Join us For A Tom's Hardware Ask Me Anything With AMD, Thursday April 6th

[cleeve]

Question
1. Intel Optane will not support?

The short answer is no. The long answer is...

1. Optane is Intel-proprietary technology, an the Optane M.2 slot is exclusive to some Intel motherboards

2. Intel partnered with Micron to create 3D Xpoint memory technology that Optane is based on. I don't know if Micron's 3D Xpoint-based memory will ever be available as an agnostic solution.
I would assume that Intel has an exclusivity clause, but I don't know how long it'd last.

3. In its current form for consumer desktop, Optane is basically an SSD cache drive with a maximum (pitiful) 32GB of storage. They don't even recommend pairing it with an SSD because you wouldn't notice a performance difference. They suggest you pair it with a mechanical hard drive. Lots of hype and little substance.
On the consumer desktop, you're better served with an SSD that actually has decent amount of storage space.

 

[PandaNation]

Two questions:
1. Why are freesync monitors so much cheaper than their G-Sync counterparts?
2. I know you won't be able to say much, but how does Vega compare to the 1080 ti and Titan XP?

Big fan of AMD, thinking of doing a mini-itx Ryzen 5 build. Keep up the good work!

 

[realJX3]

Not sure if this thing is over or not...

1. Can you clarify the number of PCIe lanes (including disabled). Are there 32 on the actual die?
2. NVMe 4k write perf is off, any upcoming software fixes'?
3. Memory BW and latency scales great and at 4000MHz Summit Ridge would somewhat catch up with Broadwell-E bandwidth as Intel has scaling issues. Any chance that at least some better chips will be able to run such DRAM clocks without Bclk OC for 24/7 setups not suicide runs?
4. Can you disclose the data path (and number of hops) for CCX to CCX communications? Can you disclose some official numbers for CCX to CCX latency and BW? Is there a trade-off AMD knowingly made here in an attempt to lower in-package latency? Is the IMC next to the GMI links?
5. What are the 2 U shaped cache units on either side of the GMI links'?
6. There is evidence that Nvidia GPUs have issues with 8 cores (from both Intel and AMD) under DX12. With quad cores going from DX11 to DX12 results in small perf gains while 8 cores see a large loss in perf. Not ideal to ask you for a comment on Nvidia's ability to fix the issue in software but, it's either that or AMD launches Vega already.... * the evidence for Nvidia's issues is hard to doubt, i can provide 3rd party data if needed.
7. Total War Warhammer got the "core count fix" but we know that F1 2016 and Ghost Recon Wildlands have similar issues - actually not certain that AMD knows about Ghost Recon. Any other titles that have the issue? And can we assume that the titles that see a large perf gain with SMT disabled are mostly suffering from this issue? -BF1, Dishonored 2 , Deus Ex and so on
8. If a CPU vendor had a 16 cores chip, would it make sense to price it low enough for a customer to be able to purchase both the CPU and a 200$-ish mobo inside 1k$?
9. Can you provide any hints on Naples off-die in-package latency?
10. AMD focused a bit too much on MT perf with the R5 quads but the ST clocks are making the choice between Intel and AMD more difficult than it could be. especially since shops don't list XFR clocks. if the quads had similar clocks to the hexa and octa cores, sales would go much better. How does AMD plan to fix that.

Ok that's all i could come up with on the spot.
EDIT:
11. The ability to add a 1-2 cores boost when overclocking would be great
12. PCIe 4.0 - Do we need it in desktop and when might AMD transition to it
13. Someone else asked about 8 core in laptop but you assumed he/she meant APU so i'll ask again. Why not an 8 core CPU (no integrated GPU) in laptop? A low power octa cores(45W maybe) paired with a mobile Vega would be just great for high end gaming and workstation notebooks.
14 Can we have a Summit Ridge die keychain please?
15. When thinking TDP, folks tend to include XFR but XFR clearly expands the TDP so maybe AMD should clarify that.
16 When Vega launches, willit be just Vega 10 at first or we might see more, like Vega 11 and dual Vega 10 from day one?
17. In theory how would one solve the scaling issues with a dual GPU(on 1 PCB)?
18. Ryzen in Chromebooks?
19. Why not release the ARM version of Zen and use the same socket? Would be nice to see that next year.
20. Is there an official number for Summit Ridge die size?
21. Why Ryzen 3 in H2 when that's a lot of time lost? Is it about capacity or a respin?
22. When can we expect better mobo availability and is it about to get much worse short term with Ryzen 5 about to launch? AMD should be able to have an estimate here based on chipset shipments.
23. Are X300 based mobos still scheduled for April or is there a delay?
24. Why are the dLDOs mostly disabled with Summit Ridge,is it a thermal issue at such high clocks but they'll be enabled with Naples?
25.Can the Infinity fabric and the HBCC improve scaling in Crossfire with APU+discrete Vega later this year?

