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

[jasaero]

Are there efforts to get more of the embedded products into products that show up more in consumer channels as some Intels competing products such as the Xeon D do in easily found mITX and other boards for home server and other uses. Also seems the Ryzen V1000 would be a great product to get into mini HTPC sorta barebones systems....but as of right now nothing is showing up and only partners for like Casino and digital signage and such have announced product plans.

 

[Markfw]

Soon(tm)

Thanks for being a fan - and buying Threadripper. I look forward to hearing your thoughts as you find out more about 2nd Gen Threadripper

Well, I can say I love the first gen TR, and the 2700X only adds to it. Mine is doing 4.1 all-core 24/7@100%load on 1.3 vcore. If I get the new TR and it has the same type of results or better, I will be very happy, and keep buying. I have used 3600 LL (cl14) memory @3466 on all my Ryzen/TR boxes, and its worth every penny.

 

[dcominottim]

Question from a Java developer's point of view: does AMD already contribute code to OpenJDK|Oracle Java JRE/JDK for Zen optimizations?

Also, how would you describe the reaction from the developer community to Zen? Have you perceived high engagement from them and received lots of feedback about improvements they need on the software stack and/or hardware?

 

[moinmoin]

Currently the secure memory encryption feature is limited to Epyc and Ryzen Pro processors. Will it ever be made available to consumer Ryzen chips?

 

[0ldman79]

Yes, and the good news is that it's enabled through an open source compute library that people already use today - OpenCL 2.0.

Now the nerd in me is thinking...

I'd love to see some testing between the heterogeneous computing of the APU vs the Ryzen 7. Seems there may be a situation where the APU could outperform the high end unit.

Any plans to add a GPU to the six or eight core Ryzen?

 

[Hitman928]

Maybe slightly off the wall question but is the interest into developing an ARM based processor dead at AMD? Not looking for specifics, just curious if AMD has completely abandoned it or is still looking into it to some degree internally.

 

[dlerious]

Just one question. Can you send a review kit to Markfw (maybe X499 Taichi, 2950X, DDR4-3600)?

 

[eva02langley]

Do you know any reason why HPET is not affecting AMD at all while Intel is cripple by it in gaming performance when the CPU is bottleneck?

 

[ET]

You mentioned OpenCL 2.0. Is AMD planning to support later versions of the API? Looks like currently Intel is the only one supporting 2.1.

 

[Obvcop]

Is there any chance you will enable gpu virtulisation on consumer GPU's or Vega Frontier edition such as SR-IOV.
It isn't just proffesional workloads that could use this, home users are more commonly building systems that can serve the entire household in one package or people who use computers in a semi-professional enviroment.
Given that AMD are constantly pushing the advantages of more cores in the consumer/prosumer sphere with threadripper and ryzen, it seems only fair that could can use these cores in conjuction with your GPU's.

 

[MajinCry]

Thanks for sharing the experiment. I'll pass it along to our engineering team, for them to look into. There's not a lot of controlled variables in your testing so drawing conclusions and saying it's an architecture issue seems rather premature, if not misleading. That chart shows a 16% advantage for Haswell over Skylake. Time for pitchforks about the regression of Skylake vs Haswell? I kid, kinda, it's a good area for AMD to look into, but I'm not sure this testing is a "smoking gun" as it were.

We had users with NVidia cards supply results, but their driver has an optimization that only applies for scenes calling the exact same draw call, with no changes, over and over (i.e, floating rocks with no materials, shadows, lights, parallax maps, etc).

We also found that Windows 10 had a performance increase for draw calls over Windows 7. It's a minor gain, a few percentage points, but it's still there.

That Haswell dual core result is an anomaly, I should have not put it in, admittedly. But if you look at the quad core Haswell result, it's much more in-line with what was expected. Skylake made quite the jump in draw call performance over Haswell.

 

[bearmoo]

It has been like five month since the anoucement of VEGA MOBILE. Can you share some update? Or maybe when can we expect to hear more about it.

 

[caveman-jim]

Question from a Java developer's point of view: does AMD already contribute code to OpenJDK|Oracle Java JRE/JDK for Zen optimizations?

Also, how would you describe the reaction from the developer community to Zen? Have you perceived high engagement from them and received lots of feedback about improvements they need on the software stack and/or hardware?

It's been a good reaction, mostly in the learning phase. I'm not sure on the feedback that's been shared back with us.

 

[caveman-jim]

It has been like five month since the anoucement of VEGA MOBILE. Can you share some update? Or maybe when can we expect to hear more about it.

We spoke with a lot of media outlets, including Anandtech, when we briefed them as part of the launch of 2nd gen Ryzen. We have new designs coming from every global OEM this year, so you'll see a lot more choice in notebooks!

 

[caveman-jim]

Currently the secure memory encryption feature is limited to Epyc and Ryzen Pro processors. Will it ever be made available to consumer Ryzen chips?

I believe it is enabled but not tested/validated on Ryzen - needs the setting to be exposed in the BIOS. I don't think it has much appeal outside of commercial or enterprise markets - meaning I don't think gamers want to run encrypted memory.

 

[ParalLOL]

What magnitude of impact does memory latency problem have on Ryzen? Can/will latency be reduced back to Bulldozer levels with Zen 2 or is it some kind of tradeoff for something much more important?

