@Medo, You are speaking from the understanding of diagrams presented by reviews sites of a microprocessor architecture discussion. The same diagrams given to them by the chipmaker. These sites are run and presented by people who are not engineers having worked in the industry, Anand Lal Shimpi for instance has a degree in computer engineering but no real experience applying any of it. He comes across as an ignorant person many a time. But that's exactly why he is a journalist now and not a real engineer on the inside working for chipmakers.


Chip designing is not something the man on the street can simply direct on his whims because some site preached that shared decoder was a bad idea or some other abstract model known to him was poor or so on and so forth. Being a CEO you are still advised by your engineering heads if your ideas are practical or even possible.

Ideas are good to have when you don't have engineering backgrounds. But many ideas go down the drain because they are not practical or possible.

You are also assuming there are some big problems with the shared architecture and that somehow Intel's better. Apart from the diagrams shared by these companies to review sites, can you or anyone else really give a technical reason why the designs are not delivering? Nope you can't because their logic blocks and what's inside them is their trade secret whether they do good or bad, you don't know it. You don't know the critical path, you don't know the timing skews, you don't know the various active history data of their cache structure, you don't know exactly what is going on in their pipeline stages etc etc. These are closely guarded secrets.

No micro-architecture out there is bad in concept because they are set out with certain goals in mind, it only fails if it doesn't meet those goals. Thats why they work on improving it when they don't deliver according to previously set goals. Statements in the reviews about how bad an architecture is are just petty arrogant statements from ignorance. If they are so smug being the apparent know-it-alls they should come and join the industry to do real engineering work. But they cant do that. They rather "engineer" running their mouths and ranting in their reviews.

Then we have many experts out there who's quick BIO review reveals they did practically nothing in the industry but now they are "analysts". And if they pass judgement it must be gospel, this then gets latched on as fact and disseminated through the internet.

You must also realize AMD has historically suffered dips in sales when Intel suffered dips in sales, so having 80% of Intel performance is of little use if its not the outright leader in the performance. PC market regardless is going to decline to such levels that workstations will be a small niche. People are moving to the mobile computing age. Gaming consoles have done enough damage to PC gaming that PC gaming is pretty much a port of the console version. There is little need for the PC now unless you develop content or want to make your own home videos or creations. Then you build that niche workstation in the future. For the vast majority the PC as a 'need' will completely go away.

You have seen AMD's dire situations for the last what, 1 year? There is a good 3-4 year period yet to be seen to truly get an idea of where they are heading in their new direction.

Also you haven't seen their roadmap for the next 3-4 years have you? If that was known you will change your opinion

darn, even i gave you a thumbs up.

good reply, it hit like a brick, and re-adjusted my views.

However i cant stop help thinking that there was a lot of missing opportunities.

Why no one foresaw the economic crisis, and instead of creating a new Arch like the FX that have shown a very dim light in power efficiency, which in today world is a big thing, and counts as much as performance for some people, including me.

The lowest and latest frequency FX processor (AMD FX-4300) can reach TDP on load of 125~135, which is like an i5 intel cores, however performance is not the same, neither multi or single.

However i have a8-3850 APU, which contains the old and Athlon cores produced in 32nm same as the FX, that can reach 130~150, GPU included and North bridge included, while the AMD FX-4300
is only modules, nothing more fancy.

Remove the Integrated GPU from the a8-3850 and believe me, that thing will be very power friendly, like 80~90w, which is like intel i3 series in power efficiency.

I know the world is going mobile, and its crucial to have everything in one chip.

however i miss buying an good old processor from AMD, with nothing fancy, just cores, and i like power efficiency so i will pair with silent HD 7750, i want something like quad core phenom or athlon, that is 65w, or something like that,and there is nothing AMD has to offer in that regards.

All the new stuff is just too fancy for me, with the modules and more, i like Mini-ITX form factor, i have never build it, but i would love too, cause i travel alot, and i still believe that desktop will remain, however it will change its size, no more big cases, just smaller i think, and quad core with small 7750 and small case for my computer usage, will be perfect for me.

What you (and I) like and what the market demands are two separate things. Companies must make profit to justify their existence.

For the latter part of your post, a good deal of your wishes will come true in some form or another. That is all I can tell you or anyone for now. AMD is changing and they have to its for their own good, there is a future in their plans. Its just a little different than what most people traditionally expect.

