What would make you happiest with AMD?

What would make you happiest with AMD?

  • Sell Bulldozer chips and stop Thuban/Deneb production

    Votes: 11 8.7%

  • Sell both lines

    Votes: 17 13.5%

  • Die shrink the Thuban/Deneb cpus for lower power consumption and higher clocks

    Votes: 98 77.8%
  • Total voters
    126
M

Mr34727

Supreme [H]ardness
Joined
Sep 1, 2008
Messages
4,149
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.
 
I want them to be in business in 5000 years and viewed as intels competitor for every moment of it.
 
8 cores of Thuban would do nothing for us. Give us higher clocked 4/6 cores with more cache, faster NB, etc.
 
You left out "Cut your losses, take extra careful R&D steps going forward, and focus on bringing a COMPETITIVE Piledriver to market ahead of time."
 
Last edited:
can I point out that if the problem lies in the manufacturing end of things, it stands to reason a die shrink wouldn't fix the problems with the process.
 
Phenom III Please. Maybe a few arch tweaks here and there. None of this "Works with REALLY highly threaded programs maybe!" BS with the FX. increase the Phenom IPCPC by 5-10% and shrink the die, turn up the clocks, paint a smile on my face.
 
Keep selling Bulldozer, fix power consumption, reduce redundant cache - make Piledriver awesome.
Phenom III as well - die shrink of Phenom II with optimizations.
 
KazeoHin said:
Phenom III Please. Maybe a few arch tweaks here and there. None of this "Works with REALLY highly threaded programs maybe!" BS with the FX. increase the Phenom IPCPC by 5-10% and shrink the die, turn up the clocks, paint a smile on my face.
Click to expand...

This
 
Dump the FX name...Have Piledriver blow us away, and give it a simple name, with a simple box. Something like the AMD A-10 (geddit?) and sell them in the old style boxes and cheap.
 
Pile Driver and Southern Isles both be beasts. That's all.

No fancy smancy gimmicks. Perhaps Rory knew his PR had failed miserably with BD. Gold painted turd is still a turd.
 
Last edited:
They should do a die shrink of Thuban, and incorporate the same clock gating that went into Llano. But it's too late for that.
 
I'm gonna be honest -- why is everyone focusing on AMD's CPU weakness? CPUs are good enough. It's GPUs that matter, and 28nm AMD stuff will kick massive ass. And on the APU front, I'd much rather have a llano or trinity than a SNB or IVB since both pairs have good CPUs, but only one pair has good GPUs!
 
Mr34727 said:
Can you explain? Llano is fine at 32nm, it isn't the process's fault
Click to expand...

you think they are produced on the same line or have the same needs?

You are probably right, they should just pull the wafers off the llano line, slap some bulldozers through there, that'll learn em.

I am pretty sure bulldozer is the first time for HMKG for gloflo. It's a pretty big deal as I understand it.
 
Last edited:
I didn't vote, because you didn't include my #1 option: for AMD to cease to exist.

;)

No honestly I haven't studied it in-depth to vote, but ultimately I'd like them to improve perf/W and performance in general. That said, die-shrinking Thuban seems like it would be better for them in the short-term (until they make a competitive BD part). Then, transition to BD once real-world workloads evolve to suit BD's multi-threaded nature.

LOL I just realized multi-threaded on desktop is gonna be a niche for the foreseeable future, so release BD once it becomes competitive with Intel, and do Thuban until then.
 
I would rather see them make a die shrink of Thuban at 32nm and give us more cache, after seeing Bulldozer's poor performance I think Thuban is a better CPU, and could be even better if it was produced at 32nm.
 
stealthy123 said:
you think they are produced on the same line or have the same needs?

You are probably right, they should just pull the wafers off the llano line, slap some bulldozers through there, that'll learn em.

I am pretty sure bulldozer is the first time for HMKG for gloflo. It's a pretty big deal as I understand it.
Click to expand...

I ask for an explanation and get sarcasm...thanks bro.

I asked because I see BD having problems with heat/speed/low IPC, but not that many actually defective chips. So, design flaws, rather than process flaws. My Llano comparison was that the 32nm transition isn't the issue, must be something else.
 
To go blow up GloFo and hook up with a fab that can actually produce something!
 
