If AMD were to release a new hexcore chip what should its specs and die layout be?

[cbn]

As I see things, two big weaknesses of AM3+ and FM2+ are huge amount of cache for the former and too large of an igpu for the latter.

If AMD can fix both those problems with a new hexcore die I think the cost would be low enough for it to be competitive.

I also think hexcore would be a good sweet spot for AMD (one reason being with six cores the chances for a dual core SKU are extremely low)

The question is what direction do they go?

1.) Do they design the hexcore for both desktop and higher end laptops (using dGPU)? Or do they go strictly for desktop with the hexcore?

2. What uarch do they use? Streamroller (which proven capability for high clocks)? Or do they use Excavator (which might clock lower due to high density libraries, but might be better for notebook using dGPU)?

In any event, I would hope for either no iGPU (integrated graphics on chipset are fine) or a very minimal iGPU (64 GCN stream processors maximum).

 

[Shehriazad]

I don't think "no iGPU" is the direction AMD will ever go again. I expect Zen to also come with iGPUs....simply because going toward SoCs is the right thing to do.

Just that AMD is too early to the party again with hardware that is too weird.

Also Excavator and HDL are not in need of eachother...you can make Excavators without HDL...AMD simply chose HDL because they're not ready to leave 28nm yet but needed space for their SoC style implementations.


But if you want to see a Hexacore...it will be neither Steamroller nor Excavator..it's Zen. (Granted it will still come)

If they HAD to make a Hexacore right now...it would be a 28nm Excavator chip with a proper L2 cache and no iGPU...but that's just daydreaming.

 

[cbn]

I don't think "no iGPU" is the direction AMD will ever go again. I expect Zen to also come with iGPUs....simply because going toward SoCs is the right thing to do.

I get the idea about iGPUs and Integrated graphics on chipset being effective cost reducers. However, with AMD being so far behind Intel with respect to manufacturing node I really question the idea of having so much iGPU. In some ways, I feel like it is really hurting AMD to try and do this.

In fact, I actually believe having too much iGPU could eventually be one of Intel's weaknesses (unless they make a return to having dedicated GT1 dies).

 

[cbn]

But if you want to see a Hexacore...it will be neither Steamroller nor Excavator..it's Zen. (Granted it will still come)

This discussion could also include Zen (assuming we ever see it).

 

[ShintaiDK]

How many threads do we need about your AMD hexcore dreams without IGP?

AMD is done in the desktop segment as well.

 

[cbn]

How many threads do we need about your AMD hexcore dreams without IGP?

AMD is done in the desktop segment as well.

They have new leadership, lets give them a chance to fix the problems/deficiencies instead of giving up on them.

 

[tential]

They have new leadership, lets give them a chance to fix the problems/deficiencies instead of giving up on them.

Lets wait for them to announce that they're interested in pursuing the desktop market before theorizing day in and day out on something they've clearly abandoned.

 

[ShintaiDK]

They have new leadership, lets give them a chance to fix the problems/deficiencies instead of giving up on them.

A hexcore without IGP isnt a solution, its a problem. And that management said goodbye to the desktop.

 

[cbn]

A hexcore without IGP isnt a solution, its a problem.

Read my opening post again.

 

[Erenhardt]

Anyone have enough knowledge to confirm or deny if FM2+ can support 3 module SR or Excavator cpu?

 

[ShintaiDK]

Anyone have enough knowledge to confirm or deny if FM2+ can support 3 module SR or Excavator cpu?

The socket isnt the limitation. Excavator isnt happening. So it would have to be SR based if any.

The only question is, how big of a die can you fit under the IHS on a FM2+ package. And how is that die usage distributed.

 

[NTMBK]

Isn't an FM2 package the same size as AM2/2+/3/3+? So at least 8 cores.

 

[SlowSpyder]

How many threads do we need about your AMD hexcore dreams without IGP?

AMD is done in the desktop segment as well.

The socket isnt the limitation. Excavator isnt happening. So it would have to be SR based if any.

The only question is, how big of a die can you fit under the IHS on a FM2+ package. And how is that die usage distributed.

There isn't going to be an Excavator core based APU/CPU?

 

[sm625]

When you say hexacor do you mean 6 FPU? ie six dual int modules? (12 threads?) If they did such a thing then I think they should reduce the L2 cache. Too much of their die was wasted on cache. If their caches were fast, this would have been a godsend, but unfortunately they are not fast at all, so their die space would be better spent on additional execution units. Hell, with 256KB L2, they could probably fit 6 modules in the same space that is currently occupied by four.

