New Zen microarchitecture details

[KTE]

http://www.hardware.fr/news/14749/amd-dit-peu-plus-zen.html

Litteraly translated :

Under Blender, wich benefit fully from multithreading, Zen was slightly ahead while consuming a little less according to AMD.

Which still means absolutely nothing in terms of how well it performs

Does everyone here believe AMD will deliver at least 40% IPC gain over EXC?

Because last time that has actually happened in recent times is... when?

And now we're hearing same power efficiency as EXV, low power.

The interesting thing for me is to see how they will balance IPC, clocks and power at launch.

It is quite obvious to see Intel isn't the least stretched these days. Their tank by mid-2016 will be holding +400MHz back for if need be on their top-end. Let alone in 8-12 months. Assuming performance and power are simply astounding, IMHO, AMD cannot compete in the mid to high end DT with Intel unless Zen clocks 3.4-3.7GHz base.

But those assumptions are huge IF in itself.

Sent from HTC 10
(Opinions are own)

 

[Abwx]

Which still means absolutely nothing in terms of how well it performs

It has a meaning, of course..

Less power consumed at equal FP throughput, you think that with higher throughput BDW comsumption will be lower..?.

Does everyone here believe AMD will deliver at least 40% IPC gain over EXC?

Those who look at the uarch certainly have a better idea than those who rely on the "derp derp AMD"..

And now we're hearing same power efficiency as EXV, low power.

Same power consumption at 40% higher IPC, this amount to the same efficency...?.

 

[Tuna-Fish]

Does everyone here believe AMD will deliver at least 40% IPC gain over EXC?

I've pointed this one out to you before, possibly even in this thread. When making a clean design, a designer can just pick the amount of IPC they target. I could design you a CPU that has twice the IPC of Skylake. It wouldn't even be hard. It would be slow, of course, because IPC is a tradeoff against clock speed. Your entire argument is nonsensical. If AMD wanted to, they could have doubled the IPC over EXC, but obviously they decided that would have cost them too much clock speed and so targeted a lower IPC uplift.

The thing to be sceptical about is not the 40% IPC gain, it's what clocks the chips run at. Because that is something the designers can't just choose.

 

[Dresdenboy]

Which still means absolutely nothing in terms of how well it performs

Does everyone here believe AMD will deliver at least 40% IPC gain over EXC?

Because last time that has actually happened in recent times is... when?

Of course, being just a second faster at maybe even lower power means nothing.

And who said "at least 40% IPC gain"? I'm sure, I could construct a scenario, where Zen performs worse than XV*1.4. Maybe it is this kind of misperception, which in the end results in disappointment.

And is your last point about something, that can't be possible because it doesn't happen every day? See it this way: XV is still a execution t'put crippled design with long latencies in the FPU, lots of resource collisions and inefficient caches. Lot's of roadblocks. Saying 40% gain over HSW would be a totally different story. While we are at it - how is POWER9 in this regard?

Another view of your logic, pre 2008: House pricing can't go down. Because last time that has actually happened in recent times is... when?

The thing to be sceptical about is not the 40% IPC gain, it's what clocks the chips run at. Because that is something the designers can't just choose.

This. And the clock thing is a mixed bag. Of course, they had clock targets in the form of FO4 delays per pipeline stage (LinkedIn is full of these timing closure things). But GloFo provides the actual FO4 delay at given voltages.

 

[DrMrLordX]

The first two were due to AMD trying to adapt to Intel's changes concerning SSE5 specs, right?

Basically. Intel switched up the situation when they went from FMA3 to FMA4, while AMD went from FMA4 to FMA3 without knowing that Intel had changed plans:

https://en.wikipedia.org/wiki/FMA_instruction_set#FMA4_instruction_set

It did not work out well for AMD.

^Even with equal IPC and equal power usage, AMD would need a base 4GHz Zen to challenge Intels fastest in 2016.

Uh, what? Since when did Intel have an octocore Broadwell-E with a base clockspeed of 4 GHz? That's the turbo frequency for the 6950x.

 

[Arachnotronic]

Uh, what? Since when did Intel have an octocore Broadwell-E with a base clockspeed of 4 GHz? That's the turbo frequency for the 6950x.

