Anyone know the specs of AMD Zen CPU's

[lolth1]

Anyone know the specs of AMD Zen CPU's

 

Thanks!

[LokiFire]

No idea on how the actual architecture is AFAIK.

[Glenwing]

No one here. You'll only find speculation. Even if we knew the specs it wouldn't really tell us anything about performance anyway. We learned that with Bulldozer.

[The_Stig]

No, we dont know what Zen will bring, that includes the specs. We can make educated guesses, but thats all (unless you own a time-travelling delorean)

[Sam Z Man]

If there claims are to be believed the IPC will be about as good as ivy bridge.

[looncraz]

Anyone know the specs of AMD Zen CPU's

 

Thanks!

 

From what I've been able to gather, from only the most reliable of sources:

 

10 pipelines:

4x ALU (1x mult, 1x div, 4x add/sub, etc.)

2x AGU (both fully store/load capable)

4x FPU (non-uniform, gangable, units)

 

Four fully-capable decoders (akin to Bulldozer's decoders, but updated, so about ~20% faster than was is needed to *support* Haswell IPC).

Most of the latest instructions sets (except AVX512, which is not a loss).

 

Some form of SMT (such as Intel's Hyper-Threading).

 

Target TDP: 95W.  For which configuration was never stated.  If this is the 8-core CPU, that's astonishing (or possibly quite worrying), if it's for the quad core with SMT (akin to a 4790k, for example), then it's an improvement, but not enough to match Intel - but enough to make it non-issue if performance is identical.

 

Smaller, faster, L2 (512KB rumored).  No information on L1 or L3, aside from some rumors about the L3 being shared in a quad-core group, however.  L1 cache is probably larger and faster at the same time.  My guess is that cache associativity has been increased.

 

Clockrate is a complete wildcard.  AMD knows how to make high-clocking CPUs, but on a new process that isn't tailored for it designed becomes extremely critical and you have to give up IPC for clockrate.

 

Going from here everything would just be speculation.

[lolth1]

From what I've been able to gather, from only the most reliable of sources:

 

10 pipelines:

4x ALU (1x mult, 1x div, 4x add/sub, etc.)

2x AGU (both fully store/load capable)

4x FPU (non-uniform, gangable, units)

 

Four fully-capable decoders (akin to Bulldozer's decoders, but updated, so about ~20% faster than was is needed to *support* Haswell IPC).

Most of the latest instructions sets (except AVX512, which is not a loss).

 

Some form of SMT (such as Intel's Hyper-Threading).

 

Target TDP: 95W.  For which configuration was never stated.  If this is the 8-core CPU, that's astonishing (or possibly quite worrying), if it's for the quad core with SMT (akin to a 4790k, for example), then it's an improvement, but not enough to match Intel - but enough to make it non-issue if performance is identical.

 

Smaller, faster, L2 (512KB rumored).  No information on L1 or L3, aside from some rumors about the L3 being shared in a quad-core group, however.  L1 cache is probably larger and faster at the same time.  My guess is that cache associativity has been increased.

 

Clockrate is a complete wildcard.  AMD knows how to make high-clocking CPUs, but on a new process that isn't tailored for it designed becomes extremely critical and you have to give up IPC for clockrate.

 

Going from here everything would just be speculation.

 

Thanks for the info. Me personally I use Intel all the time for my builds and builds for my friends. But I really want to see AMD step up their game and become really competitive so it will drive Intel to either lower prices or make better CPU's because you know that they are just taking it "easy" because they do not have much competition. (If you know what I mean) 

[King_Tuna]

No they havent released full details about them.