 

[anironbutterfly]

anironbutterfly asked:
I've been reading on the new Ryzen CPUs, in hopes that they're a good successor to the FX-series (I'm currently using an FX-8350 on an original Sabretooth FX990 motherboard with 32 GB DDR RAM). I'm not a gamer, but a hobbyist graphic artist who uses Poser and DAZ|Studio. It's starting to show it's age, and I'm looking at options to upgrade. the Ryzen series are the first new straight CPUs I've seen come out of AMD in several years.

I'm curious how this new series of chips might perform for 3d graphics rendering with the nVidia Iray render engine (and the alternative 3Delight rendering) in comparison to the i5 and i7 Intel CPUs and compatibility with the nVidia Geforce video cards. (I'm currently using an EVGA nVidia Geforce 970 GTX 4GB that will be progressing into my new build).

Thanks for your time!

***ANSWER BY DON***
I'm not familiar how the Nvidia iRay engine works - I'll assume because it's Nvidia, it's CUDA based. It may not be CPU dependant.

For any CPU-dependant renderer, though, Ryzen will give you colossal - and I mean COLOSSAL - performance increases over FX.

And in general, it's just a lot faster and enables multi-tasking in a much more responsive way.

Even the sub-$200 Ryzen 5's will give you a tremendous upgrade over the FX. But I encourage you to read the launch day reviews on April 11th.

 

[Onoref]

Question by Onorief:
I'm really hyped about the new Ryzen CPU's and am already planning to upgrade the two computers I got at home.
But there is one thing bugging me. Especially with the new M.2 SSD's comming out, will the restriction on PCIe lanes be an issue and what is your take on it?
thank you!

***ANSWER BY DON***
Our M.2 performance is very good, especially NVMe, as a lot of Ryzen's I/O goes direct to the CPU - not through a chipset.

I encourage you to check out the storage benchmarks in Ryzen 5 reviews.

 

[tanke001]

Question by Tanke001:
Good Morning.

I'm saving to update my computer from a PhenomII X6 1055 to a full Ryzen platform. I use it mainly for videogames. My question is: if it has been confirmed that L3 shared cache memory is creating bottlenecks with its actual config, can it be a good idea to split it into 4MB per CCX by software?

Thanks in advance.

***ANSWER BY DON***
There's a lot of cache FUD out there about Ryzen, and a lot of people are making assumptions based on limited info.

Just keep an eye on actual reviews. As we ramp up the platform and get rid of bottlenecks like memory speed, we get faster. And I can confidently say Ryzen is a lot faster than the Phenom II out of the box when it comes to gaming.

 

[orifiel]

Question by Orifiel:
Is AMD working with ram vendors for more ram options to be available? All I love is an AMD RYZEN 1700X with 32gigs of ram and at 3000-3200mhz. There is also issues with 4 dim solution and people buy 2 dims instead. The bad news for me is I can only get corsair ram and there is no options at that speeds and amount of ram yet. So I wait and wait and I cannot buy something that may not work yet.

can I hope for a fix with rev2 and firmware updates?

ps: I am replacing my intel pc.

***ANSWER BY DON***
Ram compatibility is one of our top priorities right now, and we're working hard to get regular BIOS updates until the situation is ideal. Expect a new Bios around April 11th, and in May to make memory work faster on Ryzen.

