AMD AMA - Starts at 12PM EST on Wednesday May 9

[R0H1T]

Hello James, can we expect a KBL-G equivalent from AMD (with HBM) anytime soon?

Also on a totally unrelated note ~ Ryzen or Threadripper, your personal favorite?

 

[cbn]

Any chance we will see an Enthusiast level X300 or B300 motherboard? This allowing the advantage of two NVMe PCIe 3.0 x 4 slots while still retaining PCIe 3.0 x 16 for graphics? (Something that cannot be done on current enthusiast motherboards)

P.S. Going a bit farther with X300/B300 I also think a board with PCIe 3.0 x 16 bifurcations* (for use with NVMe SSD cards like this one) would be great for certain tasks as well.

*Like we see on TR4 boards.

 

[Praefator]

Hi James!

Two questions for you:
1) Are there still improvements to be expected in terms of memory compatibility, either for my 1700x or perhaps for Zen 2/3? For instance, Samsung B-die seems to be the only type of memory which can reliably hit 3200MT/s when memory is rated as such, yet Hynix chips continue to suffer
2) Can you lift the veil ever so slightly on what to expect from the next (last) two years of AM4? If not, when might we expect news to make its way through the usual channels?

Thanks!

 

[Aspiring Techie]

Hey James
What do you think is the reason that Ryzen lags so far behind Intel's Coffee Lake with regards to overclocking. GloFo's 12nm process certainly helped, but there's still a ~600 MHz difference.
Do you think that Global Foundry's process is just not as good as Intel's 14nm++, or does it have something to do with the Zen architecture? If it does have something to do with the Zen architecture, will Zen 2 smooth out those issues?

 

[MajinCry]

When the GPU driver thread is on a separate CCX from the thread it's communicating with, Ryzen suffers a significant drop in draw call performance. Even when paired with blazing fast RAM, it only manages to be as fast as Core 2 at processing draw calls.

We did a benchmark to see the draw call performance of specific CPU architectures, which gave us the above findings.

And even when the driver thread is communicating to another thread on the same CCX, with blazing fast RAM used, draw call performance is around Ivybridge's. Considering that Ryzen has single thread performance on par with Broadwell, why is it that Ryzen is not as good at processing draw calls?

xLake Intel CPUs are very good at draw calls, with them being 23% faster than Ryzen at processing draw calls despite only being ~5% faster in single thread performance. Will draw call performance be targeted for significant improvement in future architectures?

 

[TGSpeed]

Hi James,

1) What is Mode0 in the latest 4011 UEFI bios? There's no info for this anywhere and I asked ASUS (I'm using a prime x470 pro board with my 2700x) but they told me it's a feature AMD added and that I would have to ask them instead.

2) ASUS also told me I need to ask this to AMD as well so here goes: Under the Precision Boost Overdrive settings of the motherboard, there are 3 settings that appear when Precision Boost Overdrive is set to “Manual.” These are PPT Limit, TDC Limit, and EDC Limit. How are these settings intended to be adjusted? I have a general idea of what these settings do, but I am not certain of what these limits are supposed to be and what is a safe range to set the limits.

Thank you.

 

[caveman-jim]

Hello there!

1. Threadripper Gen 1 had the unfortunate issue of having a fairly large latency penalty for crossing dies. I'm wondering, would there be anything with Threadripper Gen 2 like say an interposer to reduce latencies involved? I understand if this cannot be answered

2. Anandtech's review claims Pinnacle Ridge's L2 latency further dropped from 12 cycles in Threadripper/EPYC/Raven to 11 cycles. Is that correct, or did they misunderstand company messaging?

3. Are OEM's embracing Raven Ridge in prebuilt desktops? It was claimed that Ryzen Gen 1 was held back there because OEM's prefer to not having to deal with a dGPU. Raven Ridge seems like a perfect package for OEM's!

4. Can a single memory controller support both DDR4 and DDR5? I know Intel's Skylake does both DDR3 and 4 depending on motherboard, but I'm not sure if they have a single controller or multiple ones.

5. This is more of a suggestion than a question. How about an ultra binned part for overclockers? If you remember the old TWKR processors, basically that. Those things were awesome, and I'd love seeing them make a return!

1 - Can't speak to future products, but AMD engineering teams are looking at ways to take the infinity fabric to new heights, as that is the method by which the two die communicate. Currently, it seems all high core count chips have synthetically measurable latency differences for data that needs to move from one side of the die to the other, be they monolithic or multi-chip module approaches. I think what really matters is that the product offers the right level of performance for the investment, and what kind of scalability the platform as a whole offers.

