Review Index:
Feedback

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

Author:
Subject: Editorial
Manufacturer: AMD

Less Risk, Faster Product Development and Introduction

There have been quite a few articles lately about the upcoming Bulldozer refresh from AMD, but a lot of the information that they have posted is not new.  I have put together a few things that seem to have escaped a lot of these articles, and shine a light on what I consider the most important aspects of these upcoming releases.  The positive thing that most of these articles have achieved is increasing interest in AMD’s upcoming products, and what they might do for that company and the industry in general.

View Full Size

The original FX-8150 hopefully will only be a slightly embarrasing memory for AMD come Q3/Q4 of this year.

The current Bulldozer architecture that powers the AMD FX series of processors is not exactly an optimal solution.  It works, and seems to do fine, but it does not surpass the performance of the previous generation Phenom II X6 series of chips in any meaningful way.  Let us not mention how it compares to Intel’s Sandy Bridge and Ivy Bridge products.  It is not that the design is inherently flawed or bad, but rather that it was a unique avenue of thought that was not completely optimized.  The train of thought is that AMD seems to have given up on the high single threaded performance that Intel has excelled at for some time.  Instead they are going for good single threaded performance, and outstanding multi-threaded performance.  To achieve this they had to rethink how to essentially make the processor as wide as possible, keep the die size and TDP down to reasonable sizes, and still achieve a decent amount of performance in single threaded applications.

Bulldozer was meant to address this idea, and its success is debatable.  The processor works, it shows up as an eight logical core processor, and it seems to scale well with multi-threading.  The problem, as stated before, is that it does not perform like a next generation part.  In fact, it is often compared to Intel’s Prescott, which was a larger chip on a smaller process than the previous Northwood processor, but did not outperform the earlier part in any meaningful way (except in heat production).  The difference between Intel and AMD in this aspect is that as compared to Prescott, Bulldozer as an entirely new architecture as compared to the Prescott/Northwood lineage.  AMD has radically changed the way it designs processors.  Taking some lessons from the graphics arm of the company and their successful Radeon brand, AMD is applying that train of thought to processors.

Continue reading our thoughts on AMD, Vishera, and Beyond!!

View Full Size

The original Bulldozer die was pretty fascinating for the time, but that was before people knew that it was a bit of a dud.

So what exactly does this mean?  For the past several decades we saw processor design follow a fairly simple routine.  A new generation of architecture is released, and there are a few minor updates to the architecture other than moving to smaller process nodes.  Around the seven year mark a brand new architecture is then introduced, updates are applied, and then the cycle starts over again.  These massive jumps in technology are complex and expensive, and they take tens of thousands of man hours to complete.  While the advantages of going with a clean sheet design are many, if a basic decision is made that turns out to be flawed or counterproductive, then years of design work are wasted.  AMD appears to be trying to move away from this design paradigm, as the risks of making such a poor decision is nearly catastrophic for the company.  Instead, AMD is looking at a more conservative, though accelerated, route.  Essentially AMD is looking at major, yearly updates for their processor architectures.  Instead of small updates over the years culminating in a massive redesign, they are taking many smaller steps in between.  Much like AMD did with their process technology, they are applying this to the design methodology.  This means small, meaningful changes on a very regular basis.  This philosophy allowed AMD to stay within spitting distance of Intel and their formidable process lead, and their products (historically) have been in the same rough performance range with similar TDPs.

We have now seen the first fruits of this labor with the Trinity APU that was recently released.  Trinity fixes a lot of the issues that plagued Bulldozer, and combined it with a solid integrated graphics part based on the HD 6900 series VLIW4 architecture.  The first major issue that was solved was that of power.  Bulldozer was really rushed to market, and as such it was not completely optimized for power consumption.  While the top end part was rated at 125 watts, it fell well below where it was initially expected to be in terms of clockspeed.  When the FX-8150 was taken above 3.8 GHz, we saw an impressive rise in power consumption.    In my FX-6200 review, when that particular 3.8 GHz/125 watt TDP part was taken to 4.8 GHZ, I saw a 100 watt increase at the wall.  Needless to say, the design was a bit “loose” when it comes to power consumption.  With Trinity AMD was able to take the time and truly optimize the design for power consumption. 

View Full Size

Trinity, on the other hand, proved to be a very adequate CPU in a variety of performance scenarios.

