Zen hasn't taped out yet

[Boze]

At least such vague statements are good to stir up a lengthy discussion. Human nature doesn't like soft facts with attached probabilities.

I got some soft facts for us all...

By the time I could actually put a Zen processor in my hands, Intel will be two microarchitectures ahead of where they are right now with Cannon Lake. Most of us are on Haswell, a few on Skylake. Give it another two years, I imagine the majority will be on Skylake or Kaby Lake.

So Zen has to entice people away from what exists right now as well as what exists in the near future.

That's a tall order to fill. And AMD doesn't have the world's best track record with exceeding expectations of USERS. Zen might "exceed expectations" of its own engineers, but that doesn't really impress me much because I don't know what they're expecting.

I can tell you what I'm expecting. Something as powerful, with the performance per watt, as Cannon Lake. Anything less and I'm not interested.

 

[Fjodor2001]

By the time I could actually put a Zen processor in my hands, Intel will be two microarchitectures ahead of where they are right now with Cannon Lake.

Zen is 2016Q4, CannonLake 2017Q4/2018Q1.

And CannonLake-E will be much later than that.

I can tell you what I'm expecting. Something as powerful, with the performance per watt, as Cannon Lake. Anything less and I'm not interested.

More powerful in MT performance or ST performance?

And price does not matter? Regardless if the performance difference is only 10-20%?

 

[ultimatebob]

I got some soft facts for us all...

By the time I could actually put a Zen processor in my hands, Intel will be two microarchitectures ahead of where they are right now with Cannon Lake. Most of us are on Haswell, a few on Skylake. Give it another two years, I imagine the majority will be on Skylake or Kaby Lake.

I'm still wondering if AMD will still exist as an independent publicly traded company by the time Zen is ready to actually ship in volume. They might have been forced into bankruptcy by then, or will have been bought out by another technology company.

 

[mrmt]

I'm still wondering if AMD will still exist as an independent publicly traded company by the time Zen is ready to actually ship in volume. They might have been forced into bankruptcy by then, or will have been bought out by another technology company.

AMD is very cheap by the industry's standards so whoever was willing to buy them would have done so long ago. The fact is that their IP doesn't make sense on high growth market such as mobile or servers and their product line is a train wreck. Since most or all of their CPU IP will go to Intel in the event of a bankruptcy I don't think they have much value beyond graphics IP.

My guess is that if AMD runs out of cash the company will be wind down, there will be an skeleton crew taking care of the console MPU contracts and the graphics IP will be put on some IP vehicle to prevent patent trolling.

I'm really curious about Zen. AMD should be outsourcing a lot of things to third parties and using a lot of synthesis tools in order to cut down costs, on top of betting a lot of cheaper junior engineers because they lost a lot of more experienced veterans to competitors. It will be quite a feat if AMD can deliver something competitive with the shoestring budget they have and working with a subpar foundry partner like Globalfoundries. And let's not forget the incompetent management team, it will be a surprise if these guys have actually hit the mark with the product conception after screwing up so big in the last few years.

 

[Dresdenboy]

More powerful in MT performance or ST performance?

And price does not matter? Regardless if the performance difference is only 10-20%?

That's the question.

I think, those who want top ST CB or SuperPi scores might be disappointed, but those interested in modern MT workloads might get an interesting alternative at comparable prices.

 

[coercitiv]

I can tell you what I'm expecting. Something as powerful, with the performance per watt, as Cannon Lake. Anything less and I'm not interested.

Then you can leave this thread be, that will not happen.

 

[mrmt]

That's the question.

I think, those who want top ST CB or SuperPi scores might be disappointed, but those interested in modern MT workloads might get an interesting alternative at comparable prices.

What do you expect in terms of server performance?

 

[Dresdenboy]

What do you expect in terms of server performance?