2 - L2$ latency was reduced by upto 34%, compared to first gen Ryzen.

3 - We're very excited to see how "Raven Ridge" will continue to ramp up in desktop and notebook PCs, now that they notebook Ryzen 7 procesors have launched, and we've launched the desktop Ryzen 5 & 3 parts, as well.

4 - Specific memory support is driven by the fundamental design and the validation of implementation of platform design. AMDs "Carrizo" processor, for example, contained a memory controller that supports both DDR3 and DDR4. To use both capabilities requires motherboard and BIOS support.

5 - I remember the TWKR processors well... I'll put it on the list of things to consider!

 

[caveman-jim]

Hello James.
Do AMD have plan to release Athlon successor (<$90 price bracket)?

Unfortunately I can't speak to future plans or speculate on future products.

 

[caveman-jim]

Hello, James! Two questions, please!

1) In the context of the AM4 socket, what are your views on PCIe 4.0 now that the final specification has been released? Is there any chance of current PCIe 3.0 motherboards being *forward* compatible with PCIe 4.0 after a BIOS update when paired with an hypothetical CPU that supported PCIe 4.0, just as some Intel Z68 motherboards that had only PCIe 2.0 at launch eventually became PCIe 3.0 compatible when paired with an Ivy Bridge CPU?

2) The first and second gen Ryzen chips are known to have lower gaming performance in some scenarios due to high inter CCX latency, which needs special software optimizations to be partially mitigated. Does the Zen core team see this as something that eventually ought to be better addressed in hardware with further optimizations to Infinity Fabric?

1 - I think it's too early to understand what the chances are, yet. I remember some of those motherboards you mention having some issues, where not every board that initially offered the capability ended up with the full speed/performance/stability. Stability and performance would be the number one concern for trying to backport this kind of capability, so extensive regression testing would need to be performed - as well as products that contain the feature. So very hard to answer this one... personally I think it would be cool, but ultimately an engineering analysis on feasibility/expense would guide the end result.

2 - We've sampled hundreds, if not into the thousands, of Ryzen desktops to our software partners, to help games/apps offer the best possible experience on Ryzen. There's a lot of software and developers out there, and AMDs presence at shows like GDC help share the knowledge on the abilities and architecture our new core offers. From a gaming perspective, I don't know that I would blame inter-CCX latency as the reason for why gaming performance is the way it is, we saw some large gains in 2017 from game patches that didn't understand the memory architecture, the core features etc. Overall I think the gaming experience is really strong with Ryzen, and the enthusiast community deep diving into the parameters for tuning performance to take it to the next level is only a good thing. We're definitely watching the feedback and developments and finding ways to incorporate it into future products as and when we can.

 

[caveman-jim]

I do some tech support for the elderly living in the area to help them out. But last year was a nightmare when it came to "firmware/sub zero ring/out of OS control"
updates or insecure devices that work functionally 100% but from a security standpoint are bricks.

So my question is :
Why do CPU makers in general feel like putting so much extra functionality in end users products that out of OS control. Or is that me asking for a car without wheels?

Follow up question :
Will we see in the future a trend off better firmware upgrade paths in CPU design or better OS control over CPU's?

PS I know a certain company had a lot more issues than AMD

First, thank you for helping elderly people with their tech issues. I'm sure that's appreciated!

The features that go into AMD processors are there for two reasons; one, because of the need to support requests by customers (which ranges from global OEMs to individual users) and partners (software and hardware, large and small); and two, to help develop the future vision of computing that needs to be enabled. AMD works closely with the security community - you can always find the latest on AMD.com here - https://www.amd.com/en/corporate/security-updates - as well as the OS providers to give the necessary level of control where it belongs. Not everything can be achieved at once, and scope changes over time can impact schedules as well; it's always a resource and validation balancing act.

 

[caveman-jim]

Hi James,

I have a couple of questions, feel free to answer all or any of them.

1. Do you guys at AMD actually spend any money, time or research to understand what gamers want or are looking for out of their hardware?

2. Does AMD anticipate DDR5 will have a massive impact on the improvement of infinity fabric products?

3. Is AMD looking to increase single-threaded performance out of Zen2 or will it be geared more towards SMT improvements?

4. Does AMD consider Ryzen a success over Bulldozer or previous AMD CPUs?

5. Is power efficiency considered important to RTG products? Or is performance the main focus? Or is it a combination in the form of performance per watt?

