New Zen microarchitecture details

[Arachnotronic]

I would be too, but I am also cautious after the promise versus reality of Phenom and Bulldozer.

Phenom II wasn't that bad, but yes, the hype around Phenom I and Bulldozer was...well, much too enthusiastic

 

[Dresdenboy]

Phenom II wasn't that bad, but yes, the hype around Phenom I and Bulldozer was...well, much too enthusiastic

Yep, Zen will also be a nightmare. The question is: For whom?

 

[guskline]

Awe! Dresdenboy, what a wordsmith and man of intrigue👍

 

[Arachnotronic]

Yep, Zen will also be a nightmare. The question is: For whom?

Zen will certainly be a lot better than the Piledriver FX chips that are being sold into the market and, if the AMD investor slide (which has since been removed) was accurate, should be able to credibly compete with the lower end of Intel's HEDT stack in multi-threaded performance.

It'll be interesting to see what frequencies the chips will come at, how much overclocking headroom they will have, and what prices they will come at. Still a lot of unknowns right now.

 

[nismotigerwvu]

Honestly, if the chips had little overclocking headroom I wouldn't mind all that much. I'd much rather see an aggressive AMD that was wringing every last ounce out of the silicon and slapping a warranty on it than leaving 500mhz on the table. Granted, this all only applies if all that performance adds up to something competitive in both price and performance. My Phenom II 955 is a C2 so a great overclock isn't really in the cards, but it will do 3.6ghz at 1.3375V (so +400mhz and -12.5mV). Even if that was the higher side of what Zen chips would do I'd be just fine.

 

[ShintaiDK]

Yep, Zen will also be a nightmare. The question is: For whom?

I am sure Lisa got her golden parachute secured

 

[KTE]

@Tuna-Fish

In any political and corporate setting, it is very difficult to prove raw dishonesty. I don't believe anyone gets sued for claiming anything about future performance. I haven't seen it to date for AMD, ATI, Intel or nVidia?

Because excuses can be made for everything. That's why companies have lawyers, PR mouthpieces and damage control. A simple dodge is it being 40% faster only with max boost at 125W in a few highly MT apps. That's why AMDs performance claims are so ambiguous until the final stages.

If it tanks hard, all you do is apologize and some of the team jumps ship, especially the CEO. Who by then, already knew the preemptive fate, so has milked the company enough to be a mega kingpin *ahem* Hector Ruiz.

Again, I'm not saying Zen won't be 40% faster than BD, PD over EX (we don't have enough to know?). Only playing devils advocate.

Sent from HTC 10

 

[Ajay]

AMD will show the Zen uArch at HotChips it seems, so it won't be too long before we know what's going on inside & we can make some ballpark IPC estimates.

Actually looking forward to seeing how close Dresdenboy's uArch model was compared to the real thing.

So August of this year :thumbsup: Hopefully Zen will be no longer be in early pre-alpha (whatever that is).

 

[DrMrLordX]

The issue for them is that fans are raising expectations for them. 40% IPC is a simple statement, but a lot of people are reading it as 40% performance. If/when AMD fails to have 40% performance on ST loads, how many people will be disappointed?

I could give a fig what it does in single-threaded workloads. Throughput is the big question. What is the throughput of 8c/16t Summit Ridge versus a hypothetical 8m/16t XV that will never see the light of day because AMD committed all their server/workstation CPU development resources on Keller's design?

Or to put it differently, if AMD can't show superior throughput from a 2c/4t Summit Ridge (half a Zen cluster, basically) versus Carrizo at the same clockspeed over a broad range of workloads, then there is something seriously wrong.

The throughput *should* be higher, and we should know more in August (thank goodness).

 

[itsmydamnation]

I could give a fig what it does in single-threaded workloads. Throughput is the big question. What is the throughput of 8c/16t Summit Ridge versus a hypothetical 8m/16t XV that will never see the light of day because AMD committed all their server/workstation CPU development resources on Keller's design?

Or to put it differently, if AMD can't show superior throughput from a 2c/4t Summit Ridge (half a Zen module, basically) versus Carrizo at the same clockspeed over a broad range of workloads, then there is something seriously wrong.

The throughput *should* be higher, and we should know more in August (thank goodness).

If datacentres/enterprise wanted throughput then CON cores would have a much greater market share then they currently do. What they want solid IPC at solid clock with as many cores as possible with as much ram as possible so they can sell the "cloud" to everyone.

a 2c4t Zen in throughput for a high ILP code will loose to a 2m4c4t Carrizo for a simple reason. One has a total of 1024/512 load store bandwidth and the other has 512/256 load store bandwidth while having approx the same amount of int and FP resources. But as you can see the market has preferred 2 strong threads + 2 weak threads over 4 medium threads.

if your using throughput to describe many independent low ILP threads running together ( a many tab web broswer for example) then that's a different kettle of fish and Zen should be quite dominate vs Excavator in that area.

 

[jhu]

I could give a fig what it does in single-threaded workloads. Throughput is the big question. What is the throughput of 8c/16t Summit Ridge versus a hypothetical 8m/16t XV that will never see the light of day because AMD committed all their server/workstation CPU development resources on Keller's design?

That's easy. You can easily get performance/clock/module for the current Excavator chip. Then just multiply multi-threaded benchmark score result by 8. Now you have theoretical performance/clock of 8 module Excavator. Then you just sit on these calculated results and compare with Zen whenever it arrives.

 

[frozentundra123456]

I am sure Lisa got her golden parachute secured

Yea, there was one left after everyone that was associated with BW-E got theirs.

