AMD CPU speculation... and expert conjecture

Page 337 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.


There can't be too many Orochi dies where two modules are bad. If there were, we'd more than likely see a bigger price difference between FX 6000 series and FX 4000 series.

Regardless it's about to become an absolutely useless product once SR APU comes out, because now it will just be a slower CPU without the HSA and GPU.

I'd be expecting AMD to try and make the difference in CPU performance between APU and CPU much larger to separate the products right now. Someone going FX 4000 series today who doesn't know better is making a huge mistake.

 
excerpt from blackkstar's link (trans. google, originally from vr-zone chinese):
AMD Kaveri is a very important product, mainly because it imported the HSA architecture (Heterogeneous System Architecture) design.

The AMD platform will continue to follow the current FM2 + pin, but will also use A88X chipset, and this chipset will also continue to be used until 2015 Carrizo.

According to the information in the timetable, the motherboard makers have received this APU for testing, mass production will need to wait until after December, but I believe AMD will be held in November APU 13 mentioned Developers Conference More information about this product. Both are expected to have an A10 and A8 APU debut.

If it is not published in the APU 13, that the maximum possible time will be published in CES 2014, but this is only Paper Launch, official debut will have to wait until February 2014.
Click to expand...

seems that either kaveri has gone into production late, possibly due to redesigned b-core, or yields are low that amd won't have enough till feb '14. or both. i'll err on the side of caution and blame glofo as usual. :lol:
 



As an image:

2.png
 


Like seronx and others have stipulated in the past. SteamrollerA was cancelled for whatever reasons. Among the list I'm thinking

>Change of process
>Massive performance increases
>HSA features added

AMD's traditional HEDT October CPU release SHOULD HAVE been SRA but instead SRA got canned.

So maybe you guys (except for juaranga) will give me some credibility over my theory of SR being delayed and replaced with SRB and AMD just skipping SR for HEDT and going directly to EX?

If AMD were to release HEDT SR chips it would put them at a mid-summer release or so given their delays. Meaning AMD would then either have to push EX on HEDT back yet again for another 6 months or they could jump 6 months ahead and release EX earlier than they would have.

Thinking if they released HEDT SRB we would see:

summer 2014: SR HEDT
fall 2015: EX HEDT

Now AMD can rush EX HEDT out and just ride on software improvements for FX series.

Wouldn't be surprised to see HEDT EX in early to mid 2015 for desktop.

By then the gaming ecosystem will be absolutely rich with software that works well on AMD module architecture from console ports and it will put Intel in a very bad spot for gaming CPUs.
 
What are the thoughts about this Steamroller-B redesign? I think it sounds like bad news. Speculation that they delayed it by 6 months to get HSA compliance in doesn't ring true to me. I think if the general cpu performance was there 6 months ago, they would have released it and continued working on the B core design while making money on Steamroller-A. The fact that they didn't indicates that they are struggling to get the performance improvements they want.

The last time AMD delayed a core design incessantly in the search for performance improvements was Bulldozer. If Steamroller-B releases with 5-15% improvements (depending on application) over Piledriver, and massive improvements for selected HSA demonstration programs, I am going to be really underwhelmed. HSA may have massive potential in certain areas, but it isn't a solution for everything. AMD needs their general cpu performance to go up - whether it be in performance-per-watt to get more competitive in mobile, or just in pure performance. If the best thing about Kaveri is the GCN cores, AMD is not going to be in a good position.
 

afaik, process never changed. it was supposed to be on 28nm high perf. node (or super high perf./hpp whatever glofo calls it.). if non-intel foundries hadn't failed with 32nm, kaveri and sr should have been fabricated on 22nm(hp+/shp) in 2012, launch in q3-q4. but that delay happened before bd came out, so shouldnt have affected much.
hsa was part of bd from the begining. amd started 'fusion' (before they got sued and changed it to hsa) with llano and trinity.

this latest slide gives me this impression:
kaveri has had modifications in the u.n.b. after 2012 unveiling (post-rumored-cancellation), which caused delay. it also has had some laptop friendly tech implemented, similar to intel's. if anything major, i'd speculate that jim keller tweaked the u.n.b. and put in xdma.
the rest is highly dependent on glofo and their track record is bad. problem is, amd has no other foundry to make the high perf. cores. so that will keep being a major decider.
 
@blackkstar:

That has Jim Keller written all over it...which is as much as I suspected when it was delayed. Makes me wonder when we'll get our hands on the white papers for the process tweaks...wondering what the changes were from SRA to SRB.

Additionally, I am not entirely sure they'll skip HEDT for SR, it may be a delayed sort of thing, where perhaps they'll roll out the dCPUs late summer after the APUs hit 1H 2014. Although, I will say this, if Jim Keller has his hands directly on SRB, it most certainly was for the better.
 

They can delay HEDT SR until Q2 2014 and let Vishera end its run and then launch SR FX/Phenom/Whatever on AM3+, perhaps making a 1090FX refresh with something like an A88X southbridge to replace SB950 and in 2015 they will launch EX HEDT and APUs, the HEDT being on FM3 or FM2+, with a slight chance of AM3+ B/C...
 


There were already prototype 1090FX chipset MBs ready before AMD delayed that update. They were just put on hold...however, that would be something they could easily do.

Though, it does make sense that they wouldn't put out an AM4 socket right now.

Also, considering the somewhat murky rivalry between DDR4 and HMC, it may be prudent to wait that out and see what becomes more popular and widespread. DDR4 won't be much improvement in throughput over DDR3 honestly; however, HMC has some potential for massive bandwidth gains.
 


