Poll: What would make you the most happy with AMD?

[max347]

Trying to collect some information here for a hypothetical business strategy (class). Your participation is greatly appreciated.

Options-

1) Only sell Bulldozer cpus from now on and drop Thuban/Deneb production
2) Sell both Thuban/Deneb cpus AND Bulldozer cpus, as-is
3) Die shrink the Thuban/Deneb cpus for power savings and higher clocks, while selling BD on the side

Feasibility is not an issue, just seeing if the populace thinks a die shrink of the last gen would be an appreciated course of action.

Thanks!
max347

 

[OVerLoRDI]

8 core higher clocked thuban. Ideally with tweaks to improve ipc a bit.

(not very informed rant below)

It also would have been cheaper to produce. Assuming transistor count scales linearly with core count (which it doesn't perfectly), Thuban is ~950 million transistors, an 8 core Thuban would be easily be well below the 2 billion that is Bulldozer. Also to my understanding bigger chips are more likely to have yield issues, thus AMD could potentially be getting better yields with a smaller 8 core Thuban.

Potentially this 8 core Thuban would be cheaper to develop. Phenom II architecture is already designed. Shrink, tweak, and make Phenom III.

Also it potentially would be a drop in upgrade to current AM2+/AM3/AM3+ boards (Thuban works on some AM2+ boards) Now this decision may not make motherboard manufacturers happy in the short run, it would go wonders for winning loyalty with customers. Also if your product is a disappointment, you won't sell many chipsets and motherboards anyways.

Also factor in pricing. In this hypothetical world tweaked Thuban IPC > Thuban IPC. Shrunk Thuban clocks > Thuban clocks. 8 cores > 6 cores. This all points towards better performance, which means AMD can charge more, which is good for them.

Factor in the smaller die size, potentially cheaper development costs, better performance, consumer loyalty points, and potentially higher asking price. AMD would be doing a lot better.

Now my knowledge of silicon design is extremely limited so I don't know how much performance they could tweak out of Phenom II. The thing that I keep wondering is, looking at all the resources and time they threw at Bulldozer and the results we currently have; couldn't a fraction of those resources gone towards a shrunk and tweaked Thuban and yielded better results? Also this tweaked Thuban could have potentially come out a lot sooner. Did AMD not realize that Bulldozer was a mess until October?

It just feels like a ridiculous amount of fail took place.

 

[exar333]

8 core higher clocked thuban. Ideally with tweaks to improve ipc a bit.

This.

How about releasing some nice Opteron chips with 8-12 cores on the consumer side? Seriously, some folks would pay $600-1200 for a 12-core beast that they could throw in a consumer board.

 

[max347]

Thanks for the votes guys!

Yeah an 8 core Thuban would be ideal, but my proposal just includes the die-shrink assertion. Even with an older architecture, I am thinking the power savings and improved overclocking ability would be favorable to the module design.

 

[eternalone]

They need to scrap BD and just put their resources into the APU line and server area, which I believe they already have begun. They should just drop their Desktop CPU line altogether, which they will also start doing I believe. APU is the future of AMD not Desktop. Intel has won, no need to release anymore crappy BD or PD modules to save face.

 

[Idontcare]

Thanks for the votes guys!

Yeah an 8 core Thuban would be ideal, but my proposal just includes the die-shrink assertion. Even with an older architecture, I am thinking the power savings and improved overclocking ability would be favorable to the module design.

They already shrunk the stars core to 32nm, see Llano. The CPU portion of that chip is 32nm stars core.

Considering how long Intel has been shipping CPU's with integrated GPU's, bulldozer should have never happened without a GPU. Trinity should have been the first bulldozer launch with Llano holding AMD over until then.

 

[Arkaign]

Yeah they need to unleash Llano with a BE, and ramp up to offer 6 and 8 core versions of it.

 

[gmaster456]

A Phenom ll die shrink would be killer. That alone would be a SB competitor. An 8 core thuban would also be very welcome.

 

[OVerLoRDI]

They already shrunk the stars core to 32nm, see Llano. The CPU portion of that chip is 32nm stars core.

Considering how long Intel has been shipping CPU's with integrated GPU's, bulldozer should have never happened without a GPU. Trinity should have been the first bulldozer launch with Llano holding AMD over until then.

I do not see Bulldozer's lack of a GPU being a significant failing. Also given its size of 2B transistors, were they planning on making it even larger?

The target market for Bulldozer is "enthusiasts" and high performance users. These users typically do not care one bit about an integrated GPU. Maybe with some of the lower end Bulldozer offerings you have a point, but the main problem is the lack of performance, not the lack of features.

 

[Idontcare]

A Phenom ll die shrink would be killer. That alone would be a SB competitor.

 

[OVerLoRDI]

A Phenom ll die shrink would be killer. That alone would be a SB competitor. An 8 core thuban would also be very welcome.

Not really. SB IPC is significantly better than Phenom II. You'd have to ignore the incredible overclocking of SB and allow a Phenom II die shrink to be oc'd for it to be anywhere close to a SB competitor.

Phenom II would need an IPC boost.

 

[frozentundra123456]

8 core higher clocked thuban. Ideally with tweaks to improve ipc a bit.

(not very informed rant below)

It also would have been cheaper to produce. Assuming transistor count scales linearly with core count (which it doesn't perfectly), Thuban is ~950 million transistors, an 8 core Thuban would be easily be well below the 2 billion that is Bulldozer. Also to my understanding bigger chips are more likely to have yield issues, thus AMD could potentially be getting better yields with a smaller 8 core Thuban.

Potentially this 8 core Thuban would be cheaper to develop. Phenom II architecture is already designed. Shrink, tweak, and make Phenom III.

