AOA Forums

AOA Forums (
-   AMD Motherboards & CPUs (
-   -   Interesting article for AMD fans! (

TrevorS 12th October, 2013 05:48 AM

Interesting article for AMD fans!
Article deals with the future of AMD FX processors and the hUMA initiative. Unfortunately, it doesn't look good for Steamroller availability for my AM3+ boards, but I'll cross my fingers. To be seen to what degree hUMA succeeds with Kaveri and whether it leverages the HD7000 series graphics architecture.

AMD's Processor Shift: The Future Really is Fusion | PC Perspective

danrok 22nd October, 2013 03:23 PM

Seems Steamroller will be Socket FM2.

TrevorS 22nd October, 2013 11:14 PM

Seems clear at least its first incarnation will be FM2 (Kavari), and looks like Vishera will be holding the FX fort through 2014, but that doesn't necessarily rule out an AM3+ Steamroller later. There's supposed to be a road map update sometime in November and that may shed further light. I read a post on another forum by a reported AMD engineer who seemed convinced they weren't yet done with AM3+ (including the 5GHz octal), and AMD did indicate awhile back that Steamroller would be the last AM3+ release (no Excavator).

Not that I've a problem with FM2, I like the HSA concept and hope it'll really make a difference for tasks like video crunching, but I'd like to have the flexibility of one more generation processor for my AM3+ boards (hopefully lower TDP and higher throughput). I won't be buying any more Vishera, got a quad and a hex (only the quad actually built), but would definitely check out Steamroller should it become available.


Originally Posted by AMD: Recent News
New 3D GPU Documentation
Check out these new programming guides and register specifications for the 3D engine here. These documents bring AMD’s public 3D documentation up to date with all currently released asics. Developers on both Hacker News and‎ discussed this documentation & other related topics.

HSA Platform Coherency and SoC Verification Challenges
AMD's Pankaj Singh will be presenting "HSA Platform Coherency and SoC Verification Challenges" at this year's SC13 conference, which is being held Nov. 17-21 in Denver, CO. The presentation will discuss how the support for coherent shared memory across CPU-GPUs has been implemented, new design features in the first implementation of a critical feature of HSA and how verification was achieved for this complex design.

“AMD Aims To Give OpenGL A Big Boost, ‘API Won’t Be The Bottleneck’”
DSOGaming captures an interesting Twitter conversation with AMD’s Graham Sellers and other developers about OpenGL extensions in their article “AMD Aims To Give OpenGL a Big Boost, ‘API Won’t Be The Bottleneck’”. Read John Papadopoulos’ coverage and the corresponding discussion on‎.

International Workshop on OpenCL sets date for 2014 Conference in England
The International Workshop on OpenCL (IWOCL) is an annual meeting of OpenCL™ users, researchers, developers and suppliers. Their second annual conference is scheduled for May 12-13, 2014 in Bristol, England.Technical sessions reflecting OpenCL on PC, HPC, embedded, mobile and web-based platforms are expected. For more information, visit the IWOCL site.

TrevorS 23rd November, 2013 09:17 AM

Appears the SC13 roadmap shows no sign of an FX Steamroller in 2114, but it doesn't indicate FX termination either. Guess Vishera will be the FX standard bearer for a while yet. On the other hand, appears Kavari will be showing up by February. A number of FM2+ boards are already on the market and more appear to be coming.

TrevorS 20th December, 2013 10:39 PM

Appears another FM2+ APU Carrizo (Excavator) is slated to follow Kaveri (SteamrollerB) come 2015. Although no sign of another FX processor before 2016, it remains on the roster so there's no reason to rule it out. Last released Vishera FX were 5GHz and 4.6 GHz.

Leaked AMD Roadmap Q4 2013 reveals more APUs Kaveri and Carrizo with newer cores but no new discrete processor until 2015 planned OCWorkbench

I recently built a A6-5400K FM2+ micro PC and it's a dandy processor. Win7 "Experience Index" of 6.3 processor and 6.6 graphics. For comparison, the FX4300 in my server scores 7.3.

Pandaz3 24th December, 2013 03:27 AM

I am not going the APU route for now, the family's current operational fleet are a 4170, 4350, three 8350's, a 9370 and a 9590. I am currently trouleshooting the 9590 set up (New Upgade and I suspect a bad PSU), but that is 7 FX CPU's and I am happy with them. I have a APU in a laptop (MSI GX 70) and it does it's job there

TrevorS 25th December, 2013 10:52 PM

Yikes, sounds like you're loaded for bear! I've been an AMD fan since my first K6 and still have a running Athlon XP mobile (daily PC) and three running 939 Athlon 64 X2 machines (one my AGP gamer), as well as older AT hardware, some built, some not. My first recent machine is the FX4300 in my server and latest is the A6-5400K HPTC. (Got parts to build an FX6300 gamer last summer, but no games that need it :p)