Zen might be a bit weak on the FP side, but if the small core hypothesis is correct, AMD might throw lots of cores on a die. SMT, core renaming (for power management), IVR, high integer throughput, and the promised cache subsystem should give nice performance there. It would be able to stay at higher average clocks with all cores+SMT active, as a small core footprint doesn't cause that much static power consumption (while active), making efficient use of available TDP. Size comparison: ~4-5 mm² (incl. L2) vs. 8 mm² Skylake.

L4 HBM2 cache + NVM (managed as one huge address space) seems likely and might help to be competitive.

Big Data might be bottlenecked by the 2 AGUs or cache misses, whatever comes first, as it has 70 mem operands per 100 instructions.

What's not so safe to assume yet (but found in patents and papers): stack cache (for power efficiency), uOp cache, checkpointing (low branch misprediction latency), SMT thread prioritization (not that useful for servers), near threshold computing, reliable computing, redundant computing, asynchronous logic.

 

[LTC8K6]

That's the question.

I think, those who want top ST CB or SuperPi scores might be disappointed, but those interested in modern MT workloads might get an interesting alternative at comparable prices.

That's what people say about Vishera...

 

[Boze]

Zen is 2016Q4, CannonLake 2017Q4/2018Q1.

And CannonLake-E will be much later than that.

More powerful in MT performance or ST performance?

And price does not matter? Regardless if the performance difference is only 10-20%?

I would prefer MT performance honestly. Single-threaded performance is a thing of the past with Windows 10 and Direct X 12.

And no price doesn't matter. If there's a Zen part that costs $999 and competes with Kaby Lake-E / CannonLake-E, I'll take it.

 

[Arachnotronic]

I would prefer MT performance honestly. Single-threaded performance is a thing of the past with Windows 10 and Direct X 12.

And no price doesn't matter. If there's a Zen part that costs $999 and competes with Kaby Lake-E / CannonLake-E, I'll take it.

Zen 8 core should be announced in late 2016 with mass availability in early 2017.

If multithreaded performance is your thing, then a "slow and expensive" 10 core Broadwell-E might be up your alley (and it should be here in Q2 2016 and plug right into your current board). If you wait until perhaps Q2 2017, you should be able to move to the Skylake-E CPU which should feature the "Skylake Xeon" CPU cores (which are improved from the client Skylake CPU cores we have today) and probably in a 12 core config.

 

[Boze]

I think multithreaded performance is going to become everyone's "thing" by late 2016 / 2017.

The move to realistic VR is going to push both GPUs and CPUs hard again, which we haven't seen in awhile now, so I'm very excited to see what becomes of the HEDT space. We've heard that its "dying" for years now, but VR becomes the Next Killer App (tm), then HEDT might see a resurgence, because crappy computers don't cut it for VR.

 

[Fjodor2001]

I would prefer MT performance honestly. Single-threaded performance is a thing of the past with Windows 10 and Direct X 12.

And no price doesn't matter. If there's a Zen part that costs $999 and competes with Kaby Lake-E / CannonLake-E, I'll take it.

Well if price does not matter and you want max MT performance, then you should be looking at the Intel Xeon series. Perhaps the 18C/36T E7-8890v3 at "just" $7174.

Seriously, I guess even you have some price sensitivity.

Also, Zen is 2016Q4, so it'll be competing with Broadwell-E, not KabyLake-E or CannonLake-E which will be much later.

So just a hypothetical example:

-Broadwell-E performs 10-20% better than Zen
-Zen costs $600
-Broadwell-E costs $1000

Which one would you pick?

 

[AtenRa]

Well if price does not matter and you want max MT performance, then you should be looking at the Intel Xeon series. Perhaps the 20 core E7-8890v3 at "just" $7174.

Seriously, I guess even you have some price sensitivity.

Also, Zen is 2016Q4, so it'll be competing with Broadwell-E, not KabyLake-E or CannonLake-E which will be much later.

So just a hypothetical example:

-Broadwell-E performs 10-20% better than Zen
-Zen costs $600
-Broadwell-E costs $1000

Which one would you pick?

He will pick the $1000 BW-E because he wants the fastest available.

 

[Fjodor2001]

He will pick the $1000 BW-E because he wants the fastest available.

