AMD Zen (Roadmaps, Quad Core Unit, Block Diagram, 32 Core Opteron)

[patrickjp93]

great, just a little more time and AMD could catch up to intel

Emphasis on "could" please. Let's not forget Intel's pushing the boundaries on many different fronts, and the Knight's Landing chips are nothing to shake a stick at. 36 big AMD cores with 72 threads vs. up to 18 big Broadwell cores or up to 21 big Skylake cores with 36/42 threads or between 60 and 72 "small" (AVX 512 isn't exactly small, and the modified 4-thread Silvermont cores are much larger than Bay Trail and Cherry Trail cores) with 240-288 threads. And Intel's also trying to fight off IBM and gain ground on iGPU for HPC applications as well.

 

AMD is ahead on graphics architecture, but just getting back Keller and Koduri is a far cry from a guarantee AMD will catch up to the company whose very history is one of pushing all known and perceived limits in technology and crushing competition into the ground on raw merit. There are other engineers around the world just as good as those two, and some of them are on Intel's and IBM's payrolls. 

[deviant88]

" The Opteron needs all the silicon space for the L2, L3 cache as well as its Zen x86 cores."

 

The Desktop CPU's need the exact same thing,yet we get apu fucking morons.

[Fetzie]

" The Opteron needs all the silicon space for the L2, L3 cache as well as its Zen x86 cores."

The Desktop CPU's need the exact same thing,yet we get apu fucking morons.

Given that we now have a non-APU chip and an APU chip on the Zen architecture announced for the server market it isn't a stretch to imagine the same for the consumer market.

[Opcode]

Emphasis on "could" please. Let's not forget Intel's pushing the boundaries on many different fronts, and the Knight's Landing chips are nothing to shake a stick at. 36 big AMD cores with 72 threads vs. up to 18 big Broadwell cores or up to 21 big Skylake cores with 36/42 threads or between 60 and 72 "small" (AVX 512 isn't exactly small, and the modified 4-thread Silvermont cores are much larger than Bay Trail and Cherry Trail cores) with 240-288 threads. And Intel's also trying to fight off IBM and gain ground on iGPU for HPC applications as well.

 

AMD is ahead on graphics architecture, but just getting back Keller and Koduri is a far cry from a guarantee AMD will catch up to the company whose very history is one of pushing all known and perceived limits in technology and crushing competition into the ground on raw merit. There are other engineers around the world just as good as those two, and some of them are on Intel's and IBM's payrolls. 

If the validations for XV performance are true then Intel may be up shit creek without a paddle with Skylake. With Keller at the helm Zen will top XV performance along with bring SMT. What can Intel do if Zen is clock for clock faster than Skylake with the desktop skews coming with up to 8 core 16 threads? They have no yellow brick road to follow after that as they would be entirely at the mercy of being influenced by AMD. Pushing Cannonlake to offer at least 8 cores 16 threads on the consumer desktop at an affordable price. It will be K8 vs P68 all over again although maybe AMD will be smart this time and stick to refining the architecture like Intel does instead of dumping it for another.

[patrickjp93]

If the validations for XV performance are true then Intel may be up shit creek without a paddle with Skylake. With Keller at the helm Zen will top XV performance along with bring SMT. What can Intel do if Zen is clock for clock faster than Skylake with the desktop skews coming with up to 8 core 16 threads? They have no yellow brick road to follow after that as they would be entirely at the mercy of being influenced by AMD. Pushing Cannonlake to offer at least 8 cores 16 threads on the consumer desktop at an affordable price. It will be K8 vs P68 all over again although maybe AMD will be smart this time and stick to refining the architecture like Intel does instead of dumping it for another.

It won't happen. Keller is not some infallible God and Intel's not a complacent foe the way IBM was.

[Opcode]

It won't happen. Keller is not some infallible God and Intel's not a complacent foe the way IBM was.

Rumor has it Carrizo seen a quick touch of Kellers hand before they shipped the project off to their Boston facility so they can work on K12/Zen. Sources are starting to show XV being nearly as fast as Haswell clock for clock. Meanwhile that's all coming out of a small 14.48 mm² module.

[Kamina]

AMD using cluster cores again? Does that mean it'll be shit like bulldozer? 

[patrickjp93]

Rumor has it Carrizo seen a quick touch of Kellers hand before they shipped the project off to their Boston facility so they can work on K12/Zen. Sources are starting to show XV being nearly as fast as Haswell clock for clock. Meanwhile that's all coming out of a small 14.48 mm² module.

No they haven't. We've been through this on your thread about Carrizo. There's no concrete info regarding thermal solutions or throttling or actual sustained clock rates. The validations are garbage without this information. Core M looked like crap when Lenovo first used it, but then it turned out it was Lenovo's fault for not developing a proper heatsink. In the new MacBook Core M is faster than Haswell Y by a fair margin. For all we know the one Carrizo test we have is an overclocked golden chip on an active heatsink and the I3s are throttling due to improper cooling solutions. We simply don't know.