Links for readers:
https://img.purch.com/r/711x457/aHR...vbS9OL0IvNzY1NDc5L29yaWdpbmFsL0ltYWdlMzEucG5n
https://www.pcper.com/reviews/Processors/Ryzen-Memory-Latencys-Impact-Weak-1080p-Gaming
https://www.sisoftware.co.uk/2018/0...arks-2-channel-ddr4-cache-memory-performance/

 

[Markfw]

What magnitude of impact does memory latency problem have on Ryzen? Can/will latency be reduced back to Bulldozer levels with Zen 2 or is it some kind of tradeoff for something much more important?

Links for readers:
https://img.purch.com/r/711x457/aHR...vbS9OL0IvNzY1NDc5L29yaWdpbmFsL0ltYWdlMzEucG5n
https://www.pcper.com/reviews/Processors/Ryzen-Memory-Latencys-Impact-Weak-1080p-Gaming
https://www.sisoftware.co.uk/2018/0...arks-2-channel-ddr4-cache-memory-performance/

Well, if you read some of the threads we have, much of the latency problems can be fixed by using 3600 CL14 memory, and then tweaking the subtimings. Performance increases can be very significant. I don't remember all the percentages, but first, the memory makes a lot of difference, and then the subtimings make a lot, and combined its really significant, possibly on the order of 20-40% total.

Just read the other threads here.

 

[.vodka]

Well, if you read some of the threads we have, much of the latency problems can be fixed by using 3600 CL14 memory, and then tweaking the subtimings. Performance increases can be very significant. I don't remember all the percentages, but first, the memory makes a lot of difference, and then the subtimings make a lot, and combined its really significant, possibly on the order of 20-40% total.

Just read the other threads here.

Well said Mark.


For posterity...

There's not even a need to go that far, 3200 CL14 memory + one of Stilt's profiles with proper subtimings will get you 90% of the way. Of course B-die will get the best results, with some luck Hynix based memory can be pushed but not tightened as much as B-die. Best to avoid it despite it playing much nicer with so many AGESA updates since Ryzen/AM4 launched.

Subtimings are the key in this platform when you get to 3000-3200MHz memclk and above. Just loading the XMP profile and leaving the motherboard to set the subtimings is a complete waste.

So yeah, memory latency isn't really an issue with properly set up, fast memory in Ryzen 1xxx and 2xxx parts. Of course it would be nice if future Zen iterations would not be as reliant on such fast memory to get the most out of the SoC, especially with how crazy memory pricing has become...

Spoiler: mindblanktech's testing on 1xxx and 2xxx parts

https://www.youtube.com/watch?v=RZS2XHcQdqA

 

[Markfw]

Well said Mark.


For posterity...

There's not even a need to go that far, 3200 CL14 memory + one of Stilt's profiles with proper subtimings will get you 90% of the way. Of course B-die will get the best results, with some luck Hynix based memory can be pushed but not tightened as much as B-die. Best to avoid it despite it playing much nicer with so many AGESA updates since Ryzen/AM4 launched.

Subtimings are the key in this platform when you get to 3000-3200MHz memclk and above. Just loading the XMP profile and leaving the motherboard to set the subtimings is a complete waste.

So yeah, memory latency isn't really an issue with properly set up, fast memory in Ryzen 1xxx and 2xxx parts. Of course it would be nice if future Zen iterations would not be as reliant on such fast memory to get the most out of the SoC, especially with how crazy memory pricing has become...

I wish the subtimings would have a "profile", and be defaulted/tied to the XMP profile. Thats more work than just use XMP, then select the speed (max 3466 that I have seen)

 

[Muki]

I just saw the Thunderbolt 3 + Ryzen Pro announcement! How is that possible? Is there already a controller chip AMD has access to that supports Thunderbolt 3? Is it compatible with Titan Ridge?

 

[caveman-jim]

What magnitude of impact does memory latency problem have on Ryzen? Can/will latency be reduced back to Bulldozer levels with Zen 2 or is it some kind of tradeoff for something much more important?

Links for readers:
https://img.purch.com/r/711x457/aHR...vbS9OL0IvNzY1NDc5L29yaWdpbmFsL0ltYWdlMzEucG5n
https://www.pcper.com/reviews/Processors/Ryzen-Memory-Latencys-Impact-Weak-1080p-Gaming
https://www.sisoftware.co.uk/2018/0...arks-2-channel-ddr4-cache-memory-performance/

Better, more recent testing shows that the impact of sub-timings on memory is around 10-15% - https://www.computerbase.de/2018-04/pinnacle-ridge-speicher-tuning/ (GER).

The overclocked memory sphere is tuned for our competitor, so we're all learning together how to adapt and utilize these high speed memory kits to best effect. I really like the work the awesome community at overclock.net did with the Ryzen DRAM calculator. Those are some great timings suggestions, takes 85% of the effort out so you're really left with what your silicon and memory chips and motherboards can do.

Latency is one of the things we're looking at, but it's not the only thing.

 

[moinmoin]

Regarding memory timings, any chance of a return (or upgrade) of AMP?

 

[caveman-jim]

Regarding memory timings, any chance of a return (or upgrade) of AMP?

Possibly - we need to put together what that would do. Call it 'being actively scoped'? No promises but open to suggestions from the community of what an AMD or Ryzen Memory Profile program would look like. No exclusive branding suggestions, please.

 

[BlackBishop]

James - could you please answer my question about documentation submitted on page 3?
TY in advance.

 

[Yotsugi]

Will 35W APUs be available in retail boxes?