Trinity also provided per-clock performance advances.  These advances do not take the Piledriver architecture up to Ivy Bridge performance heights, but it is still a good 8% to 10% faster per thread/per clock than the previous Bulldozer.  And it is able to achieve that at lower TDPs.  Pretty exciting stuff, right?  What is perhaps even more exciting is that Trinity was shown off in working condition a year ago, and was apparently quite production ready in late 2011.  AMD held off Trinity production due to multiple factors, but primarily because they had finally fixed Llano production issues and that particular chip was selling very well in both desktops and notebooks.  It also made little sense to introduce a new socket at that time, as FM1 was only then just gaining steam.

July 5, 2012 | 01:52 PM - Posted by Chaitanya Shukla (not verified)

I hope AMD updates the aging chipset line-up that they have for their current AM3+ platform.

July 6, 2012 | 09:43 AM - Posted by Josh Walrath

I haven't heard anything about the 1000 series chipsets since last year.  They really could use PCI-E 3.0 and an updated/improved SB1000 series southbridge with better SATA 6G support and native USB 3.0.

July 5, 2012 | 02:23 PM - Posted by Prodeous (not verified)

There was one comparison from AMD's side that was missed here.

Phenom I wasn't that impressive and its performance over previous generation was low, and was extremely power inefficient. When overclocked they drank power by the bucket. When Phenom II came out it addressed most of the issues.

I expect this to be the same situation. Bulldozer vs Piledriver.

As for the article it self. Impressive.

Early Trinity vs Bulldozer show very nice 15% range improvements. And that is Bulldozer with L3 vs Trinity without :)
http://www.tomshardware.com/reviews/a10-5800k-a8-5600k-a6-5400k,3224-2.html

July 5, 2012 | 02:44 PM - Posted by 240mike (not verified)

I'm really holding out on this CPU. I hope it is powerfull enough for some good gaming and light editing. I have a perfect subsystem for it. Dual 6870's and gigabyte board and good PSU. I'ts just my 955 BE thats holding me back.

July 5, 2012 | 07:35 PM - Posted by Humanitarian

Same situation, I feel ya.

Games are starting to hover around 90% on my 960t, there's not a lot from AMD to upgrade to unfortunately. Hoping this will have the performance gains needed for a little bit of screen capturing/encoding while I'm gaming.

July 6, 2012 | 09:43 AM - Posted by Josh Walrath

Have you unlocked those other cores on the 960T?

July 5, 2012 | 03:49 PM - Posted by mAxius

Nice work Josh :D

Since the road-map for 28nm cpu for AM3+ is essentially dead.

What will replace the 8 core FX line?

Will AMD make an 8 core FX Steamroller Apu?

July 5, 2012 | 04:10 PM - Posted by franzius

Piledriver based desktop systems built by HP are already on sale at Newegg. Other OEMs should not be too far behind with their own offerings.

http://www.newegg.com/Product/Product.aspx?Item=N82E16883157362

http://www.newegg.com/Product/Product.aspx?Item=N82E16883157361

July 5, 2012 | 04:20 PM - Posted by Josh Walrath

I somewhat misspoke.  It looks as if retail availability of DIY style and small OEM style processors in a box offerings will not be available until Q4, and the question is if that will even happen with Kaveri not much farther away than that.  Also, Trinity was pushed back in this area due to a large volume of product based on llano in the channel.  Big guys like HP will have products based on Trinity, and as mentioned, they are available right now.

July 6, 2012 | 01:18 AM - Posted by IronMikeAce

Great article Josh. It's sad to hear that desktop Trinity is being delayed which in turn pushes back Kaveri. I plan to upgrade my A8-3870K APU to Kaveris highest end desktop chip. Looking forward to seeing what a 3rd generation Bulldozer chip is capable of as well as the integrated GCN architecture and not to mention the full access to system memory for both processors. I think AMD is definitely onto something with their HSA and I am very excited to see the potential of mutliple processors completed integrated together. Josh, do you think AMD will decide to integrate an ARM chip onto Kaveri or is that not planned until a lot later? I would really like to hear your thoughts on what AMD plans to do with their HSA concept in the coming years. Again, thanks for the article.

July 6, 2012 | 09:26 AM - Posted by Josh Walrath

Yeah, I think that AMD already talked about integrating that ARM processor into Kaveri, but it will only be for that trusted computing/security functionality... that we know of, so far.