As for the people not seeing the economic crisis coming and how the tech market changed by storm.

One word: Apple.

I hate apple and their products with a passion because the fanboys have a done a great job to act like cult religion hooligans. Most of the hate for apple stems from its fanboys and their petulant attitude and antics. But you have to take your hats off to those clever rascals at Apple, especially Jobs. They made the phone now a fashionable item, its more than a phone its now a status symbol. They made it ok to buy crazy expensive items like the iphone completely ok and casual. Most people takes months to decide on a car, for a phone? 1 week tops. Their initial moves into the tablets with Ipad also picked up the tablet market. Android is sure leading the technical front now but Apple started this trend. They maybe milking customers now with lackluster improvements in successive generations making money on the established name with jolly good fanboys to defend their every move but it sure is a nice time being Apple now.

They are the ones responsible for the situation now, and there is no turning back. Apple showed the true power of mobile computing and its possibilities. Others followed with even more revolutions.

People are now buying these things like its some vegetable, I cant believe my eyes seeing some people who barley have money to eat 3 meals a day are saving up to by an Iphone or an S3/S2. I am not joking.

That cult effect is catching on, now everyone can stand the occasional slowness in tablets and smart phones, its ok to be deal with slow computing which is now "just enough" computing, because its mobile its where ever you are whatever the moment.

Standing in line to board a plane? no problem, take that smartphone out. Laptops? you can forget about it, posture and ease of use disaster with netbook and laptops standing in a line.

These trends have completely ruined the laptop and PC market, no one could have predicted this to be honest, I am not sure even Jobs thought this was possible at this pace.

And because of these devices, more computing time and improved battery life is calling for low-power maximum performance computing a must. Everything is going low power now. Soon 90-100 watts CPU powered workstations/servers will also go extinct. It will be the domain of power efficient mini blades and machines with fabrics in servers.

There is much efficiency to be had to low-power computing and with efficient software written around efficient ISAs like ARM. X86 is going to have loose weight and go on a diet to stay competitive.

Many people blame Microsoft for not doing anything, they have done everything to be honest, but its not easy to push something and expect it to pick up all the time. MS saw the trends and knew they missed out big time and now they are scrambling to get a big part of that pie. Besides what can an OS do apart from giving the platform yeah? Where is that Killer app? Not games, not everyone is a gamer, there should be something that appeals to everyone.

That friends is the age of mobile computing, people want their social media where ever they go whenever they want it. In the rest rooms, in a cafe, in a bar, in a line waiting for some tickets.. you name it. Missed the name of that drug you wanted to buy and you are at the store right now? No problem google it right there on your smartphone or tablet.

People have found that killer app, only its not an app, its a concept of pure mobility. Who knows the next frontiers are already here before us, with glasses or clothing that integrated technology to make it seamless and completely hands free, you just talk to your computer and it talks aback to you with your answers. This is where the world is heading.

We have an economic crisis even now, but its hardly making a dent on smartphone sales, the marketing has really gone to people's heads, one would be foolish not to cash in on this trend as its only going to get bigger.

You are comparing crippled version of a product(FX-4300) to a non-crippled version of another product (a8-3850).

And TDP is completely different thing than average power or power consumed to do some task. TDP is a value manufacturer DECIDES to use and which he limits the chip.

Compare that a8-3850 against Trinity instead and it's very clear that Piledriver is superior architecture compared to K10, especially on single-thread performance.

K7 was a great design 13 years ago. K8 was an improved K7 and was a good design 9 years ago. K10 is just slightly improved K10 and was hopelessly outdated years ago.

And K10 as it exists only because the original planned successor of K8, K9 got into so big troubles that they had to completely abandon the design. So they just quickly added L3 cache and widened the SSE unit of K8 and got a K10 out from K8.

When you have outdated small cores, you can always add many of those and get good multi-threaded perfromance, but single-threaded perfromance which matters more still sucks. This is what AMD did with Phenom X6.


Bulldozer was the second (or maybe even third? there have been rumours of another canned cpu architecture) attempt to develop completely new cpu architecture after the K7-derived designs.

This time they understood that improving single thread performance is getting difficult, so they concentrated more on improving multi-threaded performance. But this was the wrong design choice; When something is hard, you just need to try even harded to do it, not give up. And then the first version of bulldozer had quite many small performance problems of which many, but not all, are fixed in piledriver.

