What does Zen need to be in order to get you to buy it?

[TheELF]

That graph doesnt compare a Quad vs 8-Core in DX-12/Vulkan.

If you only have 4 cores, your main thread will be slower because Game Code (Purple) and DirecX Driver (Blue) will only scale to 4 Cores. So even if you have higher IPC, your 4 Cores will need to compute more than the 8-Core CPUs.

Yeah but you (me and the graph) are talking about the 8 core CPU of the consoles (because that's what devs code for) ...and only 6 of those cores are used for the games,rest goes to OS,and of those 6 cores only 4 are on the same chip so only those 4 do the heavy lifting while the other two only run secondary tasks.

There is no game whose gamecode is written for 8 desktop level cores.
And the only thing we know about ZEN is 40% MORE IPC then EX,doesn't tell us anything about the speed of execution of those commands only about the amount available per cycle/clock.

"your 4 Cores will need to compute more than the 8-Core CPUs"
If that graph where to compare 8core FX to 4core i5 then the bars would be twice as long on the FX compared to the i5,maybe even longer depending on clocks.
That's the thing, it is able to compute more,much more.

 

[Dresdenboy]

Here is a random game "no one plays" that mostly used just 2 cores; and because of lack of multi-threading in the game, it helps to show just how far behind AMD is in IPC:

i5 6600 is 129% faster than 8150, and 89% faster than 8350:
[pic removed]
http://gamegpu.com/rts-/-стратегии/cossacks-3-test-gpu

40% gain in IPC is nowhere close to even catching up to Haswell.

40% is not a constant FPS gain. There are tens of factors to consider.

Here is a well-threaded AAA title:
[pic removed]
http://gamegpu.com/action-/-fps-/-tps/assassin-s-creed-unity-test-gpu.html

3.6Ghz 8150 = 48 FPS
4.0Ghz 8350 = 57 FPS
4.7Ghz 9590 = 67 FPS

Vs.

3.5Ghz Haswell i3 4330 = 69 FPS
3.0Ghz i7 5960X = 107 FPS

Seems Intel is about 90-100% ahead in IPC. Excavator hardly moved IPC over Vishera.
[...]
That's why AMD really needs to work closer with AAA game developers so that Zen can shine under DX12/Vulkan.

Your i7-5960X number suggests, that this won't be the worst title for Zen.

And if its behaviour comes closer to Intel's chips (mem, caches, ex units, etc.), it might simply run the games better than BD/PD out of the box. These titles aren't necessarily heavily optimized for Intels latest uarchs. It's just more likely that BD/PD uarchs with their focus on different workloads aren't well suited for current games or a DX11 driver. Even bdverx compiler optimizations would only add 10-20'% if lucky.

 

[superstition]

40% gain in IPC is nowhere close to even catching up to Haswell.

That is a great chart for showing how a game can have an i3 outperform an i7.

 

[BTRY B 529th FA BN]

For me, I guess equal performance of Broadwell-E, if not I might pick up a 6850K or 6900K

 

[AtenRa]

"your 4 Cores will need to compute more than the 8-Core CPUs"
If that graph where to compare 8core FX to 4core i5 then the bars would be twice as long on the FX compared to the i5,maybe even longer depending on clocks.
That's the thing, it is able to compute more,much more.

What the hell are you talking about ??

You have 100

Divide 100 by 4 cores and by 8 cores.

4 Cores will have to compute 2x (Double) more per Core than the 8 Cores.

The main Thread will be almost 2x more work for the Quad Core than the slower IPC 8 Core.

 

[TheELF]

What the hell are you talking about ??

You have 100

Divide 100 by 4 cores and by 8 cores.

4 Cores will have to compute 2x (Double) more per Core than the 8 Cores.

The main Thread will be almost 2x more work for the Quad Core than the slower IPC 8 Core.

Tell me one game that gets 100% usage on a 8core CPU....
"The main Thread will be almost 2x more work for the Quad Core than the slower IPC 8 Core."
That's what I'm saying the 4core will do twice the work on the main thread meaning twice the FPS.
Also,if the example from AMD where to run on a 4core CPU.
No loss in main thread AT ALL.

 

[frozentundra123456]

What the hell are you talking about ??

You have 100

Divide 100 by 4 cores and by 8 cores.