Ok, if price really does not matter at all and max MT performance is desired then I don't see why he is even looking at Broadwell-E or the HEDT series to begin with. He should get something like the $7174 Intel E7-8890v3.

 

[svenge]

Ok, if price really does not matter at all and max MT performance is desired then I don't see why he is even looking at Broadwell-E or the HEDT series to begin with.

Because said hypothetical customer is not a ridiculous AMD shill using the most unrealistic scenario available to set up a pricing-based strawman?

There's a big difference between a person wanting the fastest-available consumer product (i7 HEDT) and a 18-core chip made for 4S-8S settings (E7 88xx v3) in terms of available supporting hardware and end-user experience.

 

[Fjodor2001]

Because said hypothetical customer is not a ridiculous AMD shill using the most unrealistic scenario available to set up a pricing-based strawman?

There's a big difference between a person wanting the fastest-available consumer product (i7 HEDT) and a 18-core chip made for 4S-8S settings (E7 88xx v3) in terms of available supporting hardware and end-user experience.

You did not notice the when I first mentioned the Intel E7-8890v3?

Anyway, then price does matter after all. Which exactly was my point.

But if Boze really is that price-insensitive, then he should at least be looking at lower/mid priced Xeon models instead of the HEDT CPUs like Broadwell-E or Zen. He does not have to get the top model in the Xeon series.

 

[mysticjbyrd]

I got some soft facts for us all...

By the time I could actually put a Zen processor in my hands, Intel will be two microarchitectures ahead of where they are right now with Cannon Lake. Most of us are on Haswell, a few on Skylake. Give it another two years, I imagine the majority will be on Skylake or Kaby Lake.

So Zen has to entice people away from what exists right now as well as what exists in the near future.

That's a tall order to fill. And AMD doesn't have the world's best track record with exceeding expectations of USERS. Zen might "exceed expectations" of its own engineers, but that doesn't really impress me much because I don't know what they're expecting.

I can tell you what I'm expecting. Something as powerful, with the performance per watt, as Cannon Lake. Anything less and I'm not interested.

Cannonlake won't be out until 2018, so zen would only be competing with kabylake .

Zen's appeal will be the higher core count, sort of like the x99 series.

 

[mikk]

Cannonlake won't be out until 2018, so zen would only be competing with kabylake .

Zen APU should compete against Kabylake and Cannonlake.

 

[Fjodor2001]

NVM

 

[LTC8K6]

Taped out
Tapped out

Which will it be?

 

[cytg111]

The move to realistic VR is going to push both GPUs and CPUs hard again, which we haven't seen in awhile now, so I'm very excited to see what becomes of the HEDT space. We've heard that its "dying" for years now, but VR becomes the Next Killer App (tm), then HEDT might see a resurgence, because crappy computers don't cut it for VR.

- Im with you there 100% but that part about ST performance not being that important. I think ST performance will still be the number one metric 10 years from now.

 

[cytg111]

Taped out
Tapped out

Which will it be?

Funny guy over here

 

[Kuiva maa]

Zen might be a bit weak on the FP side, but if the small core hypothesis is correct, AMD might throw lots of cores on a die. SMT, core renaming (for power management), IVR, high integer throughput, and the promised cache subsystem should give nice performance there. It would be able to stay at higher average clocks with all cores+SMT active, as a small core footprint doesn't cause that much static power consumption (while active), making efficient use of available TDP. Size comparison: ~4-5 mm² (incl. L2) vs. 8 mm² Skylake.

L4 HBM2 cache + NVM (managed as one huge address space) seems likely and might help to be competitive.

Big Data might be bottlenecked by the 2 AGUs or cache misses, whatever comes first, as it has 70 mem operands per 100 instructions.

This plan sounds like a fast death sentence for AMD ,in other words, BD no 2. Why go with many small cores plus SMT when you can have similar MT performance but better ST performance If you go with a moderate amount of wider cores+SMT?

 

[Dresdenboy]

That's what people say about Vishera...

Well, at much lower power and higher throughput, but yes...