AMD's Future Chips & SoC's: News, Info & Rumours.

Page 111 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Then why hasn't Microsoft done some easy things to help Ryzen? Like preselecting a power plan that is actually beneficial to the CPU.

I get the scheduler is complex, but there are some easy changes Microsoft can make, but doesn't.

AMD made some deep level changes to its power management from the tech white papers I have read. You could yurn up the dials but then power management goes out the window and that may eventually work against you in burst modes.
 
A little different but I felt like saying this.
Ask GM Ford and AMC/Chrysler during the 70's and 80's or Sears JCPennys or Montgomery Ward in the 90s or Yahoo and MySpace in the 2,000s
Hopefully, Apples refusal to quickly fix the butterfly keyboard design comes back to bite them.

I was using a brand new Macbook Air the other day and I truly was surprised.

The machine feels premium and the screen is great. I also was surprised just how snappy the machine felt even with just a mobile dual core.
However, the keyboard was literally the worst keyboard I have ever felt. Basically the least travel I ever have felt and virtually no feedback. The keys are not contoured at all so I had trouble finding the correct keys. I literally enjoy the RELIABLE keyboard of my old netbook I got for free more than I like the keyboard on this 1100+ laptop. I have used a 25b year old membrane keyboard from a compaq and i liked it more.

To add insult to injury, the butterfly switches fail so commonly that apple keeps having to recall machines with the butterfly switches and now lifetime warranties them but still refuses to fix the dang design. All of their new MacBook models with the same flawed design and they have known about this issue for many years.
 
Last edited:
The butterfly keyboard issue reminds me of the ford pinto.

Ford knew their cars would blow up if rear ended, yet instead of fixing the problem they kept sellung flaws cars. They said it would br cheaper to pay the families of the dead victims of their design flaw than it would be fo fix the issue.

In the same repect apple knows their keyboard are prone to failure yet they still keep selling the laptops with bad keynoards. They find it cheaper to just pay to replace peoples dead keyboards that it is to fix the issue.
 
Round 2, FIGHT:

https://www.anandtech.com/show/14924/tsmc-countersues-globalfoundries-seeks-manufacturing-injunction

The key point here is this:

In their complaint, TSMC is demanding injunctions against GlobalFoundries, asking the courts to stop GlobalFoundries from making and selling chips using the allegedly infringing technologies. Which, given the broad nature of TSMC's claims, essentially covers all of GlobalFoundries' production lines in some form or another and would seemingly shutter GlobalFoundries manufacturing operations entirely. The company is also seeking "substantial monetary damages" for prior infringement.

Break out the popcorn.
 

InvalidError

Titan
Moderator
AMD confirms that B550A is a refresh of B450 made for the OEM's only. The actual B550 chipset will be different.
Hopefully it will support PCIe Gen 4.
AFAIK, the real B550 specs have already leaked and it will only support PCIe 3.0, which is still an upgrade over older AM4 chipsets that only supported 2.0-speed downstream lanes. Really wanted to see 4.0 CPU link to reduce the likelihood of chipset bottlenecking, better luck on B650-series boards next year on that one I guess. It is still unclear whether B550 boards will be allowed to run host-based lanes at 4.0 speeds.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
After seeing what they've accomplished with the TR 3K siblings, I'm really interested in what they have next with their Zen 3 based products.

I wonder how they'll improve on what they already have.

Cheers!

@Yuka me too me too...