Summit Ridge's problem is that it's going to be a ~3GHz Broadwell IPC-at-best octa-core competing with a 4.2GHz base/4.5GHz turbo Core i7 7700K in the consumer/client market. Also, I have a sneaking suspicion that Broadwell-E will be a better overclocker than Summit Ridge.

AMD looks like it will have a better offering for those who prefer AMD than the Bulldozer-based family, but Zen based products look like they're still going to be a tough sell in just about every segment of the PC market.

 

[raghu78]

Summit Ridge's problem is that it's going to be a ~3GHz Broadwell IPC-at-best octa-core competing with a 4.2GHz base/4.5GHz turbo Core i7 7700K in the consumer/client market. Also, I have a sneaking suspicion that Broadwell-E will be a better overclocker than Summit Ridge.

AMD looks like it will have a better offering for those who prefer AMD than the Bulldozer-based family, but Zen based products look like they're still going to be a tough sell in just about every segment of the PC market.

I would wait till Summit Ridge launches before judging it. I do not buy AMD benchmarks or marketing. Right now I don't know anything about Zen IPC and final clocks of Summit Ridge. But saying that Summit Ridge is going to be a 3 Ghz octa-core with at best Broadwell IPC is purely speculative. I do expect that AMD will be at a significant clock disadvantage vs Intel as they have chosen a density optimized 14nm process which means there is an associated tradeoff compared to a high performance process. Moreover Intel 14nm has the highest transistor performance by a significant margin and no foundry process can come close to it. I am also curious as to what process AMD has chosen.

 

[Doom2pro]

Summit Ridge's problem is that it's going to be a ~3GHz Broadwell IPC-at-best octa-core competing with a 4.2GHz base/4.5GHz turbo Core i7 7700K in the consumer/client market. Also, I have a sneaking suspicion that Broadwell-E will be a better overclocker than Summit Ridge.

AMD looks like it will have a better offering for those who prefer AMD than the Bulldozer-based family, but Zen based products look like they're still going to be a tough sell in just about every segment of the PC market.

It will be the day when you finally have something nice to say about AMD...


Trolling is not allowed
Markfw900

 

[itsmydamnation]

We will have to wait and see on clock speed but the idea that an 8 core high IPC chip that clocks somewhere in the 3-4ghz will have pressure from a high IPC 4 core 4.2 I find a bit of a stretch, its not 2010 anymore, it will be 2017. What workload does the enthusiast or general consumer run that is single threaded or can scale past 4. That tipping point has been largely passed.

If I was intel I would be more worried about Zen either side of that enthusiast market, a <35 watt APU with much closer single thread performance means GPU will matter more. On the other side (server) AMD could do stuff like make 24 core 8 channel memory SOC's with 4-8x 10gb links from Harvested cores!

AMD has agreement to manufacture with Samsung as well, so if GF is at capacity we could see bursting to Samsung happen........

 

[majord]

Summit Ridge's problem is that it's going to be a ~3GHz Broadwell IPC-at-best octa-core competing with a 4.2GHz base/4.5GHz turbo Core i7 7700K in the consumer/client market. Also, I have a sneaking suspicion that Broadwell-E will be a better overclocker than Summit Ridge.

AMD looks like it will have a better offering for those who prefer AMD than the Bulldozer-based family, but Zen based products look like they're still going to be a tough sell in just about every segment of the PC market.

Firstly, is that the confirmed, final clockspeeds for either unlreased products? , No. Infact, I think that might be a slight exageration on both ends there.

Secondly, yes, 4/8t Kabylake is going to be a tough opponent - Just as it is for Intel's own HEDT line - at least the "lower end" of it. It will have a significant lead over Zen SKU up to 4 threads. But any work load that extends beyond this, Zen should progressively claw back, and over-take at around the 6-7 thread + mark I would estimate.. Obviously one could argue till the cows come home which is more relevent to the end user, but at the end of the day, that's up to the buyer.

Again, Intel already have this situation in their own product lines.. and to be honest, I think if Top SKU Zen only offers direct competiton to lower-mid HEDT line, but in a more cost effective, mainstream platform that's got potential for success..