[patrickjp93]

AMD using cluster cores again? Does that mean it'll be shit like bulldozer?

I imagine this time around it's more about yields to be able to produce more packages without a monolithic die with lower yields.

[Opcode]

No they haven't. We've been through this on your thread about Carrizo. There's no concrete info regarding thermal solutions or throttling or actual sustained clock rates. The validations are garbage without this information. Core M looked like crap when Lenovo first used it, but then it turned out it was Lenovo's fault for not developing a proper heatsink. In the new MacBook Core M is faster than Haswell Y by a fair margin. For all we know the one Carrizo test we have is an overclocked golden chip on an active heatsink and the I3s are throttling due to improper cooling solutions. We simply don't know.

Out of hundreds of validations I think it's safe to say we can easily establish a median as these numbers will not change after so many validations. Your argument of the Haswell i3 being a "new" chip and not having enough validations from many different vendors is moot. Also the numbers for the 15w FX-8800p were validated at stock clocks so it wasn't overclocked (I haven't even seen the FX-7600P overclocked). Although for all we know the FX validation could be the baseline where it's throttling as it's only a single validation (unlike the Haswell). We can clearly establish Haswell performance although we need more XV validations as they can go either up or down.

[patrickjp93]

Out of hundreds of validations I think it's safe to say we can easily establish a median as these numbers will not change after so many validations. Your argument of the Haswell i3 being a "new" chip and not having enough validations from many different vendors is moot. Also the numbers for the 15w FX-8800p were validated at stock clocks so it wasn't overclocked (I haven't even seen the FX-7600P overclocked). Although for all we know the FX validation could be the baseline where it's throttling as it's only a single validation (unlike the Haswell). We can clearly establish Haswell performance although we need more XV validations as they can go either up or down.

Oh ffs, every laptop is different. Every thermal and electrical solution is different. And it's not super difficult to spoof a benchmark score. It's been done before.

[Opcode]

Oh offs, every laptop is different. Every thermal and electrical solution is different.

Exactly, so how can we say that this is the definite performance of the 15w FX-8800p when it's only one single validation? We have hundreds of validations for Haswell based laptops from dozens of different vendors so it's safe to say we can can establish a median for Haswell performance. Which cannot be said about a single validation.

[Opcode]

Say hello to the Zen quad core unit.

 

 

Original post updated for full coverage of Zen news.

[Deletive]

Say hello to the Zen quad core unit.

 

 

Original post updated for full coverage of Zen news.

Dude you're literally making me like a little child right now.  

This is looking extremely promising, Interested to see the pipeline though. Hope the L3 Cache is better then what I hear it was on Bulldozer.

[patrickjp93]

Say hello to the Zen quad core unit.

 

-snip-

 

Original post updated for full coverage of Zen news.

I find it interesting they doubled the L2 cache size relative to Intel. Maybe Keller found some extra magic sauce to keep cache latencies down despite the size increase?

[Opcode]

I find it interesting they doubled the L2 cache size relative to Intel. Maybe Keller found some extra magic sauce to keep cache latencies down despite the size increase?

What I find interesting is the similarities now between an Intel and an AMD die.

 

[Doobeedoo]

If I can get an 8-core with top notch IPC at solid price at that time I'll be happy

[Sharkyx1]

If I can get an 8-core with top notch IPC at solid price at that time I'll be happy

Well,what is a solid price? Are you thinking i7 4770 performance? Or 5960? 8thread or 8 core Because Intel's current prices are already pretty solid,you used to pay twice as much for the performance chips compared to what you do now.

[That Norwegian Guy]

In before all the hate, Intel does the same thing. They have chips that are just 16-32 atom cores.

 

The biggest Avoton chip is 8 cores.

[Eroda]

snip

cool cool, got an official block diagram for a quad core BD handy? be great to compare

[Dabombinable]

What I find interesting is the similarities now between an Intel and an AMD die.

 

Looking at that, Intel don't use the extra space from removing the iGPU on Xeons to increase the size of the cores, do they?

[Opcode]

cool cool, got an official block diagram for a quad core BD handy? be great to compare

Pretty much the same as Zen other than core changes.

 

[Sharkyx1]

Looking at that, Intel don't use the extra space from removing the iGPU on Xeons to increase the size of the cores, do they?

2011 cores are indeed larger, those are 1150 cores

[patrickjp93]

The biggest Avoton chip is 8 cores.

16. You're thinking Xeon D. Furthermore, Knight's Corner goes up to 61 Atom cores, and Knight's Landing goes up to 72.

[Thony]

Yes, I'm aware of hypocrisy very much (looks at Intel's 16-core)

LOL, love that Amd's chart.