The AMD Execution Thread [2007 - 2017]

Status
Not open for further replies.
I still wonder why Stars wasn't further developed. Say a 12 core Stars without IGP and a less useless Turbo. I suppose it's been discussed. :)
 
Last edited:
I still wonder why Stars wasn't further developed. Say a 12 core Stars without IGP and a less useless Turbo. I suppose it's been discussed. :)

Probably inertia. By the time people started realizing that Bulldozer sucked, the company was too invested in it and it was too late to go back to Stars without incurring enormous expenses and wasting a lot of time. Plus it was likely felt that some of Bulldozer's biggest issues were fixable, and I think Excavator shows that to be at least partially true.
 
Or maybe a stubborn person or group of people wanted to somehow prove a point.
And failed.
 
Not core for core. Sure, it had more cores than the PII X6, but only because the finer process allowed it.
I won't go as far as digging old reviews but what I remember for what it is worth is that Bulldozer sucked compared to Phenom in most metrics and the only saving grace was support of new instructions that allowed big wins in specific workloads.
It is pretty terrible when you consider that it was indeed using a newer process.
May be some issues with bulldozers were fixable but one may say that fixing issues is what AMD have been doing for now 4 years with the results we know about. It tooks lots of efforts and time to get from bulldozer to excavators, in the meantime Jaguar which looked like to me like a "simple" but sound and solid basis to build upon (different market segment than BD) ended up canned. By killing Jaguar line AMD gave up on the really low power segment (tablet, cheap notebook and what could be a growing market mini desktop/STB if not HDMI keys) by sticking to a big core only they are sure to not be in a situation on perf (and perf par watts) with Intel higher end offering and it will be as tough to fight with Intel lower end line of product (Atom, Quark). I know Intel wants to get rid of Atom but looking at the slowing down of progress on the lithography front it would not surprise me if they keep it going for longer than they want, pragmatism is always a good choice especially when it is the only one the table :LOL:
I agree with an answer 3Dilettante made to one of my post a good while ago, Amd may not have other choice as bad decisions are unwinding and time travel ain't an option :(

Other than that I don't see the departure of Keller as a good thing, foremost from a managerial pov it is never really good to have so much changes at the top, AMD might be perceived as a freaking roller coaster by the engineers /employees.
 
Last edited:
I won't go as far as digging old reviews but what I remember for what it is worth is that Bulldozer sucked compared to Phenom in most metrics and the only saving grace was support of new instructions that allowed big wins in specific workloads.
It is pretty terrible when you consider that it was indeed using a newer process.
May be some issues with bulldozers were fixable but one may say that fixing issues is what AMD have been doing for now 4 years with the results we know about. It tooks lots of efforts and time to get from bulldozer to excavators, in the meantime Jaguar which looked like to me like a "simple" but sound and solid basis to build upon (different market segment than BD) ended up canned. By killing Jaguar line AMD gave up on the really low power segment (tablet, cheap notebook and what could be a growing market mini desktop/STB if not HDMI keys) by sticking to a big core only they are sure to not be in a situation on perf (and perf par watts) with Intel higher end offering and it will be as tough to fight with Intel lower end line of product (Atom, Quark). I know Intel wants to get rid of Atom but looking at the slowing down of progress on the lithography front it would not surprise me if they keep it going for longer than they want, pragmatism is always a good choice especially when it is the only one the table :LOL:
I agree with an answer 3Dilettante made to one of my post a good while ago, Amd may not have other choice as bad decisions are unwinding and time travel ain't an option :(

Other than that I don't see the departure of Keller as a good thing, foremost from a managerial pov it is never really good to have so much changes at the top, AMD might be perceived as a freaking roller coaster by the engineers /employees.
Well you should have looked at old reviews, as it was for the most part faster, though not nearly enough considering the process jump and sheer size of it
 
Well you should have looked at old reviews, as it was for the most part faster, though not nearly enough considering the process jump and sheer size of it
Well the FX8150 outdid the Phenom II X6 thought it is in no way a fair comparison, more cores, more silicon, more power.
That is a fairer comparison, new instructions support and increased simd capabilities helped the processor to age better (you would expect that much of a newer processor). If the fx6300 is anything alike the FX8150 then power efficient has to be worse than that of the phenom.
Anyway it is an old story, lets hope Zen cores come out right hence...
 
Not sure what you mean by fair since I didn't specify any specific review, I just remembered that it was usually faster than the Phenom II X6 and checked a Techreport review to be sure. I still think it's a shit CPU design from that the company who made Intel look foolish wrt the Netburst architecture, you would think at least someone there would recall what happened to Intel when they pursued a similar route.
 
I still wonder why Stars wasn't further developed. Say a 12 core Stars without IGP and a less useless Turbo. I suppose it's been discussed. :)

Whether Bulldozer was a successful replacement is separate from whether Stars was running out of steam.
Llano was what resulted from Stars going to 32nm.