That does not mean they cannot compete with 4c/8t Intel's AT ALL , just because Intel's Halo offering is out of reach as so high clocked. There are, after all a bunch of SKU's below this which zen could theoretically aim at. Offering a 3.4base, 3.6Ghz Turbo @ 65w SKU for example, (talking Raven Ridge here, essentially) has potential to go head to head with current 65w i7's and even the entire i5 Line Particuarly if they're unlocked. Something AMD have absolutely nothing to compete with currently.. .

This continues down the line into mobile.

 

[sirmo]

If the 200mm^2 die size is correct.. than 8c/16t Zen will change the game, no matter the clocks (assuming even 3Ghz and a 40% IPC uplift). Intel will have no choice but to slash their prices.

 

[ShintaiDK]

If the 200mm^2 die size is correct.. than 8c/16t Zen will change the game, no matter the clocks (assuming even 3Ghz and a 40% IPC uplift). Intel will have no choice but to slash their prices.

What does die size have to do with it?

If it performs like a 1000$ chip, AMD will charge ~1000$.

 

[sirmo]

What does die size have to do with it?

If it performs like a 1000$ chip, AMD will charge ~1000$.

If it performs like a $1000 chip AMD will charge $900 for it. Size means there is plenty of margin no matter what and the price war is very likely.

 

[ShintaiDK]

If it performs like a $1000 chip AMD will charge $900 for it. Size means there is plenty of margin no matter what and the price war is very likely.

Why would AMD charge 900$ for it? They sure didn't the last time they had the advantage. And we didn't get a price war.

If AMD charges 900$ instead of 1000$ its because its lacking something.

 

[sirmo]

Why would AMD charge 900$ for it? They sure didn't the last time they had the advantage. And we didn't get a price war.

If AMD charges 900$ instead of 1000$ its because its lacking something.

There was absolutely a price war when Athlon 64 was released. I remember paying $250 for my Opteron 144 at release.. and it completely smoked anything Intel had in the price bracket.

Same reason there is a price war with rx480/70 and 1060s.. and why Nvidia had to disable SLI on it.

 

[ShintaiDK]

There was absolutely a price war when Athlon 64 was released. I remember paying $250 for my Opteron 144 at release.. and it completely smoked anything Intel had in the price bracket.

Yet AMD still sold K8 at 1031$. Something like 1350$ in todays money.

It wasn't a price war, it was a SKU replacement in the ranges.

 

[sirmo]

Yet AMD still sold K8 at 1031$.

It wasn't a price war, it was a SKU replacement in the ranges.

Those binned gold samples are red herrings. I am talking about volume mainstream products. AMD has always offered more value. Up until BD.

 

[ShintaiDK]

Those binned gold samples are red herrings. I am talking about volume mainstream products. AMD has always offered more value. Up until BD.

But prices in those segments didn't change.

AMD charged everything they could within the performance metrics of their products.

The last time AMD undercut anyone was when it was cloning 486 CPUs. But that's ~20 years ago.

 

[sirmo]

But prices in those segments didn't change.

AMD charged everything they could within the performance metrics of their products.

The last time AMD undercut anyone was when it was cloning 486 CPUs. But that's ~20 years ago.

Wrong.. My Opteron 144 could have been $500 and it still would have been better than anything Intel had at the time.

 

[ShintaiDK]

Wrong.. My Opteron 144 could have been $500 and it still would have been better than anything Intel had at the time.

By that definition a 6700K could have been 2000$. And your Opteron 144 wasn't the top bin. What did the Opteron 154 cost?

There is nothing post 486 class CPUs that gives merit to your claim that AMD will change the pricepoints. Instead like the last 20 years, they will fit their products into those pricepoints as they can.

 

[sirmo]

By that definition a 6700K could have been 2000$. And your Opteron 144 wasn't the top bin. It was the cheapest SKU they sold.

There is nothing post 486 class CPUs that gives merit to your claim that AMD will change the pricepoints. Instead like the last 20 years, they will fit their products into those pricepoints as they can.

The reason 6700K isn't $2000 is because they wouldn't be able to sell nowhere near the volume at that price point. Intel's bean counters calculated their wafer opportunity costs and how many 6700K they can make and the optimal bell curve point where they can charge the most and push the highest quantity of product, and they derived at the price.

AMD is in a completely different position.

In business there is an important thing called inertia. It's what keeps people from switching from an incumbent product supplier to an alternative. AMD is an alternative, when it comes to x86 they've always been the alternative.