July 6, 2012 | 01:22 AM - Posted by IronMikeAce

AMD 8-core CPUs are definitely an upgrade over any Phenom II x2 or x4 chips. I upgraded to 8120 at 4.8GHz from my 965 x4 at 4.2GHz and it was a huge upgrade. Much more snappier and completely awesome at multitasking. Also, the 8cores are awesome for BF3 gameplay. Smooth as butter. By now though you should wait until Vishera but the 8core Bulldozers are definitely an upgrade over the x2 and x4 Phenom II chips. No question about it.

July 6, 2012 | 09:27 AM - Posted by Josh Walrath

I my testing of the FX-6200, I also did finally see that chip outperforming the Phenom IIs... but only once it got over 4.5 GHz.  It does pull a whole lot more power at that speed though.

July 6, 2012 | 03:57 AM - Posted by Anonymous (not verified)

Using Phenom II 1090t with 16gb

will definitely buy steamroller with 16 or 32gb ddr4 and gddr6 GPU's and 480GB ssd.........."might" pass off piledriver. Also 3rd gen APU ultrathins is on me mind.

Wish AMD the best of luck..............i really hope they dont go KAPUT.

July 6, 2012 | 09:31 AM - Posted by Josh Walrath

I believe that AMD will be abandoning AM3+, and the Piledriver cores will be the last chips that fit in that particular socket.  AMD does look to go the route of all APU SKUs for desktops once AM3+ is totally phased out.  What I am curious about is how they will handle more modules + L3 caches with a GCN module attached at 28 nm.  Such a product would seem to be in the 360 mm squared range of die sizes, which I guess is still comparable to the Phenom II X6 series of chips.  Not so nice as compared to Intel's 22 nm Ivy Bridge which is about half that size...

July 6, 2012 | 04:42 AM - Posted by biblicabeebli

I've always thought that the bulldozer chips were marketed by a squirrel. And the Squirrel was on fire. And the fire was started with a bulldozer CPU.
(now that I have your attention…)

AMD must live with a generation gap compared to Intel. Intel has smaller process node, and then they snuck in tri-gate magic on top of it. Intel is at 22nm, AMD is stuck at 32nm... so how the f**k does AMD have 8 core consumer class chips, while Intel limits their consumer class chips to either 4 or 6 cores? (Server side Intel has 8 and AMD claims 16.) Read on.

Intel bumps up the core count of their chips with Hyperthreading, a method of simultaneous multithreading (okay, that exhausts my vocabulary) -but- everybody knows that hyperthreaded cores ≠ real cores.
AMD has this shared FPU design, a new and exciting, and apparently better, hack at getting a higher core count. They can have this clever modular design thing going, they can save a quantity of die space, and while some people on comment threads complain about floating point metrics - here's the point - nobody cares. So we let AMD sell us "8 core" chips without much of a fuss. They just do a really good job at pretending to have this many cores

---
AMD did not have to sell it like that.
---

AMD COULD have sold us "4 core" chips with a super power that lets it process 8 threads PERFECTLY. Way better than Intel's silly hyperthreading, double execution threads my arse, ours are Much better ... (as long as you don't do scientific computing workloads, those jerks can go f&#k themselves).
The masses would have applauded such humble accolades, and from a company that has to go up against big evil corporate Intel, too. Much pity. Oh and these chips work in your existing processor sockets thankyouverymuch. Oh and they cost a lot less, that too. Did we mention they are ALL multiplier unlocked?
---
But. They. Didn't.

(Okay, that ramble got away from me. Woo forumz.)

July 6, 2012 | 09:40 AM - Posted by Josh Walrath

LOL, heck of a train of thought there.  I think that biggest problem that AMD has is that the chips just didn't come out of the oven very well for the first gen of Bulldozer, and instead of spending a lot of money on marketing/advertising, only to receive a huge negative backlash about how they actually compare to recent Intel products, they saved up what good will they have built up with consumers and enthusiasts.  I think we will see more aggressive marketing with Vishera, and a ton of it with Kaveri once it hits.

I don't think AMD is going to fold anytime soon.  They aren't really losing money, they keep on producing products, and some of those are even popular (GPUs, Trinity...).  Time will tell if they have what it takes to keep competing with Intel, but I like what I am seeing with Rory at the helm.  Now if the rest of the new executive team can start making a positive impact...