4 Cores will have to compute 2x (Double) more per Core than the 8 Cores.

The main Thread will be almost 2x more work for the Quad Core than the slower IPC 8 Core.

No, the amount of "work" will be the same. The question is how many cores the work is distributed over. The only time the 8 core will have less work to do per core is if the load is evenly distributed over more than 4 cores.

 

[TheELF]

The unit of measurement for work in this scenario (game) is FPS so the only way to get the same amount of work is to limit FPS to something both CPUs can achieve.

 

[DrMrLordX]

Excavator hardly moved IPC over Vishera.

Huh? I don't see XV benched anywhere in those graphs, and unless you're running games that suffer from XV's short l2 cache (there are some), XV smashes Vishera/PD to pieces in IPC. The cache speed improvements in XV are no joke.

 

[The Stilt]

Huh? I don't see XV benched anywhere in those graphs, and unless you're running games that suffer from XV's short l2 cache (there are some), XV smashes Vishera/PD to pieces in IPC. The cache speed improvements in XV are no joke.

The average IPC improvement over PD should be 15.5% (according to AMD) and that's quite close to the actual average. However there are also MANY cases where Excavator performs the same or even worse than Piledriver. Still far from "smashing", even if it was always 15.5% faster IMO.

 

[moonbogg]

So is this thing still 6 months away? People will be using Kabylake's at 5.2Ghz...watch it happen.

 

[Arachnotronic]

So is this thing still 6 months away? People will be using Kabylake's at 5.2Ghz...watch it happen.

IMO Summit Ridge is going to be a crummy gaming chip. ST perf/clock looks like it will be lower than Haswell (based on AnandTech Cinebench result for Bristol Ridge) and clocks likely to be lower.

I don't think anybody primarily interested in gaming should wait for Zen.

 

[moonbogg]

IMO Summit Ridge is going to be a crummy gaming chip. ST perf/clock looks like it will be lower than Haswell (based on AnandTech Cinebench result for Bristol Ridge) and clocks likely to be lower.

I don't think anybody primarily interested in gaming should wait for Zen.

I wanted to believe in Zen so bad, but that joy is fading pretty hard. At least I get to look forward to Skylake-E still. I know that's going to kick ass. Its just that by the time Zen actually ends up in people's rigs, Intel will have already marched ahead two generations basically. We'll have Skylake-E and Kabylake and yet more coming not far after those. Its just always too little too late with AMD.

 

[Arachnotronic]

I wanted to believe in Zen so bad, but that joy is fading pretty hard. At least I get to look forward to Skylake-E still. I know that's going to kick ass. Its just that by the time Zen actually ends up in people's rigs, Intel will have already marched ahead two generations basically. We'll have Skylake-E and Kabylake and yet more coming not far after those. Its just always too little too late with AMD.

I don't think Summit Ridge will even compare well to a 4790K for most gaming/consumer applications. With IPC and clocks lower than Haswell, it's just not going to be that interesting for applications that require best-in-class <=4 thread performance.

I do think that Summit Ridge will be a better product, even for gamers, than the current FX chips are.

 

[moonbogg]

I don't think Summit Ridge will even compare well to a 4790K for most gaming/consumer applications. With IPC and clocks lower than Haswell, it's just not going to be that interesting for applications that require best-in-class <=4 thread performance.

I do think that Summit Ridge will be a better product, even for gamers, than the current FX chips are.

Oh. Well that's wonderful.

 

[linkgoron]

I wanted to believe in Zen so bad, but that joy is fading pretty hard. At least I get to look forward to Skylake-E still. I know that's going to kick ass. Its just that by the time Zen actually ends up in people's rigs, Intel will have already marched ahead two generations basically. We'll have Skylake-E and Kabylake and yet more coming not far after those. Its just always too little too late with AMD.

I just don't really understand why anyone with enough cash would wait for AMD on the CPU side or expect AMD to deliver high-end performance. Unless you want a Zen APU and those are late 2017, go Intel. I assume that the 6700k will easily beat Zen in gaming, and the 7700k is basically here. Sadly, I'm expecting the Zen reviews to look at best like this. I haven't been following the CPU side as much as the GPU side, as it kind of bores me these days, so please correct me if I'm wrong.