This means they have to charge less to entice people on the fence to switch.

Top binned parts are in limited supply and there is not much opportunity lost there if they don't sell many of them because they don't have many of them to sell.

Historically speaking AMD has always offered more value per dollar on the mainstream parts (until BD). For instance the example I gave you.

 

[ShintaiDK]

The reason 6700K isn't $2000 is because they wouldn't be able to sell nowhere near the volume at that price point. Intel's bean counters calculated their wafer opportunity costs and how many 6700K they can make and the optimal bell curve point where they can charge the most and push the highest quantity of product, and they derived at the price.

AMD is in a completely different position.

In business there is an important thing called inertia. It's what keeps people from switching from an incumbent product supplier to an alternative. AMD is an alternative, when it comes to x86 they've always been the alternative.

This means they have to charge less to entice people on the fence to switch.

Top binned parts are in limited supply and there is not much opportunity lost there if they don't sell many of them because they don't have many of them to sell.

Historically speaking AMD has always offered more value per dollar on the mainstream parts. For instance the example I gave you.

So in short, you hope on cheap prices of Zen, due to the performance metrics lack of previous designs post Conroe and good faith without merit in history.

Single core K8 prices wasn't alone. AMD did the exact same with X2.

CPU Clock speed L2 cache size Price
Athlon 64 X2 4200+ 2.2GHz 512KB $537
Athlon 64 X2 4400+ 2.2GHz 1024KB $581
Athlon 64 X2 4600+ 2.4GHz 512KB $803
Athlon 64 X2 4800+ 2.4GHz 1024KB $1001

AMD isn't a charity company like some believe. Zen will be priced exactly where it fits into the performance metrics and not a penny cheaper.

 

[mohit9206]

I'm going with Zen=Sandy Bridge.
Never has being optimistic about AMD products ever paid off.So keeping my expectations at this level means i'll be pleasantly surprised when Zen is faster than Sandy Bridge while most of you will be disappointed instead because its slower than Kaby Lake.

 

[sirmo]

So in short, you hope on cheap prices of Zen, due to the performance metrics lack of previous designs post Conroe and good faith without merit in history.

Single core K8 prices wasn't alone. AMD did the exact same with X2.

CPU Clock speed L2 cache size Price
Athlon 64 X2 4200+ 2.2GHz 512KB $537
Athlon 64 X2 4400+ 2.2GHz 1024KB $581
Athlon 64 X2 4600+ 2.4GHz 512KB $803
Athlon 64 X2 4800+ 2.4GHz 1024KB $1001

AMD isn't a charity company like some believe.

I will gladly pay $2000 for a 16c Zen.. I am not hoping for anything.

AMD in a vacuum will charge as much as they can get like any company, but Zen is not being released into the vacuum.
http://www.anandtech.com/show/1745/5

The victory is clear and without debate, at the $300 - $400 price point, the Athlon 64 X2 3800+ is the dual core processor to get.

 

[krumme]

Summit Ridge's problem is that it's going to be a ~3GHz Broadwell IPC-at-best octa-core competing with a 4.2GHz base/4.5GHz turbo Core i7 7700K in the consumer/client market. Also, I have a sneaking suspicion that Broadwell-E will be a better overclocker than Summit Ridge.

AMD looks like it will have a better offering for those who prefer AMD than the Bulldozer-based family, but Zen based products look like they're still going to be a tough sell in just about every segment of the PC market.

Well at aprox 80mm2 for a 4c zen core complex incl l3 it looks to me the atom celeron pentium and to a lesser degree even core i3 line will be rendered useless in 2017/2018. Eg replace l3 mm2 with gpu. So much of intel line is stuck between extremely efficient 2 wide arm a73 and a lean zen.

Surely a 7700k will be faster in many situations and a broadwell skylake e in all. But what will be the price of those? If amd intend to sell 8c in any relevant numbers 800usd is surely not an option. And hardly nessesary considering its even smaller than a polaris! They have every option to price it where there is most total profit.

If efficiency is good zen looks like a perfectly targeted product from a business perspective with good cost benefit in many segments. So big the variable x86 instruction set gets less importance (contrary to small x86) yet without fat fpu solution only needed in corner cases.