I thought I read a month or two ago that AMD already stopped their Phenom production?
Anyways, I just want their bulldozer to actually be good, I am sure there is no going back now.
 
i would like AMD to stop delaying things and get things right
also i sure wish they would announce the Operation Scorpius winners

as to the poll:sell both lines
 
AMD needs to get back in the game quickly. I vote a revision of Thuban if not's more expensive to do that than to try and fix Bulldozer.
 
Weeth said:
To go blow up GloFo and hook up with a fab that can actually produce something!
Click to expand...

From what I've gathered, AMD initially expected to improve clock speeds going from 45nm > 32nm bulldozer by 30% while maintaining the same TDP. Obviously, the chips aren't clocked anywhere near that mid 4.5ish stock speeds (about a 1ghz bump from Phenom II. This was planned in order to "maintain" the IPC of the Phenom II chips while offering better threaded performance) they were expecting. But to pin the blame on GLoFo here seems misguided. Somehow in the engineers heads they thought that a 4.5ghz chip with a long pipeline and ~2billion transistors would be competitive and cost effective to produce. To fab the chips they were asking was quite simply impossible on 32nm. I don't think anyone can make the chips they initially designed, Intel included.

Yes, GloFo has had trouble with yields on the llano and bulldozer. No, GloFo are not the ones you want to point your finger toward in an attempt to lay the blame. This was first and foremost an engineering and design blunder on AMD's part that was subsequently blamed on GloFo. Mid 4ghz stock speeds with a TDP of ~120 from a chip with around two billion transistors? Get real.
 
My .02

Go and hire some CPU engineers, take the BD cores apart, redesign them by hand and release us the CPU that you have been promising for the last 3 years.

Oh and +1 for selling in plain boxes that are understated but do more than they appear.

Whilst you are doing that, give thuban the same tweaks that you gave llano, whack it on the 32nm process and keep us happy
 
pelo said:
From what I've gathered, AMD initially expected to improve clock speeds going from 45nm > 32nm bulldozer by 30% while maintaining the same TDP. Obviously, the chips aren't clocked anywhere near that mid 4.5ish stock speeds (about a 1ghz bump from Phenom II. This was planned in order to "maintain" the IPC of the Phenom II chips while offering better threaded performance) they were expecting. But to pin the blame on GLoFo here seems misguided. Somehow in the engineers heads they thought that a 4.5ghz chip with a long pipeline and ~2billion transistors would be competitive and cost effective to produce. To fab the chips they were asking was quite simply impossible on 32nm. I don't think anyone can make the chips they initially designed, Intel included.

Yes, GloFo has had trouble with yields on the llano and bulldozer. No, GloFo are not the ones you want to point your finger toward in an attempt to lay the blame. This was first and foremost an engineering and design blunder on AMD's part that was subsequently blamed on GloFo. Mid 4ghz stock speeds with a TDP of ~120 from a chip with around two billion transistors? Get real.
Click to expand...

You make a very good point in that no fab could have successfully built the pie in the sky original BD designs. As for the current FX I'm truly wondering if anything is salvageable. Also a point of interest: I have not seen a single Interlagos benchmark anywhere which is very strange. I wonder if they're having even worse problems!
 
I would be happy with Bulldozer if it had Tri or Quad channel memory.

Honestly AMD's best route would be to Die Shrink Thuban and add a Quad Channel memory controller.
 
pelo said:
From what I've gathered, AMD initially expected to improve clock speeds going from 45nm > 32nm bulldozer by 30% while maintaining the same TDP. Obviously, the chips aren't clocked anywhere near that mid 4.5ish stock speeds (about a 1ghz bump from Phenom II. This was planned in order to "maintain" the IPC of the Phenom II chips while offering better threaded performance) they were expecting. But to pin the blame on GLoFo here seems misguided. Somehow in the engineers heads they thought that a 4.5ghz chip with a long pipeline and ~2billion transistors would be competitive and cost effective to produce. To fab the chips they were asking was quite simply impossible on 32nm. I don't think anyone can make the chips they initially designed, Intel included.

Yes, GloFo has had trouble with yields on the llano and bulldozer. No, GloFo are not the ones you want to point your finger toward in an attempt to lay the blame. This was first and foremost an engineering and design blunder on AMD's part that was subsequently blamed on GloFo. Mid 4ghz stock speeds with a TDP of ~120 from a chip with around two billion transistors? Get real.
Click to expand...