There is lots of room for improvement in bulldozer/piledriver, but K10 is already "maxed out"-design. For example FMA instructions cannot be added as an aftertought, all the register file read ports and bypass networks have to be designed for it, and you cannot make them wider later without complely redesigning them and that may then completely change critical paths, pipeline balance, clustering inside the core etc.


So AMD had to do the total redesign.
They chose the optimal pipeline length, which should have given them like 25-30% more clock speed, at cost of like 8% ipc. But something went wrong with the design and they did not reach so high clock speeds, only slightly higher than previous design. The architectre choice however was correct. And they sacrified some single-thread performance to make the cores smaller to fit more of those.

Only applies to complying with smaller power form factors. When its for high end systems its a function of Max junction temperature for the process used.



These are sweeping statements without much truth in any of it. This is what general public assumes based on what reviews sites have theorized, these sites have very little understanding of microprocessor design. If you think K10 is just a slap of L3 and wider SSE, then you grossly misinformed. The changes are much deeper than what any of the press is informed on via marketing slides provided to them by the very same chipmaker. Those will remain trade secrets for any company on their respective architectures.



How did you know "they understood"? Again sweeping statements.



This is gibberish, it makes no sense whatsoever.

Instructions are realized through logic, logic is formed by combination of gates, gates are formed from transistors. You can make logic to realize any instruction new or old. Its upto the design team to simulate and determine the results with said architecture to see if the gains are as much as they want or lesser than expected. There is nothing to stop FMA or any other new instructions being added to K10.

Adding such logic may or may not need changes elsewhere, and if there are changes needed they a case of how much return you get for the effort. It is always preferable to simply support the new instruction set and make the changes needed for it to work. But in most cases a more severe change maybe needed to get the best out of the new instructions.

'Network bypass' that is a term used specifically in certain specific aspects of design of certain specific logic. What were you referring to? It seems you read this term somewhere without understanding what it does.

Another misconception, register file ports need not be redesigned all the time. Infact most times you don't even touch them. Its a question of balance in returns. Port reduction methodologies have been existence for years now.



No, this is the lack of understanding in the media who later convey their opinions to the readers who then later digest this misinformation.

The bottlenecks and issues felt in some of the single threaded applications are elsewhere in the design. Those will remain trade secret regardless of it being good or bad. Sometimes companies will not bother to correct misconceptions because that would lead them to divulging some critical info on subjects they don't want the competition to get wind of.

You don't know unless you worked in AMD if the choice in not chasing higher per core IPC was a conscious choice made by design team.

We have in industry, simulation models in pre-silicon phase once we are close to GDS generation, to accurately model the entire microarchitecture with an error margin of no more than upto 4-5%. Even before that simulation process happens, its an ongoing process that is periodically updated from the time the project was first conceived to the point of GDS out.

This is to say AMD's engineers knew exactly what to expect even before first silicon tape out.


It is not your fault. I blame this sort of misinformation on the certain review sites who misguide readers.

But you have to transfer the operands from the RF's (or bypass then from other results) to the function unit.



FMA requires 3 operands. Other FP operations require only 2 (or 1) operands. The bypass network and register file read ports are designed to transfer certain amount of operands per cycle.

You can of course implement FMA by doing it with throughput of 2 cycles, by reading the operands in consecutive cycles, or by splitting the FMA into separate mul and add micro-ops, but then the performance will be bad, all the peformance benefit from the FMA instruction will be lost. So you practically have to designs the bypass network and add some register file reads ports when you add an FMA operations to a FPU which previously only had 2-operand instructions.


Adding instructions which do not need more operands than previous operations is easy though, operations which only affect things inside a function unit.

And this task is easier to implement unlike what you assumed earlier that is somehow was an insurmountable task for the K10 design.

Operand bypass networks (with a focus on transport) are reported widely as efficiency choices, not a requirement. Decision to use it depends on benefits seen in modelling for your specific design.

There are various results reported in literature not everything makes into commercial usage.



This depends on what usage models are taken into account, if the amount of FMA workloads are less than ideal in the given code, there is not much return in investing time, die real-estate, RTL verification etc to boost all possible performance of any instruction added.

