AMD Prepares ‘Bristol Ridge’ APU (Carrizo) For Desktop

[vm'N]

Do you think amd will try to modify x86 into a frankenstein enough design to make CMT/SMT be better than SMT? not with their money they wont

I doubt AMD would implement both CMT + SMT, unless they don't have any bigger plans for improving the individual thread performances.

[LukaP]

I doubt AMD would implement both CMT + SMT, unless they don't have any bigger plans for improving the individual thread performances.

They have plans to compete with Intels IPC by 2016, which means they have to go SMT (CMT inherently will have a lower IPC with the same technology) to reach it.

[vm'N]

They have plans to compete with Intels IPC by 2016, which means they have to go SMT (CMT inherently will have a lower IPC with the same technology) to reach it.

I doubt they have plans to compete on IPC.

Again, AMD will use the moar core to leverage the lower core thread performance.

Edited January 29, 2015 by vm'N

[LukaP]

I doubt they have plans to compete on IPC.

Again, AMD will use the moar core to leverage the lower core performance.

no, because that would cost them their life, and they know it, as does keller. They are making an 8core chip because they will be on 14nm (im talking Zen here) and will have the space to do it. Im pretty sure they can manage to get their ipc back up like they used to be

[vm'N]

no, because that would cost them their life, and they know it, as does keller. They are making an 8core chip because they will be on 14nm (im talking Zen here) and will have the space to do it. Im pretty sure they can manage to get their ipc back up like they used to be

No, it would not.

They will make 8 core to compete against 4 core skylake. Or they would just allocate more area to the iGP or make the chips smaller.

Get their IPC back up? Yes.

On the level as skylake? No.

[patrickjp93]

No, its power-efficiency comes when you actually have software that can leverage on the highly parallized processing power of the powerarchitecture.

Where do you get 80 watts from?

Intel actually makes two x86 architectures. However Intels core architecture have been designed to cover more markets.

This does have the impact that the architecture wont be tailored to a specific market.

Three if you count Edison/Galileo.

Also, there are still a number of single-threaded tasks Power 8 wins at.

[LukaP]

No, it would not.

They will make 8 core to compete against 4 core skylake. Or they would just allocate more area to the iGP or make the chips smaller.

Get their IPC back up? Yes.

On the level as skylake? No.

Lets agree to disagree here. And im pretty sure Intel will go 6core on the 6770k. its about time they add some cores to their 115x i7 range now.

[Opcode]

K12 is not been designed to take over the x86 server market.

There have been no signs to indicate that.

Then maybe you can enlighten us as to what AMD will be using K12 for. Keeping in mind that AMD officially calls Zen the K12's "sister core".

[vm'N]

Lets agree to disagree here.

Agreed.

[patrickjp93]

No, it would not.

They will make 8 core to compete against 4 core skylake. Or they would just allocate more area to the iGP or make the chips smaller.

Get their IPC back up? Yes.

On the level as skylake? No.

Zen' IPC doesn't have to match Intel's 1:1 to compete, but a $500 chip with 85% the performance of the 5960X would make a lot of people switch to AMD in a heartbeat. I'm hoping we see on-package or on-die HBM since Micron's Hybrid Memory Cube seems to finally be finished. For Skylake's successor, which may be a 14nm architecture if 10nm is delayed too much, Intel could tweak the memory controller to use it.

[vm'N]

Then maybe you can enlighten us as to what AMD will be using K12 for.

I rather wouldn't, as there aren't enough information at this time.

If you would like to make predictions, make sure your statements indicates that.

Keeping in mind that AMD officially calls Zen the K12's "sister core".

I don't see how that is relevant?

[patrickjp93]

Lets agree to disagree here. And im pretty sure Intel will go 6core on the 6770k. its about time they add some cores to their 115x i7 range now.

For Skylake? Doubt it. Cannonlake? At that point they'll be running out of excuses.

[patrickjp93]

I rather wouldn't, as there aren't enough information at this time.

