AMD Engineer: Developers Given Total Freedom with 'Zen'

[patrickjp93]

I would not be surprised at all if intel already have something faster than what zen will be. Someone would have to run the compute numbers but is zens 40% improvement enough to beat even skylake clock for clock core for core. If its not enough AMD is going to have to push 8 core really hard and maybe even go further than that.

I ran the numbers through QEMU and a virtual CPU constructed with theoretical latencies based on AMD's 40% IPC improvement over Excavator claim. Using an integer linear program as the optimization measure, Zen will be within +/- 3% of Haswell clock for clock. That means AMD has to win on clock speed.

[Humbug]

THey don't need the performance crown if they can still maintain const competitiveness.

This.

The problem right now is that AMD's architecture is so far behind Intel that AMD is forced to sell even big top of their range stuff at budget prices in low volumes. They can't make any money on it, no margins. What AMD needs is enough of an improvement that they can move volume and start making money. They don't need the flagship crown, but they can't afford a situation where i3 is close to their top part. E.g. They can't be selling 8-core CPUs for under $150.

I guess this is all or nothing. AMD has enough cash to survive until Zen launches. By if it fails they won't have the resources left to try again a couple of years later...

[Biased Opinion]

I ran the numbers through QEMU and a virtual CPU constructed with theoretical late vies based on AMD's 40% IPC improvement over Excavator claim. Using an integer linear program as the optimization measure, Zen will be within +/- 3% of Haswell clock for clock. That means AMD has to win on clock speed.

Even if your numbers are off by +/- 10% this stil puts AMD in a bad spot because it puts them up against skylake and competing in a clock speed war against those chips cant end well, at least if all the initial overclocking impressions are to be believed. This just leaves core count and price and i dont think selling 8 core chips for $150 is a good business strategy.

[SeriousDad69]

Even if your numbers are off by +/- 10% this stil puts AMD in a bad spot because it puts them up against skylake and competing in a clock speed war against those chips cant end well, at least if all the initial overclocking impressions are to be believed. This just leaves core count and price and i dont think selling 8 core chips for $150 is a good business strategy.

What people seem to forget is CPUs don't age as hard as GPUs. My buddy is still maxing out games on a stock i7 2600k no problem. All they need to do is benchmark the Zen chip against the latest Intel chip with the exact same video card and they'll probably have the exact same frame rate because games are less CPU intensive and more GPU intensive these days, especially with DX12 and Vulkan. Same performance in gaming and cheaper, everyone I know would get in on that.

[Humbug]

What people seem to forget is CPUs don't age as hard as GPUs. My buddy is still maxing out games on a stock i7 2600k no problem. All they need to do is benchmark the Zen chip against the latest Intel chip with the exact same video card and they'll probably have the exact same frame rate because games are less CPU intensive and more GPU intensive these days, especially with DX12 and Vulkan. Same performance in gaming and cheaper, everyone I know would get in on that.

Ya probably unless you turn the resolution down to 640x480 to create a CPU bottleneck discrete GPU gaming performance will be within spitting distance.

[MageTank]

Even if your numbers are off by +/- 10% this stil puts AMD in a bad spot because it puts them up against skylake and competing in a clock speed war against those chips cant end well, at least if all the initial overclocking impressions are to be believed. This just leaves core count and price and i dont think selling 8 core chips for $150 is a good business strategy.

Skylake is not a threat to Zen. Skylake-E and Kaby Lake are where things will start getting rocky for Zen. Remember, Skylake, on its best day, is only around 5% faster than Haswell when you compare IPC. In some cases, Skylake even falls a little behind Haswell, depending on the task at hand. Skylake is more so a platform upgrade, Z170 being better than Z97. The CPU selection itself isn't exactly enticing to current Haswell owners.

 

If Zen matches Haswell, or even gets close to that 3% number that is rumored by @patrickjp93, it puts AMD in the perfect position to not only take out the Z97 platform, but also the X99 platform (hypertransport can easily handle 4 way SLI and 4 way crossfire, so PCIE lane count is irrelevant. Zen also supports redundant amounts of memory) assuming those 8 core, 16 thread Zen CPU's are indeed the high end consumer grade chips. Zen being slightly delayed, puts it around the same release date as Kaby Lake (Skylake Refresh) and that could be where AMD gets hurt. If the Skylake refresh is anything like the Haswell refresh (Devils Canyon) then AMD is in for a world of hurt. Skylake-E will also be capable of matching, or even exceeding the high end Zen CPU's, which would also cut into AMD's potential sales.