And as I said earlier if the requirements for the design are met it doesn't matter if there was more performance to be had but neglected because a decision was made on what was enough as target. Should it be a design performance target, facilitating a 3rd or even 4th operand if not as complicated for design teams as some may believe.

If they wanted to put FMA in K10 its very straight forward.

There is always that tendency to squeeze in as much performance as possible, but when you are working with die size constrains and evolving not so robust EDA tools, you cant chase every drop of optimization. Trade offs are normal.

Besides FMA will take a lot time before we see it seriously used in that limited space where consumer software can benefit. It sees immediate benefits in some of the server/HPC markets.

K10 can have FMA now and it wouldn't matter in the overall picture because its adoption for the general public software space is already slow.

Port reduction takes care of the need for extra ports, impact is minimal and in some cases none, design depending of course.

The reason we have bulldozer family now and not a straightforward evolution of K10 is a mix of both business level executive decisions combined with the options advised by the design heads. This I know from AMDers themselves, their take not mine.

I can see why, their heads must have envisioned a world in general moving towards highly threaded workload model sooner rather than later, but its looking like its going to be a long wait at the moment.

The CEO needs to get sacked.

Not sure if this was addressed to me as its a little confusing to make out who replied to whom in this format.

Anyway, Rory Read along with his main team have come up with changes and new plans to take the company forward. Employees are putting faith in the roadmap, not in the management team. They have enough to worry about than to second guess the strategy. They have engineering work to do. Engineers don't care who they (management) are and what they have accomplished elsewhere. If they pull it off? great, if not the well oiled revolving door will be shown.

A lot of top notch people were let go along with some not so top people and some people who's roles had become redundant after the organizational and procedural changes. Loosing good people is bad, but it had to be done. The only wish would be if they could get their numbers right the first few times and not have to pull frequent publicized layoffs. That is bad for moral and company's out look.

All that said its a way too early to say if anyone needs to be sacked. So far its 50-50, nothing has been achieved and the time frame for it to be achieved has not yet passed. Until then its prudent to reserve judgement. I also thought he was doing some idiotic things at first but that was due to lack of insight into 'what' he was doing. I have that now and my position has since changed considerably.

"All that said its a way too early to say if anyone needs to be sacked."

Until its too late.

"I also thought he was doing some idiotic things at first but that was due to lack of insight into 'what' he was doing"

Some idiotic things??? You are being way too generous towards him.
What you should have said is way too much idiotic things done by him that has driven the company into the ground.

Do you know about the roadmaps not shown to the press? I do. Based on that alone I can see the difference plus I have input from other means to know where its been heading. Hence the change of position in my view.

Rest assured you wont do any better or even come close to what he and his team has done. Your credentials are non-existent his and his team's are there for anyone to evaluate whether its good or average or bad. If you think you can do better feel free to head directly to the AMD's board of directors to present them your elevator pitch on how you can do things much better.

"Do you know about the roadmaps not shown to the press? I do."

You haven't presented us with something or your so called information that you have so your claims are irrelevant.

"Rest assured you wont do any better or even come close to what he and his team has done."

Oh they have done something alright. They have driven the company into the ground.

"If you think you can do better feel free to head directly to the AMD's board of directors to present them your elevator pitch on how you can do things much better."

What, head directly to AMD's board?
I don't need to because they have dug their own grave already.

Give you the roadmap? In your dreams. I don't leak such info. It is not important to show you anything. Your interest in AMD's products are not my priority. I have seen it and I know what I initially thought was wrong and have since changed my views. That is about my position.

You position is that of someone who has no idea on anything. It wouldn't matter if someone did show you undisclosed roadmaps you will still dismiss it as photoshoped or fake. Its pointless.



Hang on, you were the one who suggested the CEO needs to be sacked. So if you cant handle the job, your are in no position to suggest someone's sacking. You lack any pedigree on such matters or else just step up to the plate.

PS: Oh now that I found out you are AvonX, explains alot about what was going on until now. Seen here towards the end of the thread:

http://www.xbitlabs.com/n...p_Chip_Due_This_Week.html

WOW thats a wicked setup you have there. You must really want to have a tablet I can see why.

I guess if you can get your hands on a sub 100$ tablet 7" like a single core Cortex A5/A8 based machine with 512MB Ram it should serve your needs.