Review Index:
Feedback

AMD, Vishera, and Beyond: New Design Philosophy Dictates a Faster Pace

Author: Josh Walrath
Subject: Editorial
Manufacturer: AMD

How Does Vishera Stack Up?

So what does this mean for Vishera?  Vishera is the codename for the desktop 4 module/8 core Piledriver based product for the desktop market.  It will be offered on the server market as Abu Dhabi/Seoul/Dehli, but for the sake of this article these chips are all based on the same designs as that which powers Vishera.  Looking at Trinity, one would expect to see the same type of performance and power improvements applied to Vishera.  This basically means that Vishera will be 8% to 10% faster per clock, and be higher clocked at stock while retaining the same TDP.  That is essentially correct, but it misses a major factor;  Vishera will be a Piledriver+ design.

View Full Size

The Piledriver module is very similar to the original Bulldozer, but the sum of all the small changes has made it much more competitive. (image source: VR-Zone)

AMD is changing their design philosophy, and Vishera will truly show if they are on the right track or not.  The Piledriver cores in Vishera should not only retain the improved power characteristics that were displayed in Trinity, but it should also improve per-clock performance as well as threading efficiency.  Details are scarce on what exactly AMD has done to Vishera, but there are plenty of details on what was achieved with Trinity.  L2 and L3 cache sizes have not changed as compared to the original Bulldozer, but I would expect a lot of work to be done on the L1 data and instruction caches, the four wide decoders should get a significant boost, and other smaller structures will either be expanded (like the TLBs) or simply improved (pre-fetch and prediction).

My best guess is that the upcoming FX-8350 will be a 3.8 GHz part, but Fudo over at Fudzilla claims to have information that it will in fact be a 4.0 GHz part with a 4.2 GHz Turbo.  I have heard nothing credible from either mine or his stance, but we can safely bet that it will be one or the other of those numbers.  Either way, it will be a significant improvement over the current FX-8150 at 3.6 GHz not just in raw clock speed, but also the much more powerful Piledriver+ modules that will be used in these products.

The next bit of mixed up news is that some are claiming that Vishera is still on track for a mid-Q3 launch, while others say that it will be a Q4 product.  Looking back at how AMD has handled previous launches, I am fairly confident that we will not see a mid-Q3 window.  AMD typically informs partners and industry analysts of their plans about a month to a month and a half of potential launches.  Nobody has heard anything so far about a Vishera launch.  Motherboard partners have been primarily focused on Trinity for laptops and an eventual desktop launch, and while they have Vishera samples in their offices, they are not aggressively optimizing their AM3+ products as of yet.  This could change quickly, but there is a lot of groundwork still to be done to prepare the ecosystem for Vishera.

View Full Size

A list of the changes put to Piledriver.  Expect this list to grow when Vishera hits. (image source: VR-Zone)

My gut feeling here is that AMD will do a press launch of this processor in mid-September, with general availability in an early October timeframe.  This fits in with some recent speculation that it will in fact be Q4 before these products are available.  By the end of July we will certainly know the exact timeframe in which to expect Vishera to hit the market.  But for the time being AMD is really focused on getting Trinity well established in the very important, and still growing, notebook market.  AMD also looks to have pushed back Trinity for desktop until at least late Q3.  Some other outlets have hinted at these products being available in a general Q4 timeframe.

This then begs to question what of Kaveri?  This is the update to Trinity that features the again updated Steamroller architecture with the graphics portion being the first integrated GCN part.  These processors are scheduled to be produced by GLOBALFOUNDRIES on their 28 nm HKMG process, but this is certainly not set in stone.  Earlier this year AMD was able to extract itself out of an exclusivity clause with GLOBALFOUNDRIES which would have forced AMD to only use GF for 28 nm CPU production.  We still do not know the current health of GF’s 28 nm HKMG line, but current indications call the quality of that product into question.  Indications point to AMD redesigning Kaveri to be produced on another foundry’s 28 nm line, and that has delayed further the introduction of this next generation part.  This speculation fits in nicely with the delay of Trinity for desktop, as we would not expect to see Trinity only be offered fully for one quarter only to be replaced by Kaveri in Q1, 2013.

