No New FX Processor From AMD in 2013

[btarunr]

AMD's FX "Vishera" socket AM3+ processors are in for a long haul. According to a DonanimHaber report based on a leaked company roadmap slide, the company plans no new processor architecture to succeed it in 2013. The company recently launched its FX "Vishera" line of eight-, six-, and four-core chips just an year following FX "Zambezi," leading analysts to believe the company would launch a new micro-architecture each year to keep up with Intel's "tick-tock" product development strategy.

The roadmap slide, pictured below, shows AMD FX "Vishera" continuing through 2013 as the flagship desktop platform, followed by "Richland" third-generation desktop APU, which combines "Piledriver" CPU components with "Radeon 2.0 cores" (we're guessing those are Graphics CoreNext stream processors), which maintains socket FM2 platform; and low-power "Kabini" APU, which carries the mantle from "Brazos."



View at TechPowerUp Main Site

 

[newtekie1]

Good, I'm all for a little bit of a slow down in new products being released. I mean, I'm all for progress and all, but I'd like to see a little bit more progress between generations instead of just releasing new products with marginal improvements just for the sake of releasing products.

 

[Jstn7477]

I guess we won't be seeing a PCIe 3.0 controller from them anytime soon. I wonder how the new E series APUs will perform, as I did use an E-350 for a while (gave my SFF PC to my dad to use for a home server) and it would have had potential with a stronger CPU component that wasn't reminiscent of a single core K8 Athlon.

 

[NC37]

Not a terrible idea for AMD. Haswell may force a little price drop but Piledriver has so far delivered what Bulldozer lacked. Liking my 8320.

 

[eidairaman1]

I guess we won't be seeing a PCIe 3.0 controller from them anytime soon. I wonder how the new E series APUs will perform, as I did use an E-350 for a while (gave my SFF PC to my dad to use for a home server) and it would have had potential with a stronger CPU component that wasn't reminiscent of a single core K8 Athlon.

probably ran better than the Atom Intel has. I had one and that thing was slower than my 9 year old desktop

 

[Absolution]

Wasnt the successor to trinity supposed to be a Kaveri "Steam-roller" core :/

 

[sergionography]

Wasnt the successor to trinity supposed to be a Kaveri "Steam-roller" core :/



http://lh3.ggpht.com/-YJ1VySqNd-o/UIJkd36f3qI/AAAAAAAAXEo/j3cUVpA8-38/s1600/-1.jpg

they had slides with the steamroller improvements based on simulations, it appeared to be pretty optimistic, but then after the haswell press release they probably realized it will be smarter to make more changes on the core, and probably give themselves more time to iterate the new instruction sets on haswell and release them in steamroller. especialy with the drive for parallelism in haswell with an architecture like steamroller its only better news

 

[eidairaman1]

i like the fact they aint rushing this time around. if they continue figuring out the workings of PD then SR should be a major change in design.

 

[cdawall]

I just want to point out it doesn't say there will be no new FX processors just that they will be Vishera based. We will likely see new revisions of Vishera with improved IMC's and improved power consumption/clockspeed.

 

[TRWOV]

Wasnt the successor to trinity supposed to be a Kaveri "Steam-roller" core :/



http://lh3.ggpht.com/-YJ1VySqNd-o/UIJkd36f3qI/AAAAAAAAXEo/j3cUVpA8-38/s1600/-1.jpg

Yeah, but Steamroller will require a new socket plus I guess AMD will implement DDR4 and PCIe 3.0 support.

 

[cdawall]

Yeah, but Steamroller will require a new socket plus I guess AMD will implement DDR4 and PCIe 3.0 support.

Well it wont require a new socket they already announced that. It will only require a new socket for DDR4 just like you can take Phenom II's and drop them into DDR2 AM2 boards as you very well know as a user of them

 

[Steevo]

Meh. We are about as far along as we can go with this series of chips anyway. They have serious design flaws causing the cache latency issues. A simple respin and tweak helped in both the "needs more gigahurtz" department as well the other bottlenecks, but we didn't see a massive change in the primary issues like latency.

At 4Ghz electricity can only travel 2.9 inches in a perfect vacuum, electrons being pushed and pulled through silicon, copper and gold are much slower, and each change of direction causes heat and reduces the strength of that signal, and each process is only capable of handling so much electricity. As we make the process smaller and small these items become huge obstacles to increasing the performance, both in " jiggawatts" and in "jizzawatts" are you even still reading this or are you checking my math, assholes, anyway the issues seen in Bulldozer and piledriver CPU cores are due to designing them with poor R&D on the physical constrain side of things.

 

[cdawall]

Meh. We are about as far along as we can go with this series of chips anyway. They have serious design flaws causing the cache latency issues. A simple respin and tweak helped in both the "needs more gigahurtz" department as well the other bottlenecks, but we didn't see a massive change in the primary issues like latency.

Says who? the cache is no slower than Phenom II or Intel's offering. It was designed by a computer and not a person yes...but on that note piledriver corrected a large number of those nuances. There is next to no reason why updated steppings wont clock higher Phenom II managed to reach a stock clock that the vast majority of the original chips couldn't even reach overclocked on water.