 

[cleeve]

Two questions:
1. Why are freesync monitors so much cheaper than their G-Sync counterparts?
2. I know you won't be able to say much, but how does Vega compare to the 1080 ti and Titan XP?

Big fan of AMD, thinking of doing a mini-itx Ryzen 5 build. Keep up the good work!

1. Freesync is cheaper because it's an open standard. In many cases a panel manufacturer can make a Freesync panel by changing their monitor's firmware and having it meet the spec.
For GSync, Nvidia charges a licensing fee.

Because of this differentiator alone, I think the inevitable future is Freesync.

2. It looks really nice.

3. I am also waiting for my MiniITX Ryzen board!

 

[Eric_010]

Question by Eric_10:
Hello,
Is amd working with Battlefield 1 developer DICE to improve performance on ryzen cpus?
In multiplayer I have alot of framepacing issues and some input lag.
I have 1700x 3,95Ghz with a gtx 1070.

***Answer by Don***
We're engaging with every major developer we can to make sure the Ryzen gaming experience only gets better. For now I'd play with the Battlefield 1 DirectX setting and detail to optimize the game, offhand I don't recall hearing about these issues onBF1 and Ryzen so this might be a problem specific to your system. But there's a lot of data being tracked and I apologize if it's a known issue I can't recall at the moment.

 

[fourseven]

Question by fourseven:
just simple question.
I'm live in Indonesia, so how long ryzen 5 will arrive in South East Asia especially in Indonesia after Official Launch?
I'm looking forward for ryzen 5 'cuz I have plan to upgrade my sons system A.S.A.P.
thanks.

***Answer by Don***
Official on-shelf launch day is April 11th worldwide. I'm not sure if your specific country has any challenges that would prevent that, but that's the day we expect Ryzen 5 to be available on shelf.

 

[gc9]

Hi,

Has AMD released performance monitoring MSR event code details for Ryzen? (partial if necessary, such as those identical in BKDG 16h). If not, when might event code details be released? (weeks from now? quarters from now?)

Context:

Normally Ryzen-specific MSR event code documentation would be published in the Ryzen BKDG ("BIOS and Kernel Developer's Guide for Family 17h"), but understandably that document includes many other low-level details that might not be ready for publication yet.

John Taylor (CVP Marketing, AMD) implied [1] that CPU-bound programs, such as 1080p game configurations, need to be optimized for AMD's Ryzen separately from Intel's chips. So getting performance monitoring event codes published seems high priority.

At least one performance measurement library developer [2] and CPU optimization manual writer [3] have asked for documentation on details of Ryzen performance monitoring MSR event codes.

[1] https://www.pcper.com/news/Processors/AMD-responds-1080p-gaming-tests-Ryzen

[2] "Performance Monitoring Documentation for AMD Ryzen (Zen)?"
"... the system programming manual does not contain the available events for the Zen architecture. The registers are the ones listed in the manual. But without the event codes, you never know that is measured exactly. ... event code for K16 can be a different event on K17 ..."
https://community.amd.com/thread/214140

[3] "Need BIOS and Kernel Developer’s Guide Family 17h"
"... even if just a draft."
"... I still need ... a list of performance events."
https://community.amd.com/thread/213584

 

[cstephenson]

Hello,
When do you expect the Ryzen chips to be competitive with Intel in terms of gaming? The potential is certainly there!

 

[elbert]

Will there be motherboards this summer that supports a larger capacity of RAM? I would like an 8 DIMM motherboard or 4 DIMM that supports upto at least 128GB's. Be it an AM4 or the leaks around a Snowy Owl/X390/X399 motherboard.

 

[Ditt44]

Cleeve: having just read on pcworld that AMD has a new power plan available, is this something that we will see integrated with products after "Date X" or will users have to manually download and update? I'm a hold-out until at least next Tuesday and your Vega 'stuff'. Thanks!

 

[James Mason]

Two questions:
1. Why are freesync monitors so much cheaper than their G-Sync counterparts?

1. Freesync is cheaper because it's an open standard. In many cases a panel manufacturer can make a Freesync panel by changing their monitor's firmware and having it meet the spec.
For GSync, Nvidia charges a licensing fee.