AMD is staying busy and nimble, but they are at the tender mercies of the foundries at this time.  Vishera does look to be a good upgrade from the original Bulldozer, and it will improve its standing among manufacturers and consumers alike.  While it will not match Intel’s Ivy Bridge in pure performance, it does look to close the gap and be a true successor to the popular Phenom II line from AMD.  This may not drive enthusiasts back into AMD’s corner, but it will certainly stop a lot of the negative press and the bleeding of die-hard AMD fans into the Intel camp.  Vishera does look to be the last desktop based AM3+ part.  Past Q3 of 2013, I believe that AMD will focus primarily on the APU for all desktop applications, and start pushing those products into the server market.  AMD will continue to produce non-integrated parts for the server market through 2014, but I do believe that they will push the APU into the high end as well.  This makes sense as the ecosystem for heterogeneous computing is growing and should be much more mature during that timeframe.  Now we only need to see if AMD can keep up with Intel with their new design philosophy, and their reliance on 3rd party foundries to actually produce these designs.

 

July 8, 2012 | 04:08 PM - Posted by andy (not verified)

I remember more then a few years ago, AMD would bring out a new uArch or a major uArch improvment..I'm thinking original Athlon - AthlonXP - Athlon64 - Phenom etc, which all had increase IPC over the previous uArch, but would always struggle to exceed or sometimes even match the clock speed of the previous generation uArch, then slowly over the months AMD's proccess tech would mature...sometimes with a little help from IBM, and clock speeds would slowly rise.
I wonder if now AMD are fab less if they have specifically gone down the route of designing for increased clock speed from the start, simply because they are now in a position where they have far less control over the fabing proccess.
AMD's must have realised that if they were behind Intel in process tech when they had their own fabs they were going to be even further behind when depending on the fabs of third parties.
It seems reasonable that being fabless might require a very diffrent stratagy then they had employed before, when it would have been far eaiser to test, tweak and fine tune the manufacturing process, and when transitioning to a new node.
The introduction of a more modular design with shared components would seem like an attempt to mitigate Intels proccess advantages.
Do you think it's reasonable to assume that some of the design decisions AMD have made with their latest designs are to degree influenced by their lack on fabs?

July 9, 2012 | 08:08 AM - Posted by Josh Walrath

Hard to say.  Even though they are working with 3rd party fabs, there are still a lot of on-site engineers both with AMD and with whatever fab they are working with.  It would be easy to think that AMD makes the design, works out the bugs, then sends the final masks and specifications to the foundry to get produced.  The reality is that the final design up to producing the masks is really only 3/4 of the process.  The amount of metal layer changes, tweaking to the process steps, and a whole lot of troubleshooting by engineers in the fab to get reliable yields and bins is a significant amount of work.  So, AMD might be designing a bit of leeway into their processors to make up for this (to a point), but I don't think it is as extreme as one might think.

By using the half-nodes from TSMC/GF/UMC, AMD is able to mitigate some of the advantages that Intel has with their full 22 nm tri-gate process.  We also could have a pretty hefty discussion about how CPUs have changed due to these guys putting 4 or more cores on a single piece of silicon.  How much of that space are you going to use to get another 5% improvement in IPC rather than throw on a couple more cores with lower IPC?  Isn't that space better utilized by a more powerful graphics unit with all that floating point power?  These are obviously questions that are answered quite differently by both Intel and AMD.  It is certainly fascinating to see where this all will shake out over the next two years.

July 11, 2012 | 03:46 PM - Posted by Brett Hood (not verified)

Nice write-up Josh. It's a pity that AMD didn't fix the issues with the initial bulldozer chips and release a version 2 of the bulldozer chips across their whole range where we could have seen a boost in stability and performance which may well have more ably competed better with Intel's 2nd gen Core-i5 and i7 cpu's.

July 12, 2012 | 09:03 AM - Posted by Josh Walrath

Yeah, that would have been nice, but if you think about it AMD has essentially put out two revisions of Bulldozer in one year.  The first iteration was only released last October, we have seen Trinity hit the scene, and now we only wait for Vishera.  I think considering the complexity of modern CPUs, this kind of turnaround is pretty quick.

July 12, 2012 | 11:07 AM - Posted by Anonymous (not verified)

Without the igp portion of trinity ,and pushed to 125 watts
I would say Vishera 8 core will do quite nicely.
However I don't expect AMD to continue the FX series
I suspect by the time excavator comes AMD will begin making
2 4 6 8 10 core apu models with true discrete hd 8770 level graphics with ddr 4 and a myriad of improvements.
Once the apu is brought to 10 cores The FX models will be obsolete.

July 13, 2012 | 07:51 AM - Posted by Josh Walrath

Heck, once APUs get to 4 modules/8 cores, AM3+ will be dead.  But I wouldn't doubt if they branded an APU an FX processor that is totally unlocked.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.