GloFlo made the decision on how to tool the lines, and some other decisions that did indeed effect things you are trying to say they have no effect on. It was not a first and foremost an engineering blunder, despite what you want it to be.

It was a combination of both that screwed the pooch, but a good bit, and probably half or more goes to gloflo.

And intel could make them in a heartbeat, lol that was just a silly comment. And, to point out, the longer pipeline isn't comparable to what intel tried to do with netburst, it's not that much longer than thubans.
 
leeleatherwood said:
I would be happy with Bulldozer if it had Tri or Quad channel memory.

Honestly AMD's best route would be to Die Shrink Thuban and add a Quad Channel memory controller.
Click to expand...

What do you need all that memory bandwidth for?

I am pretty sure the server parts are quad channel.
 
stealthy123 said:
GloFlo made the decision on how to tool the lines, and some other decisions that did indeed effect things you are trying to say they have no effect on. It was not a first and foremost an engineering blunder, despite what you want it to be.

It was a combination of both that screwed the pooch, but a good bit, and probably half or more goes to gloflo.

And intel could make them in a heartbeat, lol that was just a silly comment.
Click to expand...

no, you're right. GloFo probably has to hang its head in shame here, but I don't think that it would pertain to the bulldozer. GloFo didn't design a 2billion transistor chip with stock clocks of over 4ghz, loads of slow cache and a massive pipeline. These things don't scream "GloFo fab failure" but rather "what the hell were the engineers thinking?"

No, not even your mighty god Intel could produce that chip on 32nm.

I understand your argument that GloFo couldn't produce what AMD wanted/need, but if you look at what AMD initially designed the specs were too far out of reach, especially for a relatively new 32nm process at GloFo. And even if they did the IPC would still be lower than an equivalently or cheaper priced Sandy Bridge alternative.
 
of course they could, and by next year so could gloflo.

intel could probably walk into gloflo right now and tell them everything they are doing wrong, that ain't gonna happen.
 
stealthy123 said:
of course they could, and by next year so could gloflo.

intel could probably walk into gloflo right now and tell them everything they are doing wrong, that ain't gonna happen.
Click to expand...

By next year, IB is going to be top dog of the silicon universe so they'll be behind the 22nm 8ball, a process that GloFo has just about as much likelihood of mastering as I do of talking Blake Lively and Jessica Alba into a threesome. Lots a luck, GloFo. Or should I say R.I.P.
 
lots of good arguments here.

i'll echo for the short term a revision of deneb/thuban would be optimal. Those chips were actually better than a lot of people give them credit for, some tweaks and proper pricing could certainly make them competitive with a decent chunk of intel's lineup. besides, that chip is a product that had many years to mature and be revised...if it isn't broke....just make it a little better.

during that time they could do as stated, eat the cost and bring some top talent in to take apart bulldozer and start a redesign for it's next incarnation. I see what they wanted, and i agree the initial ideas of bulldozer are probably a bit too much to ask...but i think the compromise can be better than what was actually delivered...
 
Weeth said:
By next year, IB is going to be top dog of the silicon universe so they'll be behind the 22nm 8ball, a process that GloFo has just about as much likelihood of mastering as I do of talking Blake Lively and Jessica Alba into a threesome. Lots a luck, GloFo. Or should I say R.I.P.
Click to expand...

Pretty much. I have a hard time believing GoFlo would be capable of doing trigate on 22nm.

From here on, Intel's magic is in their manufacturing. Even if AMD was able to make a faster CPU, there's no "light at the end of the tunnel" for them with regards to power usage. We're looking at 14nm in 2014 for Broadwell.
 
Rather than introducing bulldozer in such a mediocre state, they shuold have waited for the next version (piledriver) with higher clocks.

MEANWHILE, they should have revised Thuban with the new "Husky" (1mb L2) cores, added per-core overclocking, doubled the L3 bandwidth, shrunk it to 32nm, and released that as Phenom III on the existing AM3 platform.
 
They need to devote their resources to improving Bulldozer (Piledriver etc).
A die shrink of Thuban would be a waste of resources since it's a dead end architecture.

It would be best to keep on producing Deneb/Thuban until a few months before Piledriver is released.
 
You must log in or register to reply here.
Back
Top