 

[ShintaiDK]

There isn't going to be an Excavator core based APU/CPU?

In an FM2+ socket? No. BGA only.

In case you missed it:
http://www.anandtech.com/show/8871/...he-desktop-depends-what-you-define-as-desktop

 

[SlowSpyder]

In an FM2+ socket? No. BGA only.

In case you missed it:
http://www.anandtech.com/show/8871/...he-desktop-depends-what-you-define-as-desktop

I didn't see that, thanks. Doesn't sound like anything is confirmed, but AMD not mentioning anything beyond the AIO's sure does point to that direction. I guess we'll have to wait and see how they sell these, but it looks like AIO desktops which aren't something gamers / enthusiasts typically want anything to do with.

 

[Shivansps]

AMD already has hexacores in servers... there just no need to bring them to desktop.

I just dont see the point.

 

[cbn]

The socket isnt the limitation. Excavator isnt happening. So it would have to be SR based if any.

Carrizo isn't happening for FM2+.

But if we are talking a new six core die, it could certainly be Excavator.

The question is six Steamroller cores vs. six Excavator cores (with or without HDL)? I would think Excavtor with HDL (rather than without HDL) would be more likely due to cost reasons.

 

[cbn]

When you say hexacor do you mean 6 FPU? ie six dual int modules? (12 threads?)

No, I mean 3 modules, 6 cores/6 threads.

 

[cbn]

AMD already has hexacores in servers... there just no need to bring them to desktop.

I just dont see the point.

AMD has Piledriver based FX-6300 and FX-6350 hexcores, but these are based on 315mm2 32nm octocore dies with 8MB L3 cache taking up a good amount of die space.

The performance is fairly decent, but I feel these will be too expensive to manufacture going forward.

 

[Phynaz]

AMD already has hexacores in servers...

Yep, both of them :twisted:

 

[cbn]

(snip) If they did such a thing then I think they should reduce the L2 cache. Too much of their die was wasted on cache. If their caches were fast, this would have been a godsend, but unfortunately they are not fast at all, so their die space would be better spent on additional execution units. Hell, with 256KB L2, they could probably fit 6 modules in the same space that is currently occupied by four.

I noticed Haswell only has 256 KB of L2 cache per core. Haswell, of course, does have L3 like the AM3+ chips. (However, the AM3+ chips also had a whopping 6MB L2 cache on the four module SKUs.) EDIT: The six core AM3+ Piledriver chips also have 6MB L2 cache listed.

 

[cbn]

So at this point, does AMD go hexcore for FM2+ or AM3+?

If FM2+, Does the hexcore die also get positioned for use in higher end laptops (with dGPU)?

Lastly, if FM2+, how many stream processors should the iGPU have on this hexcore die?

 

[Shehriazad]

So at this point, does AMD go hexcore for FM2+ or AM3+?

If FM2+, Does the hexcore die also get positioned for use in higher end laptops (with dGPU)?

Lastly, if FM2+, how many stream processors should the iGPU have on this hexcore die?

Of course it would go FM2+...they cannot make a single new chip for AM3+...that would totally destroy what little revenue they have with AM3+...sure that Hexacore chip would sell...but the dozens of other chips won't.

IF the Hexacore was done with HDL and no more than 1mb L2 cache per module then the iGPU would be about the same specs (or slightly higher) like Kaveris' higher end with some updated tech.


That said...with HDL and without iGPU AMD could probably fit a 10-12 core (5-6 modules) on there...since there are 95W/100w from the socket and HDL Excavator being rather low powered (in theory) that would probably be the better idea since the cores can't clock too high, anyway. If they ever tried to make a novelty multicore on FM2+ they would go all the way. 12 core (6 modules) at about 3.3 GHZ base and 3.5GHZ boost at a tdp of 95w.

 

[cbn]

IF the Hexacore was done with HDL and no more than 1mb L2 cache per module then the iGPU would be about the same specs (or slightly higher) like Kaveris' higher end with some updated tech.

So you think 512 Stream processors is doable at the same die size as Kaveri.

The question then is should AMD go for large iGPU on hexcore for FM2+ or go very small on iGPU for hexcore on FM2+?

IMO If AMD went very small on the hexcore iGPU it would have a much smaller die size (than Carrizo) without impacting the targets of desktop and notebook with dGPU.

So Carrizo for highly integrated lower end gamer notebook and Hexcore with very small iGPU for FM2+ desktop and higher end gamer notebook with dGPU? Does that make sense? Or do you have a differing opinion?

P.S. Regarding higher end gaming notebook with dGPU, we'd also have to think about how that hexcore would compete against Intel's offerings.