I cannot help thinking AMD will probably be ready for whatever Intel throws at them, I believe they are very aware of the threat and have not taken their foot off the gas at all...
That SMT4 is lurking in the wings probably getting perfected, no need to drop it yet I guess especially if it is not working properly... why spoil a perfect record. I'm sure there's 3d stacking being looked at as well an who knows what butter do-nought chiplet design is next, not to mention heterogeneous compute... what a great time to be alive in Tech....
Full steam ahead... warp 10 :)

View: https://youtu.be/A2BLLBSd3Yc
 
Last edited:

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
Guy's interested to hear your thoughts on an SoC for laptops with say two 6c chiplets instead of one chiplet and a GPU, small Navi can then be used on the MB for graphics... This would be a beast of laptop right... what are power requirements for such an SoC ? ?
It's is not impractical in my mind as 6c chiplet's (you can verify with a simple google search) are not constrained, that's point !
In fact it would make more sense than using up precious 8c chiplets that would make more money if sold into the server market or even used in 16c desktop (or HEDT etc) sold for top dollor etc, no ? Two defective lower binned 6c put in an SoC without a GPU sounds like an excellent idea to me.
I don't think it's a wish, (I'm sure it has been thought of and possibly planned for also, hopefully on the way, depending on power requirements etc) I think it will happen, it is just a question of when it will happen. If not on 6th Jan maybe later in the year. But it makes a lot of sense, to penetrate the laptop market the same approach that is being taken in the server market may be required, produce a product that is so much better they simple cannot ignore it and must sell it to remain competitive.
 

InvalidError

Titan
Moderator
In fact it would make more sense than using up precious 8c chiplets
8c chiplets aren't "precious" by any means, the whole point of chiplets is to be able to treat them as a more disposable commodity in the first place. Most of the die area (~60%) is L3 cache and half of the remaining area is fabric logic, reducing the number of cores per chiplet wouldn't make much of a difference in total cost per chiplet and would divert production capacity away from EPYC/TR/Ryzen.

There is no point in making a separate 6c part when the 6c demand can be covered by 8c defects and surplus without incurring the added re-design, re-qualification, split manufacturing, split inventory, split support, etc. costs to save 5-7% on die size.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
8c chiplets aren't "precious" by any means, the whole point of chiplets is to be able to treat them as a more disposable commodity in the first place. Most of the die area (~60%) is L3 cache and half of the remaining area is fabric logic, reducing the number of cores per chiplet wouldn't make much of a difference in total cost per chiplet and would divert production capacity away from EPYC/TR/Ryzen.

There is no point in making a separate 6c part when the 6c demand can be covered by 8c defects and surplus without incurring the added re-design, re-qualification, split manufacturing, split inventory, split support, etc. costs to save 5-7% on die size.

Because the higher binned 8c chiplets are going to the server market.. an frankly they are the most valuable to AMD an most profit is achieved by selling these in server/hedt and desktop in that order...

While lower binned 6c Chiplets are in affect defective 8c chiplets and are not in short supply right now or in as much demand either.

I am not inferring that reducing the cores reduces space, in fact I'm aware that it does not... actually I wasn't concerned about space at all...I'm more concerned about "power requirements" & Efficency etc...

Edit: 8 Core Chiplets are very precious to AMD especially the higher binned ones... the are the most valuable of the lot... an achieve the most profit when sold into the server market... Demand for ROME is huge !
 

InvalidError

Titan
Moderator
I am not inferring that reducing the cores reduces space, in fact I'm aware that it does not... actually I wasn't concerned about space at all...I'm more concerned about "power requirements" & Efficency etc...
Disabled cores are powered down and don't consume any meaningful amount of power, so there is no point in wasting resources making and managing completely different chiplets specifically for that.
 

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
After seeing what they've accomplished with the TR 3K siblings, I'm really interested in what they have next with their Zen 3 based products.

I wonder how they'll improve on what they already have.

Cheers!

Hey Yuka :)
Faster cache and more of it is going to have a big affect on IPC from what I am hearing so far.

View: https://youtu.be/6Hn0ML-8Xjw


This chap has became a little bit biased IMO over time... his take is always still interesting... but from what I know he wanted a marketing job off AMD and never got it... so he has developed some sour grapes over it again imho, but still a good analyst although now clearly pro Intel an spending his time dreaming up ways they can come back and beat AMD... while still it is worthwhile listening all theories of coarse but jus be mindful to remain objective etc...He actually came out in one of his video giving out about AMD not giving him his marketting job.
 