 

[24601]

I am sure Lisa got her golden parachute secured

The only reason companies put females in leadership positions is because their focus groups show that females will shield the company from criticism when it crashes and burns taking everything with it.

No one puts females in leadership positions in US companies for any reason other than that.

If you see a female being put in a leadership position in the US you know the company is in pure damage control mode.


Threadcrapping. off-topic and trolling are not allowed
Markfw900

 

[dark zero]

The only reason companies put females in leadership positions is because their focus groups show that females will shield the company from criticism when it crashes and burns taking everything with it.

No one puts females in leadership positions in US companies for any reason other than that.

If you see a female being put in a leadership position in the US you know the company is in pure damage control mode.

Not necessary... Intel has a female leader and is a good one.... on the other side the Atom Team....

 

[Avalon]

The only reason companies put females in leadership positions is because their focus groups show that females will shield the company from criticism when it crashes and burns taking everything with it.

No one puts females in leadership positions in US companies for any reason other than that.

If you see a female being put in a leadership position in the US you know the company is in pure damage control mode.

My company's CEO is a female and has been CEO for the past 6 years. In that time frame, the company has doubled in value and we're making record profits.

/OT

 

[SarahKerrigan]

The only reason companies put females in leadership positions is because their focus groups show that females will shield the company from criticism when it crashes and burns taking everything with it.

No one puts females in leadership positions in US companies for any reason other than that.

If you see a female being put in a leadership position in the US you know the company is in pure damage control mode.

What the hell?

 

[24601]

The replies so far showcase exactly why this practice is commonplace as it's a psychological trick that works even if people know that it's happening and exactly why it's happening.

 

[jhu]

The only reason companies put females in leadership positions is because their focus groups show that females will shield the company from criticism when it crashes and burns taking everything with it.

No one puts females in leadership positions in US companies for any reason other than that.

If you see a female being put in a leadership position in the US you know the company is in pure damage control mode.

Yeah… that's why Pepsi Co. and IBM are crashing and burning, right?

 

[Exophase]

The only reason companies put females in leadership positions is because their focus groups show that females will shield the company from criticism when it crashes and burns taking everything with it.

No one puts females in leadership positions in US companies for any reason other than that.

If you see a female being put in a leadership position in the US you know the company is in pure damage control mode.

Or, you know, a good person for the job also happens to be a woman. Or in this case, the next best remaining option who rose up through the ranks. AMD has been churning through CEOs over the past several years and that's the real sign that they're sinking, at least if you don't notice all the other signs. It was inevitable that they wouldn't all be males.

Meanwhile, there are other companies that have done perfectly well under female CEOs.

 

[DrMrLordX]

That's easy. You can easily get performance/clock/module for the current Excavator chip. Then just multiply multi-threaded benchmark score result by 8. Now you have theoretical performance/clock of 8 module Excavator. Then you just sit on these calculated results and compare with Zen whenever it arrives.

I know that. But we don't know if Summit Ridge can best that score yet, and that's what I'm waiting to see. All this talk of single threaded performance is basically pointless.

If datacentres/enterprise wanted throughput then CON cores would have a much greater market share then they currently do.

No they wouldn't. Their TCO is entirely too high even taking into account throughput.

What they want solid IPC at solid clock with as many cores as possible with as much ram as possible so they can sell the "cloud" to everyone.

If that were true, neither Intel nor AMD would even bother selling HT-capable CPUs as datacenter products. Total throughput is still going to matter. You are also failing to take into account the fact that Zen will have to have significantly higher performance from the first thread of each core vs. the first thread of each XV module in order to exceed XV's throughput given identical clockspeeds. AMD has to deliver on both counts.

a 2c4t Zen in throughput for a high ILP code will loose to a 2m4c4t Carrizo for a simple reason. One has a total of 1024/512 load store bandwidth and the other has 512/256 load store bandwidth while having approx the same amount of int and FP resources. But as you can see the market has preferred 2 strong threads + 2 weak threads over 4 medium threads.

Each Zen core has twice the fp resources of a single XV module.

The market is getting higher throughput from "2 strong threads + 2 weak threads" when those two strong threads happen to be coming from Intel, so I do not think the market has rendered any verdict as to whether or not it prefers SMT or CMT. What it wants is better performance, lower TCOs, and so forth and so on. Intel has been delivering all of the above since 2006.

 

[crashtech]

The only reason companies put females in leadership positions is because their focus groups show that females will shield the company from criticism when it crashes and burns taking everything with it.

No one puts females in leadership positions in US companies for any reason other than that.

If you see a female being put in a leadership position in the US you know the company is in pure damage control mode.

I'm willing to get infracted just to call you a misogynistic ass.

 

[Arachnotronic]

The only reason companies put females in leadership positions is because their focus groups show that females will shield the company from criticism when it crashes and burns taking everything with it.

No one puts females in leadership positions in US companies for any reason other than that.

If you see a female being put in a leadership position in the US you know the company is in pure damage control mode.

Come on dude, not cool

 

[sxr7171]

My company's CEO is a female and has been CEO for the past 6 years. In that time frame, the company has doubled in value and we're making record profits.

/OT

Not HP I'm sure.

 

[Qwertilot]

To be fair it could very easily be read as a statement about the normative culture in companies in the US....

Probably an unfair one even in those terms of course but an awful lot less offensive.

 

[NTMBK]

VIA, Zhaoxin ZX-C+ FC-1080/1081 프로세서, ZX-100S 칩셋보드 발표.
source: savage3d.tistory.com

I think you clicked on the wrong thread

 

[SAAA]

Naa that's just Zen early pre-prototype