New Zen microarchitecture details

[MiddleOfTheRoad]

Lynched due to a 6700K purchase instead of Zen

Yeah, well Zen will probably be a chip that is capable of running Prime95 without crashing on its launch date.

Can't say that Skylake could do that when it launched.

 

[Exophase]

Yeah, well Zen will probably be a chip that is capable of running Prime95 without crashing on its launch date.

Can't say that Skylake could do that when it launched.

I'm sure it'll have its own errata. Much better to have a quickly fixed issue like the Skylake Prime95 one than the Phenom TLB bug that was fixed with a big performance regression until the real hardware fix came several months later.

 

[el etro]

(BS)

One is that AMD will match or come close to intel performance. The second is that *if* that performance is even close to the hype, that AMD will sell it at prices much lower than intel. I think the second assumption is even less likely than the first, especially initially, when Zen will be in short supply.

1)I maintain my confidence in the leakers and the rumour jornalists. Those last ones set the bar high as AMD never set for Zen, but i still believe then. One thing Fottemberg says is that the 8-core SR chip already have a 3Ghz clockspeed in its A0 revision, corroborating the last rumor mill.

2) I don't think Zen will be cheaper in any way. If it were going to be cheaper, it would be made at 28nm or at least have a 28nm version and there wasn't any need for making Bristol Ridge(this last described as a upgraded Carrizo with even better electricals).

And a 3) hardly Zen will have 95W TDP for 3.5+Ghz with the Eight Cores, it likely will come at a much bigger TDP like 140W to do all of this.

 

[ElFenix]

If a frog had wings, he could fly. Two huge assumptions here, and in every Zen thread apparently, even one started by an admitted April Fool's prank.

One is that AMD will match or come close to intel performance. The second is that *if* that performance is even close to the hype, that AMD will sell it at prices much lower than intel. I think the second assumption is even less likely than the first, especially initially, when Zen will be in short supply.

i think there's maybe one person holding both those assumptions around here. it's certainly out of the mainstream.

 

[itsmydamnation]

but this is evidently what you all want because no one seems to want to discuss zen architectural details. clearly what you all want is yet another rehashing of the same old arguments you've had in thread after thread, where no one's mind is changed and many end up showing their @$$.

I've asked time and time again for them to put up or shut up. Not once have they came up with anything of substance on a technical front, you can literally boil their argument down to four words "derp derp coz AMD".

 

[myocardia]

probably like fury x which had very limited supplies for a few months.

Sure, and it's 100% natural with a new process. Intel and nVidia have lesser numbers available, after switching to a new process node, as well. It's just par for the course in this industry.

Am I the only one who thinks that Zen is not aiming for big Xeon, but instead is aiming for the Xeon-D market? I can honestly see Zen being a runaway hit, if it goes up against Xeon-D, instead of the big Xeons. We'll see, I guess!

 

[Phynaz]

You make it sound like AMD is some small fry. These guys have been around since 1969 and introduced several revolutionary tech to the industri.

just top of my head:

-first to break 1ghz barrier with its Athlon cpu to consumers
-first to have a 64-bit extension with the X86 architecture
-first gpu with DirectX 9 hardware support
-first gpu to use GDDR5 memory
-first gpu to use HBM

Do you really think somebody that has been a member here for 10 years hasn't heard this argument a hundred times?

 

[Phynaz]

I've asked time and time again for them to put up or shut up. Not once have they came up with anything of substance on a technical front, you can literally boil their argument down to four words "derp derp coz AMD".

The best predictor of future performance is past performance.

 

[Adored]

The best predictor of future performance is past performance.

Was this also true in 2006? You should know, you've been around for 10 years.

 

[rainy]

The best predictor of future performance is past performance.

Do you really think somebody that has been a member here for a couple years hasn't heard this argument a hundred times?

 

[Phynaz]

Was this also true in 2006? You should know, you've been around for 10 years.

Yes.

 

[Phynaz]

Do you really think somebody that has been a member here for a couple years hasn't heard this argument a hundred times?

Nice try.

 

[itsmydamnation]

The best predictor of future performance is past performance.

thats fine, but CPU's are made up of many components placed together to perform a task

so whats wrong with the existing performance of amds:

decode unit
prefetch and predictors
load store pipeline
physical register file
ALU's
FPU's
cache system
int and fpu schedulers

some of these have problems and we know those ones are changing (ALU, cache for example) . So what about the rest. How about actually contributing something of worth, or how many times are you going to repeat "derp derp coz AMD".

 

[Phynaz]

thats fine, but CPU's are made up of many components placed together to perform a task

so whats wrong with the existing performance of amds:

decode unit
prefetch and predictors
load store pipeline
physical register file
ALU's
FPU's
cache system
int and fpu schedulers

some of these have problems and we know those ones are changing (ALU, cache for example) . So what about the rest. How about actually contributing something of worth, or how many times are you going to repeat "derp derp coz AMD".

Let me know when you can buy those individually.

AMD's history outside of Athlon 64 is one of over promising and under delivering. Going all the way back to K5.

 

[itsmydamnation]

Let me know when you can buy those individually.

AMD's history outside of Athlon 64 is one of over promising and under delivering. Going all the way back to K5.

So what are you doing in the Zen microarchitecture details thread when you dont want to talk about mircoarchutecture :thumbsup:

So you've got nothing but being a typical internet though guy cant back down. I went to the trouble of showing you my evaluation at your request, now you lack the capability to do one yourself when asked. Instead you keep running to Derp Derp coz AMD.

Also your blanket statement is wrong, K7 , jaguar and bobcat performed exactly how amd said they will.

 

[Phynaz]

So what are you doing in the Zen microarchitecture details thread when you dont want to talk about mircoarchutecture :thumbsup:

What's there to discuss? AMD will never talk about 99% of it, and of what's left AMD hasn't said anything substantial.

What do you know about the following that hasn't been said a dozen times already

decode unit
prefetch and predictors
load store pipeline
physical register file
ALU's
FPU's
cache system
int and fpu schedulers

You're just upset because I won't be an AMD shill.

 

[DrMrLordX]

You're just upset because I won't be an AMD shill.

So who are the AMD shills? And he has a point: to date, nobody but maybe The Stilt has cited reasons why Zen might not be so great that have any technical basis whatsoever.

 

[itsmydamnation]

im not upset about anything, back you can't even in the slightest from a technical standpoint backup what you are saying. There is plenty of stuff around CON cores i have read about i haven't bother to talk about. For example like load and store forwarding behavior or bulldozers odd loop behavior but whats the point when you cant even reach first base.

 

[Phynaz]

So who are the AMD shills? And he has a point: to date, nobody but maybe The Stilt has cited reasons why Zen might not be so great that have any technical basis whatsoever.

You know who they are.

The opposite is true also - nobody has shown any technical details on why Zen will be any good. Why isn't AMD releasing any information?

IF Zen is all that why isn't AMD punching hard to Osborne Effect Intel? Why didn't they knock one out of the park at ISSC?

So that leaves us with recent history. Barcelona will be 40% faster than Cloverton. Bulldozer IPC increases. Fury is an overclockers dream.

Edit: Carrizo has 2.4x perf/watt improvement over Kaveri. That was a good one.

 

[itsmydamnation]

You know who they are.

The opposite is true also - nobody has shown any technical details on why Zen will be any good. Why isn't AMD releasing any information?

IF Zen is all that why isn't AMD punching hard to Osborne Effect Intel? Why didn't they knock one out of the park at ISSC?

So that leaves us with recent history. Barcelona will be 40% faster than Cloverton. Bulldozer IPC increases. Fury is an overclockers dream.

Then why dont you disprove what i have linked to you now 3 times.
I have said why i think Zen will be good, the removal of verified bottlenecks that existed in the fundamental design of CON cores, L1 cache and policy, the WCC, L2 size and latency, execution resources, long pipeline no op cache, removal of shared prefetech/predictors, shared L1i, hopefully far more associative L1i etc.

Look at Dresden boys diagram and the split of FPU and ALU into what looks like separate register files ( like jaguar) which means having 4 ALU's isn't going to require a really high amount of ports into a shared PRF for the entire core. look at what we know about the cache system, L1 size and latency , l2 size , l3 is locality aware.

also you know there are patches with Zen pipeline configurations and instruction latencies so why dont you pull out all the bad stuff about Zen from their, there is plenty of detail to look at.

There is lots of info coming out, Your just being willfully ignorant.

AMD isn't going to present to early, remember bulldozer, bobcat and jaguar were all done at Hotchips which is in AUG which aligns far more to Zens release then a conference in march...........

 

[PPB]

Core width is some indicator of performamce, Desdrenboy showed is that info and we can think of better capabilities specially on FP compared to BD.

Cache policy has been shown to be changed in order of better latencies. That indicates more performance 99% of the time.

FPU width has been shown to be lacking compared to haswell and post. From intel. However, this applies only to avx+ workloads and thus Zen could be competitive in the mayority of fp code without the power efficiency and die size penalty of a wide fp pipeline.

And there are more like the existence of a uop cache of sorts to reduce branch misses, etc.

On the other hand, we wont ever see ONE technical argument to think Zen will fail except the "cause its AMD!1!one!!! Eleventy1". No that I expect one from some forum parasytes in here.

 

[mysticjbyrd]

So what are you doing in the Zen microarchitecture details thread when you dont want to talk about mircoarchutecture :thumbsup:l.

They troll every single AMD thread, and the mods don't do anything about it, because they are just as bias. Every single AMD thread devolves into "herp derp AMD sux".


Mod callouts are not allowed
Markfw900

 

[Phynaz]

Core width is some indicator of performamce, Desdrenboy showed is that info and we can think of better capabilities specially on FP compared to BD.

Cache policy has been shown to be changed in order of better latencies. That indicates more performance 99% of the time.

FPU width has been shown to be lacking compared to haswell and post. From intel. However, this applies only to avx+ workloads and thus Zen could be competitive in the mayority of fp code without the power efficiency and die size penalty of a wide fp pipeline.

And there are more like the existence of a uop cache of sorts to reduce branch misses, etc.

On the other hand, we wont ever see ONE technical argument to think Zen will fail except the "cause its AMD!1!one!!! Eleventy1". No that I expect one from some forum parasytes in here.

You were doing great, and then in came the meta commentary. You destroyed your own posts' credibility.

The only company you should get emotional about is the one you work for.

 

[mysticjbyrd]

You were doing great, and then in came the meta commentary. You destroyed your own posts' credibility.

The only company you should get emotional about is the one you work for.

Look at this blatant trolling.

 

[itsmydamnation]

You were doing great, and then in came the meta commentary. You destroyed your own posts' credibility.

The only company you should get emotional about is the one you work for.

still waiting

your all tip and no iceberg