Just about the only thing i agree with so far when it comes to Juanrga no i feel like their position on the market is great.
 
If AMD were to release HEDT SR chips it would put them at a mid-summer release or so given their delays. Meaning AMD would then either have to push EX on HEDT back yet again for another 6 months or they could jump 6 months ahead and release EX earlier than they would have
Click to expand...

I pinned the next iteration of desktop CPU's for late spring to mid summer. APU's and low power parts are AMD's #1 priority as that's where it's biggest market is at. It won't ignore the high performance arena though, it just has it at a lower priority.
 


Interesting that a 30% performance boost in the same power envelope translates to a 10-15% efficiency gain.

as for your other marketing beliefs. sure, AMD will gain ARM server market share. Say at best case ARM gains 20% market share. 5 other vendors, say AMD's share is 25%, that gives AMD a 5% market share, while losing 6% x86 market share for an overall gain of (1%). now say samsung & qualcomm puts out some heavy marketing and gains 50% of the server ARM sales. AMD is only stuck with 15%, so overall their market share is a measly 3%.

So why does fieldman keep saying the same thing? easy. Its called marketing.

He wants you to believe it to be true. If it doesn't pan out, AMD is fked since they stopped developing x86 and have a failed ARM attempt.

sheer genious abandoning x86 isn't it?
.
 
As things stand now, i3 cannot compete with FX6xxx, while FX8xxx equals the i5 and reaches i7 levels for heavy processing tasks.

In the earnings call, rory Read stated:

We need to move stronger into desktop and as we talked about a year ago, we worked on the inventory on the desktop segment and we built and repaired that and we have seen two quarters of consistent revenue growth in that segment and we believe that we have the right product stack to continue to make progress and that’s part of our business as well.
Click to expand...

Which means that AMD remains committed to x86 desktop, but the talk of having the right product stack suggests they don't see a need to introduce new products right now. Perhaps their goal, given that Intel performance is remaining static, is to just produce more PD chips cheaper.
 
Just a simple term mix up efficiency does not immediately just mean power envelope if performance improves at the same amount of power as last gen its efficiency went up.

" achieving maximum productivity with minimum wasted effort or expense."
 


Yeah, PD performance is going to be there for next gen console ports but the die size is still massive compared to what Intel offers.

Intel IB-E 6 core is 217mm^2 (IIRC) and FX CPUs are 315mm^2 (IIRC).

Assuming GloFo 32nm is not ridiculously cheaper than Intel 22nm, AMD is in a lot of hurt on profit margins for FX chips.

It needs to be fixed, badly. It's also stopping AMD from releasing a 3m/6c APU with a decent GPU.

They need to reign in die size for their x86 module architecture. It is out of control.
 


god i hope. i'm so tired of the cpu race stuck at a standstill.
 

Are you kidding, true?


The immense majority of the market wants a APU.



Define high-performance CPU.
 


In my opinion, those would be paired with a high-performance 8-core APU. But before AMD releases one of those, you can use a FX-8000/9000 series CPU. The 9000 series is at the same performance level than eXtreme six-core Intel CPU on BF4, for instance.

Also AMD can sell those cards and obtain revenue pairing them with products of the competence. Yes this is not optimal strategy, but it is happening today. Pay attention to AMD FirePro cards in the HPC/Workstation market. AMD is only selling the cards.

Hint: take a look to last Mac Pro worsktation. It is an AMD Firepro+ Intel Xeon conf.



It must be difficult to find an x86 engineer saying "yes the AMD64 ISA that we designed is a complete mess. Aarch64 is much more modern, elegant, and clean ISA".

It must be also difficult for them to accept that the first 64-bit chip released by ARM engineers already offers the same performance than top x86 design developed and optimized during decades.

Take a look at Intel engineers. They did spend years and billions on developing the more efficient x86 chip that they could do, just to be humiliated a week after by Apple engineers. Ask Intel engineers his opinion about that.

Finally, if I was an AMD top manager and my plans were to migrate slowly from x86 to ARM, the last thing that I would do is reveal my plans to the x86 engineers.

About investors, they aren't idiot. Look what happened at market share of AMD just after the Q3 2013 results were made public. AMD did beat market expectations still its share has dropped by about 18% or so. Why? AMD did better than expected, but investors know that AMD still has one feet in a falling market.

http://www.reuters.com/article/2013/10/17/us-amd-results-idUSBRE99G18Z20131017
 


I think you mean Warsaw. First, AMD servers head, Feldman already explained which is the goal of Warsaw:

He also told us that “Warsaw”, AMD’s follow up in the 2P/4P server market, will cater for a declining but still large market with institutional customers that will take longer to migrate to an ARM ecosystem.
Click to expand...

Second, if you pretend that ARM is low-performance then pay attention to what AMD says:

Sometimes there’s a misconception that ARM means very low power and very low performance,” says Kamal Khouri, director of product management for AMD’s embedded division, ” but this is truly a high performance SOC taking advantage of ARM’s first 64-bit architecture.”
Click to expand...
 


During several posts I explained how sending to the trash those dies could be less expensive for AMD than trying to sell them if the demand is very low.

But again just take a look to Warsaw. It comes only as 12/16 core. This means they are only using 3M/4M. The 2M are being sent to the trash. AMD is not using them for any new server product.

Warsaw replaces the previous Opteron 6300/4300; the Opteron 4000 series were 4 PD core CPUs. I suppose that are eliminating the 4 PD core Opterons due to lack of demand.

They can do something similar in the desktop. Substitute the 4/6/8 core FX desktops by 6/8 core new brand desktop. Of course this is only speculation. Time will say what are AMD real plans.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom