(Rumour) - Zen chips have already been tested, met all expectation, no significant bottlenecks

[TechGod]

No bottlenecks=sandybridge. That's not really anything to be proud of. 

Did you EVEN research how Zen is gonna work? 14nm node and SMT pretty much confirms it WILL be better than Sandy bridge. Expect Haswell and maybe Skylake performance. 

[Notional]

oh hai patrick... since you are here. with a source too.

 

@MageTank

PATRICK POSTED A LINK... we must write this down in the history-books.

 

 ​

[Tomsen]

Did you EVEN research how Zen is gonna work? 14nm node and SMT pretty much confirms it WILL be better than Sandy bridge. Expect Haswell and maybe Skylake performance.

You are both making a ton of assumptions.

How does no major bottleneck relate to sandys performance?

How does 14nm and SMT confirm its performance?

[Prysin]

Did you EVEN research how Zen is gonna work? 14nm node and SMT pretty much confirms it WILL be better than Sandy bridge. Expect Haswell and maybe Skylake performance. 

in theory, if all AMDs claims are true. it CAN be better then skylake in single core performance.

 

in reality, i would say it will be between Ivy Bridge and Haswell.

 

Zen+ should be around skylake

[TechGod]

 

You are both making a ton of assumptions.

 

How does no major bottleneck relate to sandys performance?
How does 14nm and SMT confirm its performance?

Because one FPU unit isn't being shared between two cores. That was a major bottleneck with Bulldozer.
14nm is gonna help with maxing performance and efficency.

 

 

in theory, if all AMDs claims are true. it CAN be better then skylake in single core performance.

 

in reality, i would say it will be between Ivy Bridge and Haswell.

 

Zen+ should be around skylake

I expect it to be between Haswell and Skylake. Hopefully.

[Dabombinable]

"significant bottlenecks"....

[Tomsen]

Because one FPU unit isn't being shared between two cores. That was a major bottleneck with Bulldozer.

14nm is gonna help with maxing performance and efficency.

Now, it is shared between two threads.

There were good points and negative points in regards to sharing the FPU.

I wouldn't call it a major bottleneck. For general use, it is sufficient, and FPU operations aren't that heavy in demand for general usage.

[Prysin]

I expect it to be between Haswell and Skylake. Hopefully. 

well you see it is ALL based on excavator.

 

AMD claims Zen is 40% faster then excavator.

They claim Excavator is 30% faster then steamroller.

 

however THERE IS NO PROOF that Excavator IS 30% faster then Steamroller.

Following AMDs history of predicting their own IPC, they have always shot for the moon, and gotten halfway....

 

So if excavator is only 15% faster in single core then Steamroller. that will automatically place ZEN slightly below Haswell.

However, if we apply similar logic to ZEN... that anything AMD claims is about twice that of actual performance... then ZEN will be 20% faster then excavator...

 

ive done this math a few times around this fourm.. Bottom line is;

at this moment, FX is 80% slower in single core then Skylake.

 

let that sink in... 80%

 

So thinkin that AMD will catch up to skylake. Litterally improving by 80%, is beyond ridiculous.

If there was any chance for AMD to reach such a feat, Intel would have made moves already. Intel wouldnt have delayed Cannonlake. they would have pushed hard for it.... Because if AMD can catch up to Skylake. They can catch up to Kaby Lake... and THAT is bad news for intel.

 

If you think Intel doesnt know roughly how ZEN will actually perform by now, then you are naive. These companies spy on eachother. That is how they stay ahead and undercuts eachother....

All major companies do this in some way.

[Tomsen]

We have to be careful in how we intepret this news.

Basically, this means that zen is not an unbalanced micro-architecture. Doesn't really say anything about performance alone.

You could try to "predict" performance, by going off with actual released information for some factors to performance.

[TechGod]

 

well you see it is ALL based on excavator.

 

AMD claims Zen is 40% faster then excavator.
They claim Excavator is 30% faster then steamroller.

 

however THERE IS NO PROOF that Excavator IS 30% faster then Steamroller.
Following AMDs history of predicting their own IPC, they have always shot for the moon, and gotten halfway....

 

So if excavator is only 15% faster in single core then Steamroller. that will automatically place ZEN slightly below Haswell.
However, if we apply similar logic to ZEN... that anything AMD claims is about twice that of actual performance... then ZEN will be 20% faster then excavator...

 

ive done this math a few times around this fourm.. Bottom line is;
at this moment, FX is 80% slower in single core then Skylake.

 

let that sink in... 80%

 

So thinkin that AMD will catch up to skylake. Litterally improving by 80%, is beyond ridiculous.
If there was any chance for AMD to reach such a feat, Intel would have made moves already. Intel wouldnt have delayed Cannonlake. they would have pushed hard for it.... Because if AMD can catch up to Skylake. They can catch up to Kaby Lake... and THAT is bad news for intel.

 

If you think Intel doesnt know roughly how ZEN will actually perform by now, then you are naive. These companies spy on eachother. That is how they stay ahead and undercuts eachother....
All major companies do this in some way.

We'll see about that.