We have a whole thread dedicated to explaining the differences about the various Syncs here on tom's:
http://www.tomshardware.com/forum/id-3355120/sync-free-sync-sync-adaptive-sync-fast-sync.html

 

[Tech_TTT]

I missed the place

I hope I am not late

Okay, some Questions :

1- Are we expecting an AMD APU with onboard HBM2 Memory as shared memory for both System and GPU and no DIMMs slots any time sooner? That would fit in a very small box 1/4 the Area of the Mini-ITX case. for example : 16GB of HBM2 Memory ?

2- Why dont you introduce a new form factor to the Market ? Mini ITX needs to have 2 slots for 2 GPU systems while still short (170mm) and not long like the Micro ATX (245mm) and .. in short (3 slots motherboard for 2GPU and still compact 170mm length)

3- Why dont you manufacture AMD motherboards ?

4- What are your plans for very low voltage CPU ? The Ryzen managed a good 65 watts for 8 cores .. can we expect a 15 Watt 4 cores Ryzen APU to compete with Intel low voltage CPU ?

5- Why did you choose to go dual channel memory and not quad or eight channels for the Ryzen ?

6- the same goes for the PCIe lanes , why just 16 lanes while the server CPU can reach up to 128 lanes ? and the competing Intel i7 X99 CPU offers upto 40 lanes ...

7- I have this Idea , When you Design the CPU , why dont you space out the components a little bit inside the cpu so that the cpu has more surface Area and hence better cooling ? When I look at the CPU , no matter the die shrink , the "spacing" design is the same .. why dont you take advantage of the better process to space between the components internally giving you better cooling potential and bigger surface area for cooling ? I understand the need of smaller CPU Area inside phones and compact machines , but for the desktop we have the space for a big CPU... so why dont we take advantage of the smaller process by keeping the same old bigger area and space out the cpu Areas internally?

Thats all , Thanks.

 

[cleeve]

Hello,
When do you expect the Ryzen chips to be competitive with Intel in terms of gaming? The potential is certainly there!

I'd argue that Ryzen is *already* competitive. There's been a lot of reviewers talking about outliers where we don't do as well, and bringing a lot of focus to them. And those outliers are where we've focused our first-round of developer engagements. Games like Ashes of the singukarity, Total War: Warhammer, and DOTA2 already have improvements.

With the new faster memory support, we average game performance delta between Ryzen and Kaby Lake is a lot closer than you'd think over a wide swath of games at 1080p. And Ryzen can hit over 60 FPS in pretty much every game I've seen at 1080p, and usually over 80 FPS and 120 FPS. It's never slow, it's just not the fastest. At 1440p, 4K, and in VR, the delta becomes insignificant between Ryzen and Kaby..

Based on that I think it's fair to say we're already quite competitive, we're just not just not the fastest at 1080p gaming.

Saying Ryzen isn't a competitive gaming CPU because Kaby is a bit faster is like saying the Ferrari 488 isn't a competitive sports car because the Bugatti Veyron is faster. It's gross oversimplification.

 

[cleeve]

Cleeve: having just read on pcworld that AMD has a new power plan available, is this something that we will see integrated with products after "Date X" or will users have to manually download and update? I'm a hold-out until at least next Tuesday and your Vega 'stuff'. Thanks!

For now its a manual download. Our long-term goal is definitely to get it automatically in Windows, but I don't have a target date on that yet, sorry.

 

[cleeve]

I missed the place
I hope I am not late

Okay, some Questions :

1- Are we expecting an AMD APU with onboard HBM2 Memory as shared memory for both System and GPU and no DIMMs slots any time sooner? That would fit in a very small box 1/4 the Area of the Mini-ITX case. for example : 16GB of HBM2 Memory ?

2- Why dont you introduce a new form factor to the Market ? Mini ITX needs to have 2 slots for 2 GPU systems while still short (170mm) and not long like the Micro ATX (245mm) and .. in short (3 slots motherboard for 2GPU and still compact 170mm length)

3- Why dont you manufacture AMD motherboards ?

4- What are your plans for very low voltage CPU ? The Ryzen managed a good 65 watts for 8 cores .. can we expect a 15 Watt 4 cores Ryzen APU to compete with Intel low voltage CPU ?