Architecturally, Stars was behind in terms of memory speculation, cache hierarchy (it's another question if BD was a step forward, and some of the poor elements are holdouts from Stars), and an integer pipeline with reservation stations instead of a physical register file, which everything that succeeded it has moved to.
Stars also didn't have SMT. If Bulldozer is criticized for only making it halfway to supporting SMT like its ostensibly superior successor Zen, then Stars can be knocked for stopping short of that.
The architecture targeted a certain set of nodes, since manufacturing and physical constraints changed so much over time, and it was dragged past even those by the time 32nm came around.
 
So you think Stars was just too obsolete to bother with developing any further? Bulldozer sure went badly though.

It's going to be interesting to see if Zen is an extremely smart design or another collection of good and bad ideas.
 
Last edited:
So you think Stars was just too obsolete to bother with developing any further? Bulldozer sure went badly though.
This is why I made the point that the first sentence is a different issue than the second.
The need to replace something does not make the success of its replacement's design effort more likely.

It's not like Stars didn't have generations of lukewarm performance scaling. That 90nm FXs maintained their clock leadership for so long is a data point in favor of the idea that the base pipeline was conceived with the constraints of that era's process and physical needs.
AMD failed repeatedly to replace Stars (at least 2 BD attempts, there were rumors of a speed racer and brainiac besides those), but the reality is that it kept trying all the way to Bulldozer and lost ground to Intel the more it failed.

Llano had various headwinds against it since it was an APU, which makes the comparisons imperfect, but it had some of the most serious problems on the process it moved to compared to Bulldozer.
 
Chipmaker AMD to cut about 500 global jobs

Struggling chipmaker Advanced Micro Devices Inc said it would cut about 500 jobs, or 5 percent of its global workforce, as the company looks to rein in costs amid weak demand for its chips used in personal computers and intense competition.

The company said it would record $41 million of the expected $42 million charge in the third quarter ended September.

AMD said it expected savings of about $58 million in 2016 from the restructuring plan.

The company has also been shifting to gaming consoles and low-power servers, but progress has lagged Wall Street expectations due to intense competition from Intel Corp and newer companies.

Shares of AMD were flat at $1.74 in after-hours trading.
http://www.reuters.com/article/2015...20151001?feedType=RSS&feedName=technologyNews
 
What does that mean in plain english
It means that the total cost of the restructuring is expected to be $42M and that $41M of that cost will be put in the books for the September quarter.

So expect a $41M line item in the upcoming 10K report and another $1M in the one of January.

Even more concrete: the Oktober quarterly loss will be $41M larger than without this restructuring, and there will be at least a $41M difference between GAAP and non-GAAP profit/loss.
 
AMD veteran, HSA president Phil Rogers leaves company for Nvidia
Twenty-one-year AMD veteran and HSA (Heterogeneous System Architecture) president Phil Rogers has decamped from AMD to join Nvidia. Rogers will be taking over as Chief Software Architect of Nvidia’s Compute Server division, at a time when Team Green is rolling out features like NVLink and continuing to push forward with its own plans to make CPU and GPU compute more capable. With HSA 1.0 now complete, you could argue Rogers is taking a page from Jim Keller’s book and moving on now that he’s finished his work.



http://www.extremetech.com/extreme/...esident-phil-rogers-leaves-company-for-nvidia
 
I don't know about looking at version 1.0 of anything and feeling like there's not enough to look forward to. He was an old hand at AMD, so it must have been quite a draw to take him out of AMD and the HSA foundation, if not a push.

An AMD that was more secure along those lines would be one where a critical element of HSA--the GPU--was more integrated with the direction of the company (like having part of its direction covered by the CTO), was expanding its software development, and whose other critical element--the next-gen CPU--was coming in the form of an APU sooner rather than later.
 
It's evident AMD is preparing to split up in a CPU and a GPU business part, selling one part to save perhaps the other.

There is plenty of powerful / low power GPU IP these days to integrate with a CPU.
 
It's evident AMD is preparing to split up in a CPU and a GPU business part, selling one part to save perhaps the other.

There is plenty of powerful / low power GPU IP these days to integrate with a CPU.

Like ARM ones ? Im sorry but i dont really see how AMD will fight Intel IGP ( who become more and more powerfull each time ) with a imagineon or Adreno gpu's inside. let alone it will contradict completely the strategy of the company on the computing side. How will they include support of DirectX ( 12 etc ) and other computing stuff needed in their strategy of APU for server / computing / gaming.

The ARM based SOC IGP start to be interessant on a mobile gaming point of view, but they are at some thousands of miles behind what can offer them the actual AMD GPUs for APU's.. for computing or for gaming.

And when i speak about strategy for their APU, the first thing who come in my mind, his the Exascale race: http://www.hpcwire.com/2015/07/29/amds-exascale-strategy-hinges-on-heterogeneity/
 
Last edited:
Status
Not open for further replies.
Back
Top