If you would like to make predictions, make sure your statements indicates that.

I don't see how that is relevant?

AMD has said K12 is meant to be how they stir up the server world. If they can get enough perf/space, perf/watt, and perf/cost, ARM will take over very quickly.

[LukaP]

For Skylake? Doubt it. Cannonlake? At that point they'll be running out of excuses.

oh yes, i meant cannonlake... the 10nm one that is since they will have all the space for them

 

also, i think that 5xxxk will be skylake anyway (skipping broadwell K) since its supposed to be released at a similar timeframe

[vm'N]

Zen' IPC doesn't have to match Intel's 1:1 to compete, but a $500 chip with 85% the performance of the 5960X would make a lot of people switch to AMD in a heartbeat. I'm hoping we see on-package or on-die HBM since Micron's Hybrid Memory Cube seems to finally be finished. For Skylake's successor, which may be a 14nm architecture if 10nm is delayed too much, Intel could tweak the memory controller to use it.

That $500 chip should pretty much already exist.

Just not from AMD.

[Opcode]

I rather wouldn't, as there aren't enough information at this time.

If you would like to make predictions, make sure your statements indicates that.

I don't see how that is relevant?

That's the proof in the pudding that K12 is being designed as a high performance architecture. Not something you would use for cell phones or tablets (what Amur is for).

[patrickjp93]

oh yes, i meant cannonlake... the 10nm one that is since they will have all the space for them

also, i think that 5xxxk will be skylake anyway (skipping broadwell K) since its supposed to be released at a similar timeframe

Broadwell K is next Quarter. Skylake is 2H, though I'm guessing mid Q3, and Skylake K is not coming until 2016.

[LukaP]

Broadwell K is next Quarter. Skylake is 2H, though I'm guessing mid Q3, and Skylake K is not coming until 2016.

We will see, i just dont see any use in Broadwell K... even tho intel has been known to milk its users alot

[patrickjp93]

That $500 chip should pretty much already exist.

Just not from AMD.

No, no it wouldn't. The 5930K is not 85% the performance of the 5960X. It's about 75% and costs $700. The 5820K? Now you're dropping PCIe lanes.

[vm'N]

That's the proof in the pudding that K12 is being designed as a high performance architecture. Not something you would use for cell phones or tablets (what Amur is for).

Never indicated that the K12 would be a phone architecture.

But it doesn't indicate that it is to take over the x86 server market.

[patrickjp93]

We will see, i just dont see any use in Broadwell K... even tho intel has been known to milk its users alot

It's purely an enthusiast toy and a debut of their graphics architecture on desktop. People will overclock it and see how it does, and Intel will get some much needed feedback from the gaming world.

[Opcode]

Never indicated that the K12 would be a phone architecture.

But it doesn't indicate that it is to take over the x86 server market.

It's apparently slated for AMD's come back into the server industry. Now I'm not saying it will take over the x86 ecosystem tho that's apparently what AMD's plans are for it.

[patrickjp93]

Never indicated that the K12 would be a phone architecture.

But it doesn't indicate that it is to take over the x86 server market.

I think you confuse AMD's hope for the product(which @Opcode is referring to) with a definitive use/performance of it. That said, In this space where power is the same King as performance, ARM is becoming a threat.

[vm'N]

AMD has said K12 is meant to be how they stir up the server world. If they can get enough perf/space, perf/watt, and perf/cost, ARM will take over very quickly.

Regain traction in the server market. Not take over.

 

No, no it wouldn't. The 5930K is not 85% the performance of the 5960X. It's about 75% and costs $700. The 5820K? Now you're dropping PCIe lanes.

Implying perfect core-scaling?

What about the frequencies differences?

As if PCIe lanes matter to this discussion.

[vm'N]

It's apparently slated for AMD's come back into the server industry. Now I'm not saying it will take over the x86 ecosystem tho that's apparently what AMD's plans are for it.

Where does it show that it is AMD plans/hopes to take over the x86 server world with K12? Edited January 29, 2015 by vm'N