Thanks again James!

1 - Yes, of course. What we actually do ranges from third part blind focus groups, to engaging with other partners in the gaming space, as well as watching comments and sentiment in online forums. We take gamer needs very seriously.

2 - Too early to say - obviously, anything that is bandwidth bound will benefit from more bandwidth in the memory subsystem. There's a couple ways of addressing that, you can add more channels of memory (i.e. quad channel on Threadripper, or eight channels on EPYC).

3 - Sorry, can't disclose any details on Zen 2.

4 - I'm personally very pleased with the launches of all the Ryzen products so far. I joined AMD in 2013, so don't have a lot of insight into how the original thinking around previous products compares. But if you look at some of the things AMD has focused on - like "Building Great Products" - I think Ryzen checks that box firmly.

5 - I'm a CPU guy, so can't speak to RTG... I can say that our engineering division, all consider perf/w in every decision, every day.

 

[caveman-jim]

Could the chipset driver branding be changed? Right now it's confusing as it looks like you're installing Radeon drivers. I've had a lot of friends message me after they got Ryzen about why I'm telling them to install Radeon drivers when they have NVIDIA.

Good feedback, I'll check with the software team.

 

[eva02langley]

I have 3 questions:

1. Is Vega 7 nm refresh is in the plan for the consumer discrete GPU?
2. Is AMD planning to develop a high-end APU for gaming on a single chip? Like a console without the need of discrete GPU for 1080p PC gaming?
3. Is Navi aimed at being a mid range GPU or a high end GPU?

 

[caveman-jim]

Hello Will there be any improvements to the first gen Ryzen CPU's. I.e. improved precision boost , any 1st gen features. Will we get further improvements to ryzen master for 1st gen cpus so that we also can tinker just like the new 2nd gen cpus?

The improvements in Precision Boost 2 - the new algorithm - require the faster transistor and new frequency voltage curve of the new process technologies we used for Ryzen 2000 series. So, unfortunately not possible to respin the 1st gens to add it in. However, you will see some of the BIOS improvements for memory apply to both 1st and 2nd gen.

 

[caveman-jim]

Another question: some engineers from motherboard makers have talked about a major pain point related to the way AMD handles AGESA updates, because it seems AMD usually changes some settings' default values between AGESA revisions without fully specifying the changes in any documentation that is disclosed to its partners. This causes a lot of trouble for end-users because every new AGESA has to be tested by pure trial and error, which contributes to platform instability. Does AMD have any plans to improve the AGESA documentation that is disclosed to motherboard makers?

The feedback we get directly from the motherboard manufacturer teams is on the whole positive. I'm sure there are areas for improvement on both sides of the fence.

 

[Dragonstongue]

Hello James hope you all are doing well over at AMD ^.^
amazing year since Ryzen was launched and now we have Ryzen 2 (I think the naming could have been better orchestrated in regards to Ryzen 1000 would be called Ryzen 1xxx, Ryzen 2 would be called Ryzen 2xxx and so forth less likely to potentially confuse people)

not trying to bait or flame any questions of course, just want this put out there before any reading is done, thank you however for being here for folks to ask questions and get some burning questions answered ^.^

---------------------------------------
------------------

My question are, is there any plans to have RX 500 also on 12nm (refresh coming from the RX 5xx which currently use GF 14nm)
Also in this regard, are there any plans to have a direct replacement of my much loved but as of yet not replaced Radeon 7870 (1280 shaders 256 bit memory bus and whatever TMU/ROP is deemed "appropriate" ? )

^ when are we likely to see closer to MSRP for RX 500/Vega seeded to the market for us gamer types that have been wanting to upgrade for awhile and cannot either find any model or the ones that are out there are quite a bit above pricing they should be.

along this context, with the new Ryzen 2xxx models such as Ryzen 2600-2700 even though was claimed to be slightly lower cost I have been tracking the pricing of them for a new build hopefully in the near future and the pricing of them seem to often be above what they should be (Canada) whereas the day of launch they were within ~$5 of price they should be, and since this point the Ryzen 1xxx pricing has been all over the place which is odd...us Canadians are computer users as well ^.^

I really wish there was more control on what resellers can charge for "fair pricing" but alas little can be done about this without killing any potential places to buy/list the products.

Not sure why AIB can slap a dinky little cooler on their "custom cards" that they have used many times before and jump the price over reference by $25-$100 however, seems there needs to be at lest some "quartermaster" in the mix to make sure everyone is "being fair"

------------------------------------------------
----------------------

why does AMD not offer to sell the AIB cards directly from AMD website (have not noticed this of course, but there were many generations in the past the reference coolers were amazing looking vs the ones that were on store shelves, is a pride thing) ?