 

 

Now, it is shared between two threads.
There were good points and negative points in regards to sharing the FPU.
I wouldn't call it a major bottleneck. For general use, it is sufficient, and FPU operations aren't that heavy in demand for general usage.

Then what else caused it to perform so terribly? Surely having to share resources between two cores would be a significant bottleneck.

[VagabondWraith]

I think we all know once AMD releases Zen, Intel will release something 3-4 times faster thus sending AMD right back where they are... today.

[Inheritance]

Zen has to be some what decent or AMD is going to be in a metric shit ton of trouble.

[Tomsen]

Then what else caused it to perform so terribly? Surely having to share resources between two cores would be a significant bottleneck.

There are many factors to its bad performance.

It is not as simple as pointing out a single reason.

Basically, it was lacking in to many areas, giving great under-utilization of the cores.

Caches are often shared between cores. Sharing resources doesn't mean it will give significant worse performance.

[Inheritance]

I think we all know once AMD releases Zen, Intel will release something 3-4 times faster thus sending AMD right back where they are... today.

Maybe not 3-4 times faster, Maybe 1.5-2. But yes Intel has to be sitting on faster chips considering their last few gens have given us nothing to be shocked by. 

[VagabondWraith]

@Prysin

I agree Intel knows what Zen is and its performance. Companies do keep tabs on the inside of one another.

Another example; nVidia knew all about the Fury X, thus releasing the 980 Ti to get to the market first. I believe nVidia will also be the ones to release a dual flagship GPU first as well. Smart companies know when to strike.

[TechGod]

There are many factors to its bad performance.

It is not as simple as pointing out a single reason.

Basically, it was lacking in to many areas, giving great under-utilization of the cores.

Caches are often shared between cores. Sharing resources doesn't mean it will give significant worse performance.

Right, but the FPU would be a pretty important component of the CPU that is being shared right? Its more necessary than cache, right?

[Tomsen]

Right, but the FPU would be a pretty important component of the CPU that is being shared right? Its more necessary than cache, right?

Yes, it is a rather important component, however, most FPU's to this day (even bulldozer) is fine for the average user.

Caches is more of an important area. It is a big factor to performance.

[TechGod]

Yes, it is a rather important component, however, most FPU's to this day (even bulldozer) is fine for the average user.

Caches is more of an important area. It is a big factor to performance.

AMD seems to be going all out. 

http://wccftech.com/amd-zen-cpu-core-microarchitecture-detailed/2/

 

Would all the changes they've made boost performance? 

[Ivo]

Weren't there similar rumors for Bulldozer? And we all know how that train wreck ended.

 

@Arcanekitten The quote in your signature is literally invisible to Night Theme users, might want to remove the coloring on that text.

yeah, but bulldozer made intel launch Sandy Bridge which was amazing, hopefully intel shits their pants and finally gives us a worthwhile upgrade. 

[Tomsen]

AMD seems to be going all out. 

http://wccftech.com/amd-zen-cpu-core-microarchitecture-detailed/2/

 

Would all the changes they've made boost performance?

Yes, it is going to be rather interesting to see how it actually gonna perform.

Again, this is very limited information, and only gives few factors to performance. Many factors are still unknown.

However, this shows that zen is a very wide core. It is clear AMD is going for higher IPC.

I think it will do rather good in general applications. If single-thread performance lands within ivy-haswell, would mean that Intels big selling point, better IPC, would be irrelevant.

[Dabombinable]

I still can't believe that K12 will be ARM.

[Tomsen]

I still can't believe that K12 will be ARM.

There have been surprisingly little information in regards to K12.

[MageTank]

People are acting like Intel can really release a CPU twice as fast as Zen as soon as it comes out. Do you people not understand that the law of diminishing returns is already hitting Intel hard? We are reaching the point now where the gains we are getting each generation is starting to slow down. I have no doubt Intel has something up their sleeve to answer Zen, but i do not think it will be a magnitude of "2x faster". 

 

Zen does not have to beat Skylake. Skylake is only 11% faster AT BEST over Haswell, and that's not in regard to gaming. DDR4 has improved Skylake's performance in area's such as compression. In gaming the number is closer to 5%. If AMD gets within spitting distance of Haswell with Zen, then they have a winner. Need i remind you people that hypertransport is fast enough to handle 4-way SLI without needing a PLX8747 bridge? This means you do not need a $400-$500 motherboard to do 4 way SLI. Quite an advantage, don't you think?

 

We also do not know the exact configuration of the FPU's, or if they can be merged into larger sets to handle AVX3. If they can, it will be yet another advantage that Zen would have. However, with more information being released, i feel less hopeful for this happening.

 

If we assume AMD isn't lying, and that Keller is as magical as people make him out to be, then that 40% number over excavator might just land AMD near Haswell. That is all it will take for the platform to be successful. That, and competitive pricing. If they release a 4 core, 8 thread Zen chip at i5 prices, that can handle 4-way SLI on cheap enough, then how can that possibly go wrong?

[Dabombinable]

There have been surprisingly little information in regards to K12.

Well it is ARM, and AMD isn't know for efficiency. I reckon K12 should have been x86, and Zen ARM.

[Guest]

-

Fox Loli God told me that CPUs will be constructed of cheese next year.