I'm very interested in HSA and hUMA and am hoping to see that initiative translate into very powerful crunchers. Hopefully, some of my recent hardware will prove leverageable. People tend to reflexively view APU graphics capability purely as something to hold the fort before getting a good video card -- basically for a starter or office PC. That's not the design concept at all. The intent is to closely bind the parallel processing power of a GPU, with the more serial general purpose processing power of a CPU, eliminating as much latency as possible through merging their memory space. This extends to compatible video cards as well. I think that's pretty exciting and while Trinity was a step back from Llano in HSA resources (fewer shaders) Kaveri is expected to really up the game via GCN graphics, SteamrollerB compute modules, and more powerful memory sharing.

Also, note that installing a video card similar to the APU graphics makes a big difference in gaming performance (Dual Graphics).,3224-4.html

This link demonstrates the effect of utilizing graphics compute ability to improve non-graphics workload performance, particularly with quad-core APU's. (Note: OpenCL also works for Intel processors and n'Videa graphics -- not an AMD exclusive.),3224-19.html

TrevorS 26th December, 2013 05:47 AM

An interesting article from a little over a year ago -- can't speak on the veracity of the insider input, but reads believable. Taking on Intel is tough business and it's very clear AMD screwed up mightily following the success of the Athlon 64 X2. The original CEO is gone, but he really shafted the company with his egotistical incompetence. They're still fighting to climb out of the hole he dug.

Go Steamroller! :)

AMD Pushes Steamroller and Excavator Forward, Bullish about Performance Increases

TrevorS 26th January, 2014 07:46 AM

The first Kaveri APUs were released Jan 14 and there are a number of reviews out. I found this one to be especially informative, both in the introductory discussion and the well considered Intel comparisons.

AMD A10-7850K And A8-7600 Kaveri Review - APU - Tom's Hardware

The performance findings are reasonably consistent, though it's clear the Intel i7 processors do not belong in the comparisons (except maybe some situations with 'Mantle' substituted for MS DirectX, but that requires cooperation from the software provider as in the case of "Battle Field 4").

Appears AMD is positioning Kaveri primarily as an entry to mid-level gaming processor. At least three reasons for that are the strength of the integrated graphics, the paucity of available HSA/hUMA applications, and the fact that in moving from Global Foundries SOI 32nm to SHP 28nm processes, it was necessary to compromise between iGPU density and CPU clock speed. Unfortunately, the consequent clock reduction partially offsets the more efficient Steamroller architecture, though integer performance is less impacted due to no longer sharing one instruction decoder between two cores.

A current APU iGPU limitation is being limited to DDR3 Vs GDDR5 memory speeds. This could be ameliorated by including a fast L3 cache (currently there's none), but that will probably require moving to a 22nm process (the 28nm Kaveri chip is already larger than Richmond in die and especially transistor count). A major Kaveri advantage is its reduced TDP, however, there are indications of throttling when both iGPU and CPU are loaded -- another argument for a 22nm process.

So, it's clear there's more work to be done to achieve an optimal combination of process and architecture, but all-in-all, seems like AMD accomplished a lot with this double iteration, both adapting a new process and implementing a genuine HSA architecture -- a large bite! Meanwhile, Kaveri really does appear a respectable low to mid gamer and is perfectly capable of handling typical home and office chores with aplomb! The A8-7600 is expected to release later this quarter.

Pandaz3 26th January, 2014 08:53 AM

Well it is looking more promising, but I still like the traditional setup with more speed. I have been too busy with Family issues to continue my troubleshooting of my 9590 machine, maybe tomorrow.

TrevorS 26th January, 2014 10:41 AM

The A8-7600 is appearing the gem in this release, in most cases its performance is not too far removed from the A10s and its ability to run at either 45 or 65W TDP could be useful for HTPC and other quiet or small form factor applications. To whatever degree thermal throttling is an issue, moving from the factory HSF to more capable coolers would have to make a substantial difference.

Just as Richland is basically a tweaked Trinity, I could easily see a tweaked Kaveri appearing before Carrizo brings Excavator -- the engineering staff probably has a growing list of improvements that could be made. At worst, many of those adjustments will likely be rolled into Carrizo, but if Carrizo happens to be looking late 2015, then a second Kaveri pass could make sense. Although I doubt they can add an L3 cache without further process reduction, anything else that would reduce memory latency could be helpful. I expect all the glue that enables HSA/hUMA to function is under serious scrutiny. Together with ongoing evaluation of SHP process compromises and possibilities -- it takes experience to make the best decisions and they've now an iteration under their belts.

PS. Ran across this article relative to AMD and GlobalFoundries 20nm and 14nm processes, but it's not clear which processors will be targeted.

Also ran across this article regarding AMD plans regarding the FX family.

