it easy to pick a fan. True power draw is naturally dependent on how well a chip is binned, and super badly binned chips may still run with enough volts so they usually add 10 to 20 watts for the thermal headroom to make that possible.
    Reply
  • TJ Hooker
    @genz I never said TDP is equal to power draw, in fact I explicitly said there are reasons why it isn't. I simply said that "thermal watts" (heat being generated by the CPU) are equivalent to "electrical watts" (power being consumed by the CPU). At any given moment, the power being drawn by the CPU is equal to the heat being generated.

    I'll admit, I'm sort of nitpicking a small part of the answer given in the AMA regarding TDP, I just felt the need to point it out because this is a misconception I see on a semi regular basis.
    Reply