 

 AMD's biggest hurdle in my eyes, will be pricing and board features. AMD needs enticing board features to compete with the fancy offerings the current Z170 boards have. Granted, most of this will be left up to board partners, so i wont go into details on that subject. Pricing needs to be taken seriously too. AMD's reputation is hurt, so they cant just come out of the gate with a $300-$400 CPU and expect people to take them seriously. They have to offer relative performance to Intel, and undercut them slightly on pricing while also maintaining a decent profit margin. This might be hard to do. Zen should be a giant leap in the right direction, i only pray it won't be too late, with Intel being hot on their tails with both Kaby Lake and Skylake-E. I hope AMD has some extra tricks up their sleeves.

[Clyne]

Skylake is not a threat to Zen. Skylake-E and Kaby Lake are where things will start getting rocky for Zen. Remember, Skylake, on its best day, is only around 5% faster than Haswell when you compare IPC. In some cases, Skylake even falls a little behind Haswell, depending on the task at hand. Skylake is more so a platform upgrade, Z170 being better than Z97. The CPU selection itself isn't exactly enticing to current Haswell owners.

 

If Zen matches Haswell, or even gets close to that 3% number that is rumored by @patrickjp93, it puts AMD in the perfect position to not only take out the Z97 platform, but also the X99 platform (hypertransport can easily handle 4 way SLI and 4 way crossfire, so PCIE lane count is irrelevant. Zen also supports redundant amounts of memory) assuming those 8 core, 16 thread Zen CPU's are indeed the high end consumer grade chips. Zen being slightly delayed, puts it around the same release date as Kaby Lake (Skylake Refresh) and that could be where AMD gets hurt. If the Skylake refresh is anything like the Haswell refresh (Devils Canyon) then AMD is in for a world of hurt. Skylake-E will also be capable of matching, or even exceeding the high end Zen CPU's, which would also cut into AMD's potential sales.

 

 AMD's biggest hurdle in my eyes, will be pricing and board features. AMD needs enticing board features to compete with the fancy offerings the current Z170 boards have. Granted, most of this will be left up to board partners, so i wont go into details on that subject. Pricing needs to be taken seriously too. AMD's reputation is hurt, so they cant just come out of the gate with a $300-$400 CPU and expect people to take them seriously. They have to offer relative performance to Intel, and undercut them slightly on pricing while also maintaining a decent profit margin. This might be hard to do. Zen should be a giant leap in the right direction, i only pray it won't be too late, with Intel being hot on their tails with both Kaby Lake and Skylake-E. I hope AMD has some extra tricks up their sleeves.

Me= Zen that's around Haswell or higher or lakes in the sky will be a hell of a upgrade over my aging Vishera FX 8320. 

Just hope that AMD release it soon......from what i heard Zen is basically done at this point, but there's some problems with the 14nm processing node.

[MageTank]

Me= Zen that's around Haswell or higher or lakes in the sky will be a hell of a upgrade over my aging Vishera FX 8320. 

Just hope that AMD release it soon......from what i heard Zen is basically done at this point, but there's some problems with the 14nm processing node.

Honestly, if it matches Haswell in IPC, i'd take it over a 6700k. If they really are making a consumer grade 8 core, 16 thread chip, yeah, sign me up. I just hope we see some AM4 ITX boards more than anything. The AM3 selection on ITX boards was god awful x.x

[Deletive]

Even if your numbers are off by +/- 10% this stil puts AMD in a bad spot because it puts them up against skylake and competing in a clock speed war against those chips cant end well, at least if all the initial overclocking impressions are to be believed. This just leaves core count and price and i dont think selling 8 core chips for $150 is a good business strategy.

 

This puts AMD in a great spot, because they can finally gain some ground in the Server space so they can actually do something. I would take ddr4 + haswell IPC and save 200-300$ per chip (lets say like 8 core) then yeah probably thats a great idea. Thats the success they really need, and this'll put them in a great spot, if Zen fails on consumer market well they could probably live through it anyways.