TrevorS 28th January, 2014 12:15 PM

Here's a great article on Kaveri, as in 'meet Kaveri'.

Here's an article on "Mantle", the soon to be released AMD high-performance real-time drawing API for discrete and integrated GCN graphics (a more cpu efficient alternative to Direct3D and OpenGL.) It's expected to provide a considerable FPS boost to supporting games, "Battlefield 4" for example (for a limited time, being given away to early Kaveri A10 purchasers.)

Delving deeper into AMD's Mantle API - The Tech Report - Page 1


TrevorS 27th April, 2014 03:25 AM

Been watching for the Kaveri A8-7600 and still no sign of it. My guess is AMD is out to achieve max sales from the A10 versions before opening the A8 door. Or perhaps, a 2nd generation tweaked production Kaveri will be introduced and perhaps that will include the A8? Watching continues :)!

cloasters 29th April, 2014 10:49 PM

Thank you for the update, Trevor.

TrevorS 30th April, 2014 06:34 AM

Stumbled across an interesting read on the new AMD API "Mantle."

AMD’s Mantle benchmarked: The biggest innovation in gaming since DirectX 9 | ExtremeTech

PS. Hiya George :)! Hope all is well!

TrevorS 14th May, 2014 12:11 AM

AMD had a press conference May 5 talking about their ARM and x86 plans for the non-desktop market. Some interesting stuff including enhanced Puma cores and 20nm APU's with continued development/expansion of their hUMA/HSA topology. Too bad it doesn't broach the question of further FX development, but still, the slides indicate it continuing.

AnandTech Portal | AMD Announces Project SkyBridge: Pin-Compatible ARM and x86 SoCs in 2015, Android Support

TrevorS 28th May, 2014 11:25 PM

Ambidextrous apparently includes a next gen AMD x86 architecture with possibility of an all new FX release -- fingers crossed!

AMD Confirms Development of High-Performance x86 Core With Completely New Architecture

(AnandTech Portal | Apple A4/A5 Designer & K8 Lead Architect, Jim Keller, Returns to AMD)

cloasters 29th May, 2014 11:34 PM

A fair business environment( I know, that'll be the day) would take twenty five billion from 666, give it to AMD and pray like hell. Plus 666 would shut down all research, development and selling activity for two years. Sigh.

TrevorS 30th May, 2014 04:00 AM

As an AMD fan since the K6 alternative to Pentium, I found the Athlon and Athlon X2 days almost exhilarating. The Pentium 4 was hurting and AMD broke through with K8, not only the best desktop gamer of the time, but also at least competitive in other applications. Downside being the price climbed accordingly -- my X2 3600+ was a huge extravagance and yet still a step down in the Socket 939 lineup.

It would be great if the under development x86 architecture turns out fully competitive with i7 (single & multiple thread plus thermal), but I don't think that's necessary for a worthwhile FX upgrade. I'm guessing the single most important attribute is that it be scalable for both Puma+ SOC/APU and FX (plus, of course, pin compatibility with its sister "K12" ARM processor). Seems Keller's instruction to the team is to "take the best" from Bobcat, Puma, and SteamRoller/Excavator and expectation is a move from CMT ...

AMD Bulldozer architecture (page 1: Introduction) - BeHardware
The Piledriver Architecture: Improving On Bulldozer - AMD FX-8350 Review: Does Piledriver Fix Bulldozer's Flaws?
AnandTech Portal | AMD's Steamroller Detailed: 3rd Generation Bulldozer Core
AMD Excavator Core May Bring Dramatic Performance Increases - X-bit labs

... to more traditional SMT.
Simultaneous multithreading - Wikipedia, the free encyclopedia

Regarding CMT, the primary current problems (Kaveri) seem to be FPU inefficiency (hurts throughput), transistor leakage (hurts power consumption), Bulk Silicon Vs SHP (reduced clock speed), and first iteration (learning curve). A significantly more capable FPU with 20um and FIN-FET technology could make a substantial difference. From what I've read, AMD CMT accepts increasing multitasking load more gracefully than Intel SMT -- better preserves responsiveness. Would be interesting to further develop CMT, though limited resources may prevent that, and so to whatever degree it survives will depend on Keller and his team's evaluation. Interesting times!

(AMD Announces Ambidextrous Computing Roadmap | techPowerUp
AMD to Introduce New High-Performance Micro-Architecture in 2015 – Report - X-bit labs)

dabaerman 9th June, 2014 12:27 AM

WOW, there's some good reading material. I have to thank you for the links.

I too have been an AMD fan for a while (pre skt 370). have 955 on 990FXA and another on the bench. :banghead: want 8 core cpu, but waiting for a good reason to upgrade.


All times are GMT +1. The time now is 12:45 AM.

Copyright ©2001 - 2010, AOA Forums

Search Engine Friendly URLs by vBSEO 3.3.0