5- Why did you choose to go dual channel memory and not quad or eight channels for the Ryzen ?

6- the same goes for the PCIe lanes , why just 16 lanes while the server CPU can reach up to 128 lanes ? and the competing Intel i7 X99 CPU offers upto 40 lanes ...

7- I have this Idea , When you Design the CPU , why dont you space out the components a little bit inside the cpu so that the cpu has more surface Area and hence better cooling ? When I look at the CPU , no matter the die shrink , the "spacing" design is the same .. why dont you take advantage of the better process to space between the components internally giving you better cooling potential and bigger surface area for cooling ? I understand the need of smaller CPU Area inside phones and compact machines , but for the desktop we have the space for a big CPU... so why dont we take advantage of the smaller process by keeping the same old bigger area and space out the cpu Areas internally?

Thats all , Thanks.

1. We're definitely considering different HBM implementations, but we haven't announced anything I can talk to

2. Cool idea, but it's not the kind of thing I think AMD could drive by itself. We'd need a bunch of partners onboard, including system integrators/OEMs who feel the idea would have significant demand. I like the way you think though.

3. Frankly, our partners do a better job and offer more differentiation and flavor than AMD would want to. We're happy to concentrate on the processors and leave the boards to the specialists.

4. In a lot of ways the Zen architecture gets more impressive as you provide less power. I can't comment on unannounced laptop parts, but there are great things coming!

5. We really decided to focus on whats best for the market. Our goal is to have a platform that competes with low-end Intel boards all the way up to high-end Intel Extreme. After analyzing the benefits, the real-world advantage of quad-channel RAM doesn't outweigh the extra costs or tradeoffs. The vast majority of users will never see the difference. Heck, the dual-channel 1800X can still beat the tar out of the quad-channel 6900K in many benchmarks. I think it was a good compromise for the vast majority of users. From an enthusiast perspective, it's always nice to have more, though, so I get it.

6. Basically the same as answer 5. Rather than jacking up specs just to have the highest number, we really looked at what people needed, and what gives the best experiences. I think our compromises are very well balanced.

7. I won't pretend to be a processor architect and give you an answer with any authority. I do know the engineers are orders of magnitude smarter than I am, so I'm going to defer to their expertise. I do think one of their objectives is to keep trace lengths short to reduce latency. But from a layman's standpoint what you're saying makes sense for sure.

 

[Tech_TTT]

I missed the place
I hope I am not late

Okay, some Questions :

1- Are we expecting an AMD APU with onboard HBM2 Memory as shared memory for both System and GPU and no DIMMs slots any time sooner? That would fit in a very small box 1/4 the Area of the Mini-ITX case. for example : 16GB of HBM2 Memory ?

2- Why dont you introduce a new form factor to the Market ? Mini ITX needs to have 2 slots for 2 GPU systems while still short (170mm) and not long like the Micro ATX (245mm) and .. in short (3 slots motherboard for 2GPU and still compact 170mm length)

3- Why dont you manufacture AMD motherboards ?

4- What are your plans for very low voltage CPU ? The Ryzen managed a good 65 watts for 8 cores .. can we expect a 15 Watt 4 cores Ryzen APU to compete with Intel low voltage CPU ?

5- Why did you choose to go dual channel memory and not quad or eight channels for the Ryzen ?

6- the same goes for the PCIe lanes , why just 16 lanes while the server CPU can reach up to 128 lanes ? and the competing Intel i7 X99 CPU offers upto 40 lanes ...

7- I have this Idea , When you Design the CPU , why dont you space out the components a little bit inside the cpu so that the cpu has more surface Area and hence better cooling ? When I look at the CPU , no matter the die shrink , the "spacing" design is the same .. why dont you take advantage of the better process to space between the components internally giving you better cooling potential and bigger surface area for cooling ? I understand the need of smaller CPU Area inside phones and compact machines , but for the desktop we have the space for a big CPU... so why dont we take advantage of the smaller process by keeping the same old bigger area and space out the cpu Areas internally?

Thats all , Thanks.