Also it potentially would be a drop in upgrade to current AM2+/AM3/AM3+ boards (Thuban works on some AM2+ boards) Now this decision may not make motherboard manufacturers happy in the short run, it would go wonders for winning loyalty with customers. Also if your product is a disappointment, you won't sell many chipsets and motherboards anyways.

Also factor in pricing. In this hypothetical world tweaked Thuban IPC > Thuban IPC. Shrunk Thuban clocks > Thuban clocks. 8 cores > 6 cores. This all points towards better performance, which means AMD can charge more, which is good for them.

Factor in the smaller die size, potentially cheaper development costs, better performance, consumer loyalty points, and potentially higher asking price. AMD would be doing a lot better.

Now my knowledge of silicon design is extremely limited so I don't know how much performance they could tweak out of Phenom II. The thing that I keep wondering is, looking at all the resources and time they threw at Bulldozer and the results we currently have; couldn't a fraction of those resources gone towards a shrunk and tweaked Thuban and yielded better results? Also this tweaked Thuban could have potentially come out a lot sooner. Did AMD not realize that Bulldozer was a mess until October?

It just feels like a ridiculous amount of fail took place.

In retrospect, your reasoning makes perfect sense. However, I dont defend AMD often, but they probably felt they would be criticized for not bringing out a new architecture, and felt they had to do it. And I think they would have been criticized for not bringing out a new architecture. The problem is just that they did a bad job. Since I dont know what goes on inside AMD, I cant really say whether they knew far enough ahead how poor Bulldozer would be, to have scrapped or delayed it further and brought out a die shrunk Thuban.

 

[frozentundra123456]

In regards to the OP, I didnt vote because it seems none of the options you gave really fit.
1. Sell bulldozer only-- not a viable option for the present time
2. Sell both lines-- what they will have to do for the time being
3. Die shrink--obvious answer now that the results of Bulldozer are in.

Seems to me that this would have been a more appropriate case study before bulldozer came out. Dont mean to offend you, but just seems sort of late to be planning this strategy. More relavent question for AMD now seems to be how to leverage Llano into more low power markets and if they should pursue ARM.

 

[Kenmitch]

I voted for number 2 before I read your post. Guess it shoulda been number 3

If Bulldozer is trully a shitz design then it should be scrapped. If it's just a matter of some tweaking and another spin or two then carry on.

What AMD does need to do for now is find a niche in the market and exploit it to the fullest.

 

[max347]

1. Sell bulldozer only-- not a viable option for the present time

I included this, because that is what they are doing now- they are shipping BD and slowing/stopping older production

 

[Arachnotronic]

Their stock price to double. Also, if they fix up the bugs in BD, then it'd be fine.

 

[drizek]

They already shrunk the stars core to 32nm, see Llano. The CPU portion of that chip is 32nm stars core.

Considering how long Intel has been shipping CPU's with integrated GPU's, bulldozer should have never happened without a GPU. Trinity should have been the first bulldozer launch with Llano holding AMD over until then.

Is it too cynical to think that Zambezi chips are just Opterons that didn't make the cut?

 

[soccerballtux]

what about "roll over and die" does your business class have a scenario for thoroughly disenfranchised clientele?

 

[Idontcare]

Is it too cynical to think that Zambezi chips are just Opterons that didn't make the cut?

I guess that would be all of them then since I haven't seen an Interlagos SKU for sale yet.

 

[LoneNinja]

We already know Bulldozer has slow cache problems. We've seen from more than one cpu architecture that increasing/decreasing cache sizes won't dramatically impact performance. We know cache impacts power consumption and die size much more than performance.

Personally I wonder what Bulldozer could be with half the cache or less. It should perform similar, have lower power consumption, and become cheaper to sell. They could probably ramp up clock speeds and actually have something decent until they improve core IPC.

 

[Dadofamunky]

This.

How about releasing some nice Opteron chips with 8-12 cores on the consumer side? Seriously, some folks would pay $600-1200 for a 12-core beast that they could throw in a consumer board.

Agree strongly with this. Love to have a cheap VM system like that. I'm thinking about using a Thuban to do that as it is.

 

[sandorski]

The Basic BD design is here to stay. They need to at least do 1 respin of it before deciding to drop it.

 

[Dadofamunky]

Not really. SB IPC is significantly better than Phenom II. You'd have to ignore the incredible overclocking of SB and allow a Phenom II die shrink to be oc'd for it to be anywhere close to a SB competitor.

Phenom II would need an IPC boost.

It'd need a lot more than that to be competitive. Still, there's room in the market for such a thing. BD - what a disaster. I agree with Sandorski though - a respin might fix a lot of problems, particularly the thermals. Sure hope so.

 

[max347]

The Basic BD design is here to stay. They need to at least do 1 respin of it before deciding to drop it.

Yeah, this isn't Thuban arch vs BD arch, its just a theoretical 3 quarter solution until PD

 

[gmaster456]

Not really. SB IPC is significantly better than Phenom II. You'd have to ignore the incredible overclocking of SB and allow a Phenom II die shrink to be oc'd for it to be anywhere close to a SB competitor.

Phenom II would need an IPC boost.

Only 3 SB chips on the consumer end can be overclocked, 2500k, 2600k, 2700k. How is a Shrunken Down Phenom 2 BE for $150 with 90% the performance of a similarly priced i5 2400 NOT competition? Not to mention the phenom can be overclocked where the 2400 can't. Yes the single threaded peformance would be less, and if one relies heavily on single threaded apps, perhaps they should look at SB. But for the Overclocker/gamer a 32nm Phenom ll would make an excellent choice.