TR Forums

Well folks, the recent release of more details regarding Trinity (just today) had me thinking about Bulldozer, and AMD's future iterations of it. We all know that Bulldozer came out with guns blazing only to know later that it was firing plastic pellets. Thing is, Piledriver looks to be in good shape, and Steamroller and Excavator will hopefully build further on the BD architecture. Haswell is coming next year, Broadwell in 2014, Skylake in 2015, Skymont in 2016. Whatever will succeed Skymont probably has been started already as well if Intel plans to start selling it in 2016, unless they plan to just keep evolving the current architecture. As for AMD, the same thing applies. Unless they plan to evolve Bulldozer further down the road (i.e. evolve Bulldozer into a sleek, light, fast Ferrari!), they had better have started Project X years ago already.

Purely speculation. Should be fun.

AMD will certainly evolve bulldozer but I'm inclined to think that if they are working on another whole new architecture that it has only been started after bulldozer taped out and they realized that it sucked so bad. It will be interesting to see if bulldozer can be evolved into something great or if it was a totally misguided idea thatAMD eventually abandons.

Possibly more interesting to see is if AMD rolls out something ARM based.

Before thinking that Piledriver is in good shape, we need to see the following:

Quad-core Trinity @ clock X
FX-41xx @ same clock X
Control: Deneb (unlocked variety) @ same clock X

All quad, set at the same clock, then benchmark them. Then we will see what improvements have been made, and by how much. I have a horrible feeling (after reading numerous predictions) that clock for clock IPC improvements between Piledriver and Bulldozer will only be sub 5% (like only 3%). I have a feeling that AMD will be relying heavily on being able to clock Piledriver higher than Bulldozer (because of supposed leakage repairs), in order to bump overall IPC into double digits (10-15% overall). And my worst feeling of all: Piledriver at best, will only be able to equal Deneb/Thuban, and that's still with clocks about 1GHz higher than the previous gen (Deneb/Thuban). I am of course thinking of single and lightly threaded applications.

My only hope is that they've patched up the electrical leakage issue enough, so that it can be pushed to 5GHz, without consuming insane amounts of electricity. That way, it should be able to be competitive, if not finally be able to squeak past, in lightly threaded apps (at least, hopefully, with Deneb/Thuban).

Anyway. I think we need to see the above benchmark scenario in order to really get a good picture of what Piledriver brings to the table.

I think bulldozer can be evolved into something great. The front end dispatch needs to be improved (to be able to dispatch to both "int threads" at the same time per core and FPU), the FPU needs to be accessible by both threads in the core at the same time even if they only get 50% of the FPU, they could widen the FPU by say 50%, and improve the L3 cache.

I think if they do those things it would be a good improvement all around, assuming I am remembering all things bulldozer correctly. Hell with the size of the GPU they are putting in piledriver I would almost consider getting that as a general platform IF they can improve the single threaded performance on the CPU side.

sircharles32 wrote:

I have a feeling that AMD will be relying heavily on being able to clock Piledriver higher than Bulldozer (because of supposed leakage repairs), in order to bump overall IPC into double digits (10-15% overall).

IPC=instructions per clock=not affected by clock speed.

Fine, then I must have meant "performance" instead of IPC, when referring to frequency increases. Happy?
You know what I was getting at by the context.

sircharles32 wrote:

Fine, then I must have meant "performance" instead of IPC, when referring to frequency increases. Happy?
You know what I was getting at by the context.

It was fair enough to point it out. You get it, but it may have just been confusing others.

Things could go either way for them. I hope they manage to recover from the Bulldozer mis-step and continue to offer an alternative to Intel.

ARM is an interesting wrinkle. As the world becomes less PC-centric, ARM's role will continue to grow as our dependence on x86 for everyday computing shrinks. However, AMD is rather late to the ARM party; I don't know whether they are *too* late to make money at it, but it seems to me the best they can hope for is to become just another "me too" ARM vendor among many. Not exactly a fitting legacy for the last man standing among alternative x86 vendors...

sircharles32 wrote:

I have a horrible feeling (after reading numerous predictions) that clock for clock IPC improvements between Piledriver and Bulldozer will only be sub 5% (like only 3%). I have a feeling that AMD will be relying heavily on being able to clock Piledriver higher than Bulldozer (because of supposed leakage repairs), in order to bump overall [performance] (ed) into double digits (10-15% overall).

Agreed; I'm expecting that when AMD says there's a 10-15% performance increase, they mean at a given price point and not at a given clock speed.

just brew it! wrote:

but it seems to me the best they can hope for is to become just another "me too" ARM vendor among many.

They could license GPUs and SeaMicro designs. Play the "me too" game, but on the opposite side of the table.

Who knows, maybe they'll even come up with their own CPU architecture after a while? They certainly have a few years to toy around with that before it would become an absolute necessity.