1. We're definitely considering different HBM implementations, but we haven't announced anything I can talk to

2. Cool idea, but it's not the kind of thing I think AMD could drive by itself. We'd need a bunch of partners onboard, including system integrators/OEMs who feel the idea would have significant demand. I like the way you think though.

3. Frankly, our partners do a better job and offer more differentiation and flavor than AMD would want to. We're happy to concentrate on the processors and leave the boards to the specialists.

4. In a lot of ways the Zen architecture gets more impressive as you provide less power. I can't comment on unannounced laptop parts, but there are great things coming!

5. We really decided to focus on whats best for the market. Our goal is to have a platform that competes with low-end Intel boards all the way up to high-end Intel Extreme. After analyzing the benefits, the real-world advantage of quad-channel RAM doesn't outweigh the extra costs or tradeoffs. The vast majority of users will never see the difference. Heck, the dual-channel 1800X can still beat the tar out of the quad-channel 6900K in many benchmarks. I think it was a good compromise for the vast majority of users. From an enthusiast perspective, it's always nice to have more, though, so I get it.

6. Basically the same as answer 5. Rather than jacking up specs just to have the highest number, we really looked at what people needed, and what gives the best experiences. I think our compromises are very well balanced.

7. I won't pretend to be a processor architect and give you an answer with any authority. I do know the engineers are orders of magnitude smarter than I am, so I'm going to defer to their expertise. I do think one of their objectives is to keep trace lengths short to reduce latency. But from a layman's standpoint what you're saying makes sense for sure.

6- Actually , more lanes means better SSD PCIe Cards and RAID cards , I wished for at least 16 (GPU) +8 (Raid/x8 PCIe SSD cards) =24 Lanes with the Ryzen ... many people use multi 4 lanes SSD Cards , or 8 lanes SSD cards , and the M2 is not enough for this . Keep in mind that 8 lanes SSD cards will for sure come in the next 2 years for the consumer market , and that they do exist for servers as we speak ... given that AM4 will stay for 4 years , this is not good. 8 lanes SSD are coming.

7- Can you please pass this idea to your AMD Engineers ? I think you can space out between Areas inside the CPU , like space between cores, the Memory controllers , I meant area Spacing not every component spacing. I think it is possible to a certain length .

Thanks for your reply

 

[James Mason]

7- Can you please pass this idea to your AMD Engineers ? I think you can space out between Areas inside the CPU , like space between cores, the Memory controllers , I meant area Spacing not every component spacing. I think it is possible to a certain length .

Thanks for your reply

I really doubt a few nanometers difference is going to be enough to do anything about heat. Power efficiency is where you'd find heat savings, as even incredibly old CPUs still run as hot or hotter than today's CPUs.

 

[Tech_TTT]

@James Mason ,

I am not talking in nano meters total result of the spacing , I think you can space between the cores more than that. I am talking about more surface area for the final product , I hope they can double the surface Area. and this will make heat pipes of the coolers closer to the cpu and not far away from it , transferring heat faster to the cool area ... in real today only one heatpipe is directly over the CPU , the others are just near it and not over it .. if it is bigger , the next heatpipe will be closer , or partly over it as well.

 

[James Mason]

@James Mason ,

I am not talking in nano meters total result of the spacing , I think you can space between the cores more than that. I am talking about more surface area for the final product. and this will make heat pipes of the coolers closer to the cpu and not far away from it , transferring heat faster to the cool area ... in real today only one heatpipe is directly over the CPU , the others are just near it and not over it .. if it is bigger , the next heatpipe will be closer , or partly over it as well.

if they're that far apart, they'd likely suffer HUGE performance impacts because of the difference in time it takes to reach the other cores/modules

 

[Tech_TTT]

@James Mason ,

Thats why I am trying to have an Engineer to answer this ... and to know the maximum length that you can separate them without noticeable impact on performance.

 

[realJX3]

@James Mason ,

Thats why I am trying to have an Engineer to answer this ... and to know the maximum length that you can separate them without noticeable impact on performance.

Just to end this since it's offtopic, cost. Do you want to pay 2x for next to no returns?
The old school metric is PPA - perf, power , area and that's what matters. The chip maker will do its best to balance those and reach its goals..