That's not to say that AMD will not have a compelling product for the price. But, if you're not going for a lower-end rig I don't see any reason to wait for AMD.

 

[jpiniero]

IMO Summit Ridge is going to be a crummy gaming chip. ST perf/clock looks like it will be lower than Haswell (based on AnandTech Cinebench result for Bristol Ridge) and clocks likely to be lower.

Stock clocks yes it doesn't look good. But if it can overclock to like 4 Ghz the 6/8 core would be reasonably decent that if you could make the case for it if it was cheap. You only have to look at where Sandy Bridge-E lands on a lot of the GameGPU CPU charts for that.

 

[Arachnotronic]

Stock clocks yes it doesn't look good. But if it can overclock to like 4 Ghz the 6/8 core would be reasonably decent that if you could make the case for it if it was cheap. You only have to look at where Sandy Bridge-E lands on a lot of the GameGPU CPU charts for that.

I would be very surprised if 4GHz were in the cards for Summit Ridge/Zen. AMD made the machine a lot wider, which should make hitting higher frequencies a much more daunting task than it was on the 'Dozer lineage. That, coupled with AMD's explicit statements that Zen is a server-first core, leads me to believe that we'll be lucky to see these things overclock to 3.5GHz.

 

[witeken]

It should hit 4GHz quadcore, right? Maybe not out of the box, but certainly OC?

 

[Arachnotronic]

It should hit 4GHz quadcore, right? Maybe not out of the box, but certainly OC?

Not necessarily.

 

[witeken]

Not necessarily.

So the Jim Keller hype has tapered off?

 

[Arachnotronic]

So the Jim Keller hype has tapered off?

Jim Keller is just one man. Even if he were the cleverest architect ever to grace God's green Earth, it is still a very large team effort. Hitting high IPC at very high frequency is tough.

 

[itsmydamnation]

I would be very surprised if 4GHz were in the cards for Summit Ridge/Zen. AMD made the machine a lot wider, which should make hitting higher frequencies a much more daunting task than it was on the 'Dozer lineage.

This is really inaccurate
first the int pipeline grew by two pipes so i assume the iPRF grew by 4r2w that grew. But after that everything else is either similar or much less complex.

FPU/Frpf are the same width
There is no need to stich two int cores to one FPU unit
front end decode is the same width
L/S pipeline is the same width
int schedulers appear to be 4 times smaller
the bypass/forwarding network is unknown at this point
retirement grew but i would be really surprised if updating architectural registers is clock limiting
L1D doubled but so did its associativity so that should be the same
L1i stayed the same but doubled in associativity
L2 is significatly smaller
there is no write coalescing cache

Add in that it looks like Zen's pipeline has grown by 4 stages

That, coupled with AMD's explicit statements that Zen is a server-first core, leads me to believe that we'll be lucky to see these things overclock to 3.5GHz.

Yes but you use faulty data in your assessments.
I wouldn't even try to guess Fmax at this point what makes you so sure other then Derp Derp AMD?

 

[witeken]

Hitting high IPC at very high frequency is tough.

Yet it isn't a miracle ?

 

[Arachnotronic]

This is really inaccurate
first the int pipeline grew by two pipes so i assume the iPRF grew by 4r2w that grew. But after that everything else is either similar or much less complex.

FPU/Frpf are the same width
There is no need to stich two int cores to one FPU unit
front end decode is the same width
L/S pipeline is the same width
int schedulers appear to be 4 times smaller
the bypass/forwarding network is unknown at this point
retirement grew but i would be really surprised if updating architectural registers is clock limiting
L1D doubled but so did its associativity so that should be the same
L1i stayed the same but doubled in associativity
L2 is significatly smaller
there is no write coalescing cache

Add in that it looks like Zen's pipeline has grown by 4 stages


Yes but you use faulty data in your assessments.
I wouldn't even try to guess Fmax at this point what makes you so sure other then Derp Derp AMD?

Bottom line is that if AMD's Zen could hit Intel-like frequencies they would have pushed their Zen ES to match Broadwell-E base clock in their Blender demo rather than nerf the Broadwell to make their point.

Stilt also says current Summit Ridge ES chips are shipping at 2.8ghz base and 3.2ghz Max single core turbo. Max single core turbo on BDW-E is 4GHz.