--------------------------------------------
-----------------

A suggestion/question, are there plans to allow/enable more direct control of GPU clock/voltage settings with new Radeon drivers, what I mean is through my experience having went through driver after driver
some of them seem to be more able to allow the user full control of constant clock/voltages whereas other ones no matter what you do you cannot control these

so for example, watch a youtube video it will automatically through the clock into high 3d settings and highest voltage number even when it does not need it
whereas when play a game it will sometimes remember what you wanted it set to and other times it will auto jump the clocks/voltages for numbers you did not tell it to use.

if there is a idle clock, mid (low 3d) clock and a high performance clock, it would be nice if there was the ability to control this just like it appears that Ryzen is now allowing for the CPU side
so the question is, are there any plans for future Radeon GPU drivers to allow an even finer granularity of clock/voltage settings or at the very least to "force" to maintain the settings the user selects?

 

[caveman-jim]

Another question: according to recent comments made by AMD to media outlets, it seems Precision Boost Overdrive is a feature whose official implementation is still being worked on by AMD. Is there any ETA on when it might be ready?

Soon(tm)



We'll update with more details when ready - sorry not be more precise, but for right now it's a pre-release feature and we want to make it a good as possible before unveiling it completely.

 

[eva02langley]

Maybe another one...

Is infinity fabric able to be implemented on GPUs?

And by the way, how is the morale at AMD lately?

With Spectre-NG around the corner and the CTS-labs story, do you feel that flaws disclosure process should be overlooked by government agencies since they are one of the biggest stakeholders.

 

[jwapk]

Will the Q2 launch of more notebook models based on Raven Ridge contain models with higher TDP (>15W) processors (e.g. YM2800C3T4MFB), or models with a 25W cTDP up enabled? It would be great to have better competition against Intels H/U-series.

 

[caveman-jim]

Q1: Are there any plans to improve driver vs software / games as part of support for APUs ?
Q2: Any lessons learned from Ryzen 2xxx for new APUs?
Q3: For second generation of Zen APUs, is it going to be just process refresh or any chance to see increase in CU count ?
Q4: Is AMD addressing the lack of small form and very small form boards for Zen+ and beyond ?

1 - Yes, all of those are areas we want to make even better.

2 - We could have been clearer on how to communicate the boot kit process. I think that the approach of not forcing new motherboards, but allowing BIOS updates to maintain compatiblity was the right choice, but more details on how/where to get support would have helped reassure some people.

3 - Can't discuss future products, sorry.

4 - The motherboard manufacturers are focusing on the volume purchase areas of the markets first. That said, there should be new options for SFF boards this year. However, because of the BIOS compatibility strategy, all the existing boards can be compatible as well - but regression tests on BIOS take time, so the BIOS don't appear all at one time.

 

[caveman-jim]

Why has there been so few pre-builds with Ryzen processors from vendors such as HP/Dell/Lenovo?

It's early days yet, more designs will appear over the course of the year.

 

[caveman-jim]

On the topic of the newly released AMD Combat crates. Will we see more variety among them? I'd like to see some with second gen Ryzen chips, an Asrock motherboard, and either an RX 580 or Vega 56 graphics card included.

We're looking into expanding the regions, and product partners, for the AMD Combat Crates. Glad you like the idea!

 

[caveman-jim]

Another question: do you have any info on how many software companies are already using uProf, especially game studios? Is AMD actively promoting uProf to game studios?

We are working with many software partners to promote the use of uProf; not sure how many/who, or if I can disclose that.

 

[Justinus]

Why has there been a distinct lack of quality micro ATX boards for X370 and X470? Does AMD have any influence over what the board partners do with regards to their product lines, or is this simply the board partners decision entirely?

 

[Vincent]

Hello James,

Do you know of any motherboards that support ECC RAM with ECC correction enabled for Raven Ridge?

I ask because I have tried multiple motherboards and found that they support ECC correction with Summit Ridge but not with Raven Ridge. ASRock, for example, specifies that ECC correction is enabled only for Raven Ridge Pro CPUs but not vanilla Raven Ridge CPUs. On reddit AMD employees have said that Raven Ridge is just like Ryzen CPUs where ECC is available and merely requires motherboard support. Can you shed some light on Raven Ridge ECC support?

Thank you.