July 6, 2012 | 12:17 PM - Posted by Anonymous (not verified)

I still <3 AMD

July 6, 2012 | 01:58 PM - Posted by Anonymous Coward (not verified)

I just want AMD to do something that's enough of a squeeze on intel's balls that they release ivy-e, with a 22nm 8-core extreme edition. They could have easily released that by now, but what reason would they have. AMD is KILLING the high performance CPU market with how bad their CPUs are.

So pardon me that I don't get my pants wet over trinity or whatever else is the hot new thing in laptops costing as close as possible to $0.

Do something AMD. Create something. These days you're nothing more than a patent troll with an x86 license.

July 7, 2012 | 11:42 AM - Posted by Nilbog

You are an idiot

First of all, these are no longer x86 processors. They are x64, they just backwards compatible.
Second, AMD INVENTED x64. Intel pays for a licence just like how AMD pays to use x86 instruction set.
So if you read a little, you would know they they sometimes help each other out. If someone has a great idea, like x86 or x64, you share it. Its better for everyone.

Also, wtf do you call Bulldozer? That is an entire new way of looking at processing. They literally redesigned the processor completely. You have to try new things to move forward. You cant just write this stuff down, and realize that something isn't right. You go for it, if it fails. Oh well, at least now you know for sure.
Somebody has to try new things, if Ivy Bridge flopped you would be saying the same thing about Intel.
They have created a lot, its just not always a win.

Also, who is AMD suing over patents? They cant be a patent troll if they aren't suing anybody.
Idiot

July 7, 2012 | 02:06 PM - Posted by Josh Walrath

Lol, easy there, tiger!

Licensing is a sticky point with this.  AMD based AMD64 on Intel's x86, so from what I gather Intel does not actually have to pay for that technology, since it is wholly and primarily based on an Intel license/technology.  AMD and Intel did agree to a cross-licensing agreement a few years back due to all the legal issues that Intel ran into with being anti-competitive (the evidence was pretty overwhelming, which is why Intel paid AMD that $1.2 billion or so, plus the 5 year cross-licensing agreement).

The Bulldozer architecture was very, very agressive considering the new ground they were covering.  We can look back at the original Phenom, and though that had a lot of technology that was based on the Athlon 64 series, it had a very hard time of it.  Bulldozer is obviously a step well above that implementation, and when we consider the resources that AMD has, even getting the processor out the door in a somewhat timely manner is a minor miracle.  So yeah, there are things that AMD could have done to help mitigate this situation.  For example, they could have taken the Phenom II, ported it to 32 nm, and integrated the new front end and memory controller that was developed for Bulldozer.  I think we would have seen a much more competitive product without as much rework that had to have been done with the dual integer units and the large, shared FPU/MMX/SSE unit.  Oh, and made each core a 1 MB L2 cache unit rather than the 512 KB unit we see on the Phenom II.  I think that product would have helped AMD hold steady and be a lot more competitive, but they simply did not choose that direction due to monetary and design team issues.

July 7, 2012 | 07:49 PM - Posted by Nilbog

Lol, my apologies. I'm finally just sick of people like that guy

July 9, 2012 | 07:02 PM - Posted by Anonymous Coward (not verified)

Freedom of speech is a bitch, isn't it?

July 9, 2012 | 06:57 PM - Posted by Sam Wallace (not verified)

Gee, name calling and wrong. Correct nomenclature is x86-64 where 64 is an extension of the x86 instruction set.

July 9, 2012 | 10:47 PM - Posted by Josh Walrath

Heh, nomenclature is variable depending on the OS.  FreeBSD and others use amd64, others use x86-64, and there are a couple other ways of expressing it.

July 6, 2012 | 02:07 PM - Posted by IronMikeAce

The point of the Bulldozer arch with the modules having two integers cores and a single fpu each is meant for their APU(HSA) chips. Since GPUs are so great at FPU workloads the Bulldozer arch will offload these workloads onto the GPU which is why the FPU on Bulldozer is week and the Integer cores are strong. Once AMD fully implements this concept onto their APUs they will be closing the performance gap on Intel fast. Kaveri will allow both CPU and GPU full access to system ram which is a big step towards their HSA concept and we should really start seeing what HSA and APUs are really capable of.