Last edited:

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
Disabled cores are powered down and don't consume any meaningful amount of power, so there is no point in wasting resources making and managing completely different chiplets specifically for that.

Apparently it is the IF that is too power hungry to do it... but Asus jus dropped in a Desktop CPU, So I don't see why they cannot simply drop in a 3900x and design a MB around it, battery would not be great but it would be excellent for certain workloads... my m8 for example the photographer that moves around a lot, is looking for core count over anythings else an he does not care about battery life etc... he is using Premier and editing video an rendering etc... He was looking at a mini itx but he'd have to drag a screen around with him a laptop is ideal but core counts are still pretty low unfortunately.

Asaik the 6c are not purposely made as 6's they are failed 8c that get binned for lower sku etc... but two 6c would make 12c laptop (4 more than the 8c that is coming out in Jan Renoir) and two 8c would be a savage 16c 24t beast but jus too power hungry I guess... if power consumption is not a concern and you don't care about batterie life, then it would be awesome in a "Laptop Workstation" I think... what a beast of a machine. I hope someone builds it :)
 
Last edited:

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
Excellent discussion here, very in dept look at possible future strategy and hardware, SMT4, Zen3, Zen4 even, Raytracing, RDNA, Resource Contention.... more or less everything in detail. Coreteks is really technical IMO and full of really good insights, I like Moore's Law's take on the Market and Market Strategy also.

View: https://youtu.be/kC-Sjy9h_fg
 
Dec 27, 2019
37
8
35
Amd sure learned their lesson from faildozer i'm so happy to see them back and beating Intel it's like the Athlon FX-53 days all over again.

Papermaster. “You have to be very thoughtful when you add cores because you don’t want to add it before the application can take advantage of it. As long as you keep that balance, I think we’ll continue to see that trend.”

https://www.pcgamesn.com/amd/ryzen-cpu-core-saturation-point

Zen 3 is rumored to have 12% higher IPC over Zen 2 this would put it so far ahead of Intel in IPC probably higher then Amd was back in 04.

Amd is focusing so much on IPC first and that means they will continue to kick butt. Chasing frequency isn't a smart choice as the smaller the node will not allow it with high yields Intel is facing this issue i have no idea how they will get their 4Ghz 10nm part to perform even on par with their 5ghz core in just 1 gen.
 
  • Like
Reactions: jaymc
Dec 27, 2019
37
8
35
Look at Tiger Lake leaks, Intel may be ready for its comeback on 10nm++.
I read it only hits 4.0Ghz that means for it to come even to their 9900K it would have to have 25% higher IPC.

Not only that when is the desktop version coming out i just hear about their laptop CPU's? Like every article i read briefly talks about desktops if at all.

Zen 4 will be here in 2021 and then Zen 5 in 2022. So i'm just confused a bit.
 

InvalidError

Titan
Moderator
I read it only hits 4.0Ghz that means for it to come even to their 9900K it would have to have 25% higher IPC.
Ice Lake has ~18% better IPC than Coffee/Comet Lake and Tiger Lake is ~30% faster than Ice Lake watt-for-watt. ~10% of that is clocks, which leaves ~20% to IPC. Looks like a 4GHz Tiger Lake should be able to eat a same core count 5GHz Coffee/Comet Lake CPU for lunch.
 
Dec 27, 2019
37
8
35
Yes but overall per core performance will be less plus does it scale beyond 4ghz? Not doubting that it can just questions to keep in mind the last thing Intel wants to do is ruin their only advantage.
 

InvalidError

Titan
Moderator
Yes but overall per core performance will be less plus does it scale beyond 4ghz? Not doubting that it can just questions to keep in mind the last thing Intel wants to do is ruin their only advantage.
Clocks alone are meaningless. Throughput = IPC * clock. If you have 30-40% more IPC at 4GHz in typical workloads, you get per-core throughput equivalent to the old chips at 5.2-5.6GHz, usually at a lower energy cost.