 

[TRWOV]

Well it wont require a new socket they already announced that. It will only require a new socket for DDR4 just like you can take Phenom II's and drop them into DDR2 AM2 boards as you very well know as a user of them

I could swear that AMD stated Vishera was the last AM3+ CPU but then again AMD has switched its roadmap a lot. Maybe I missed that.

 

[eidairaman1]

Yeah, but Steamroller will require a new socket plus I guess AMD will implement DDR4 and PCIe 3.0 support.

SR might verywell be DDR3/DDR4 meaning AM4+

 

[cdawall]

I could swear that AMD stated Vishera was the last AM3+ CPU but then again AMD has switched its roadmap a lot. Maybe I missed that.

It depended on the roadmap on which stated what. Like you said they like to play musical maps so I guess only time will tell.

 

[eidairaman1]

It depended on the roadmap on which stated what. Like you said they like to play musical maps so I guess only time will tell.

well since its near end of year anyway, its expected to see SR in 2014 then, as a DDR3 Model then eventually a DDR3/DDR4

 

[cdawall]

well since its near end of year anyway, its expected to see SR in 2014 then, as a DDR3 Model then eventually a DDR3/DDR4

Knowing AMD's luck with memory controllers I wouldn't be surprised if we saw a set of SR chips released with disabled bad DDR4 controllers on them. Similar to the Phenom II X4 940.

 

[Steevo]

Says who? the cache is no slower than Phenom II or Intel's offering. It was designed by a computer and not a person yes...but on that note piledriver corrected a large number of those nuances. There is next to no reason why updated steppings wont clock higher Phenom II managed to reach a stock clock that the vast majority of the original chips couldn't even reach overclocked on water.

Have you even read the other parts of any reviews?

http://www.anandtech.com/show/4955/the-bulldozer-review-amd-fx8150-tested/6


10ns is a huge amount of wasted time for L3, 10 more processor cycles wasted than the Intel offering every time the L1, L2 cahces are missed.

http://www.tomshardware.com/reviews/fx-8350-vishera-review,3328-7.html


Looks like shit to me.That was one of the most massive differences in the "Bulldozer" "upgrade", its horrible and cannot be engineered out of the design with a minor respin and tweak. It will require replacement of the caches, which will cause replacement of almost everything on the die.


Transcoding and other very sequential non-branching items do well only due to the higher processor frequency that masks the failure. Its like a shit truck rolling by at 5 MPH or 55MPH, it may stink the same but it does it faster.

 

[eidairaman1]

Knowing AMD's luck with memory controllers I wouldn't be surprised if we saw a set of SR chips released with disabled bad DDR4 controllers on them. Similar to the Phenom II X4 940.

AM3+ Supported, AM4+ Supported

 

[cdawall]

Have you even read the other parts of any reviews?

http://www.anandtech.com/show/4955/the-bulldozer-review-amd-fx8150-tested/6


10ns is a huge amount of wasted time for L3, 10 more processor cycles wasted than the Intel offering every time the L1, L2 cahces are missed.

http://www.tomshardware.com/reviews/fx-8350-vishera-review,3328-7.html


Looks like shit to me.That was one of the most massive differences in the "Bulldozer" "upgrade", its horrible and cannot be engineered out of the design with a minor respin and tweak. It will require replacement of the caches, which will cause replacement of almost everything on the die.


Transcoding and other very sequential non-branching items do well only due to the higher processor frequency that masks the failure. Its like a shit truck rolling by at 5 MPH or 55MPH, it may stink the same but it does it faster.

Add more speed 99% of problems go away. Cache latency also drops. No it is not even faster than Phenom II, but those performed fine and no one blames that for Phenom's misgivings. L3 cache on AMD has been that same latency and speed since Phenom was dropped many moons ago.

 

[devguy]

Read this. Now continue speculating.

 

[NC37]

Read this. Now continue speculating.

Fusing ARM cores into it...already?! Man I thought it would take them longer to start that stuff.

 

[cdawall]

Fusing ARM cores into it...already?! Man I thought it would take them longer to start that stuff.

Hell why not show off the modular architecture I guess? Would be a killer tablet PC core. Shut down all of the cores until only the highly efficient ARM cores are running for OS support and kick the x86-64 cores on to do anything real. Standard battery life would kill.

 

[Steevo]

Add more speed 99% of problems go away. Cache latency also drops. No it is not even faster than Phenom II, but those performed fine and no one blames that for Phenom's misgivings. L3 cache on AMD has been that same latency and speed since Phenom was dropped many moons ago.

You are missing the point.

A CPU core is just that, a X86 procesing unit, there are only so few ways to make it work.

The latency is a huge issue every time their inferior banch prediction fails. 5% worse hit rate added to a 10 cycle penalty due to the latency if it is even in L3 and you have your 15-20% slower per clock than Intel.

Bend of the knee my friend, AMD is there and it isn't a good place to be.