July 7, 2012 | 11:15 AM - Posted by Nilbog

Great article as always, thank you Josh.
This has got me thinking about the shared Cache. Correct me if i'm wrong, but wasn't the assumption about Bulldozers poor performance the shared Cache? I remember reading about people assuming or guessing, that the shared Cache was in fact the issue with multi-threaded applications. That the shared caches weren't big enough for all the cores to use at once efficiently?
Am i confused?

This may sound silly, but i don't see why AMD isn't talking to the guys at Texas Instruments. I literally just saw an updated "How Microprocessors are made" that was filmed at one of the foundries. Looks like they actually know what they are doing, and seem quite confident.
Maybe even more silly, but why not ask Intel? AMD has stated that they aren't trying to make "needlessly powerful CPUs" So that doesn't sound like much of a competitor, or threat to Intel.

I hope that somebody will finally be a REAL competitor to Intel. One company dominating the high end market is never good for consumers.
I've got my fingers crossed for AMD.

July 7, 2012 | 02:22 PM - Posted by Josh Walrath

I think Bulldozer is a good basis for a next generation architecture, but the combination of design lessons learned and the 32 nm process they are stuck with, the first iterations are just not as impressive.  A lot of talk about the caches has been bandied about, and that is true to a point.  There does need to be larger L1i and L1d caches.  Associativity does need to be addressed as well, as the different threads look to be somewhat bottlenecked with cache accesses (though I can't remember offhand what level of caches is the most troublesome in this regard... possibly L1).

Trinity did primarily improve thermals and power characteristics, and that was the big jump.  A bunch of smaller changes were obviously made to many different parts that has improved performance.  Again, Vishera should see more of these optimizations all the while keeping TDPs in check at 4 GHz and possibly beyond.  So, AMD will push Intel in the budget and midrange, but they still have yet to have an other Athlon 64 where they were head and shoulders above what Intel had.  But AMD is trying to focus on growing markets, and the top end enthusiast market surely is not that.  We really are at a spot where pervasively good enough computing power is not holding us back.  So AMD is starting to really push into the lower power areas that Intel has ignored until recently.  Bobcat was a great start in that direction, and the upcoming next gen parts should do a whole lot better.  AMD will certainly stick around and be an option for folks, but they are not nearly as worried about the enthusiast desktop as they once were.  When the currently lineup of APUs have good enough performance in both CPU and graphics, plus a decently performing (and featured) southbridge to help drive it, they are not nearly as worried about the high end where it costs a lot to compete for very little actual gain.

July 7, 2012 | 07:44 PM - Posted by Nilbog

Sometimes i read the comments from you and wonder why you dont work at AMD.
I think see what AMD is trying to do, they are going from the bottom up. They want to own the low power market, so that they basically have a guaranteed income. Then most likely they will go crazy with R&D when they save up a nice $ cushion. Then i would imagine, go after the high end.
My worry is that by then they will be associated with cheap, just powerful enough chips. Sound familiar?
If ARM made a desktop CPU would you have any hopes for it, besides drawing like 5W from the wall? I would chuckle and never think about it again
Even worse by then Intel could be many more years ahead of them, and they could just never quite catch up. Back in the same hole as today.

Im not against this new road map, i just think it would be foolish to literally let Intel own the high-end market.
To quote a certain someone here; "Seems like a cop-out to me man..."

July 9, 2012 | 10:38 AM - Posted by Josh Walrath

Hmm, I'm not entirely sure that is what they are doing, or that is what their plan is.  I think it boils down more to they have an advantage in integrated graphics technology, and the price points of those chips that are actually using the integrated portion is well under $200.  When people start looking wholly at CPU performance and are willing to spend over $175, their only real choice is Intel.

ARM will be making a server/desktop chip, but this will be spearheaded by NVIDIA.  These chips won't sit at 5W TDP.  They will be 65 watts+ for the higher performance versions.  It will be interesting to see where this goes, as MS is going to push hard on Windows RT.  If NV does make a successful product, and MS has a nice installed base of applications for WinRT, then the dynamics of laptops, desktops, and servers will irrevokably change with the addition of a 3rd large fab-less semiconductor company producing large, high performance CPUs.

Vishera will show us what AMD is capable of, but I think the jump to the 28 nm Kaveri will be a much bigger release.  CPU performance as well as the fully integrated (talking shared memory architecture) GCN unit on a more bulk oriented process will show if AMD's gambles over the past few years will pay off.

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.