 

Me= Zen that's around Haswell or higher or lakes in the sky will be a hell of a upgrade over my aging Vishera FX 8320. 

Just hope that AMD release it soon......from what i heard Zen is basically done at this point, but there's some problems with the 14nm processing node.

It'll be out to Servers by late spring probably

[Clyne]

Honestly, if it matches Haswell in IPC, i'd take it over a 6700k. If they really are making a consumer grade 8 core, 16 thread chip, yeah, sign me up. I just hope we see some AM4 ITX boards more than anything. The AM3 selection on ITX boards was god awful x.x

Well as long it beats Haswell or around it in IPC, then it's a upgrade from my 8320.

If that happens, i'll try to give my 8320 one last run in oc before it blows up from fried VRMs....

[MageTank]

Well as long it beats Haswell or around it in IPC, then it's a upgrade from my 8320.

If that happens, i'll try to give my 8320 one last run in oc before it blows up from fried VRMs....

Heh, i'll join you. Got an 8320 and Gigabyte 990FXA UD3 lying around. HEAR US AMD? WE SACRIFICE OUR FX'S IN THE NAME OF ZEN! 

[Clyne]

Heh, i'll join you. Got an 8320 and Gigabyte 990FXA UD3 lying around. HEAR US AMD? WE SACRIFICE OUR FX'S IN THE NAME OF ZEN! 

Too bad my board is only 4+2.....but i'll have "extra cooling" around the motherboard....

[MageTank]

Too bad my board is only 4+2.....but i'll have "extra cooling" around the motherboard....

All i got is an old school Hyper 212 Evo. This thing served every Phenom CPU i've ever owned. I trust it to give my 8320 a good sendoff. 

[Clyne]

All i got is an old school Hyper 212 Evo. This thing served every Phenom CPU i've ever owned. I trust it to give my 8320 a good sendoff. 

Yep, or i may just keep the 8320 to compare Vishera to Zen in performance wise.

My goal=4.3GHZ at least.

[MageTank]

Yep, or i may just keep the 8320 to compare Vishera to Zen in performance wise.

My goal=4.3GHZ at least.

Not exactly impossibe. As long as your board is decent (i've seen a couple 970 boards with not so great VRM's reach 4.2ghz, though they required some aggressive cooling to remain prime95 stable). 

[Prysin]

My guess is you can put FM3 in FM2+ or if it is AM4 then in AM3+ but not the other way around probably...

FM3 was dropped A LONG time ago. All new desktop parts will be using AM4. backwards compatibility is not yet revealed, nor is it that likely to support older CPUs (due to DDR4 being in there).

Remember, the reason we have a transition from DDR3 to DDR4 like we currently do (skylake and ZEN supporting both) is because the price of DDR4 is still a tad too high. And the majority of people willing to upgrade sit on DDR3 RAM already...

[Litargirio]

If only we didn't have to wait another year for its release... 

[jos]

I ran the numbers through QEMU and a virtual CPU constructed with theoretical latencies based on AMD's 40% IPC improvement over Excavator claim. Using an integer linear program as the optimization measure, Zen will be within +/- 3% of Haswell clock for clock. That means AMD has to win on clock speed.

Is the 40% improvement on IPC is per core basis or considering parallelism in mind?

[Briggsy]

that HBM cache sounds interesting.

[patrickjp93]

Is the 40% improvement on IPC is per core basis or considering parallelism in mind?

IPC is always per core. Parallel scaling depends on IPC, but there are other factors as well including bandwidth, cache size & speed, and the structure of the workload itself.

[jos]

IPC is always per core. Parallel scaling depends on IPC, but there are other factors as well including bandwidth, cache size & speed, and the structure of the workload itself.

That is how it is usually done, but this is AMD, who is desperate now.. hence asked... but IPC is most important ... the cache speed and size will be provided to handle the IPC.. else there will be a lot of cache miss...there will be optimal amount.... even the bandwidth between processor and memory will be adequate.. which turns IPC being most important performance factor.. but for parallelism multi core architecture is important as well... 

[Nena Trinity]

I think it would be the opposite, you could put current FM2+ in the new socket, not vice versa. But even if they call it AM4 it is supposedly not a direct AM3 derivative, but derived from FM2 and will more closely reflect that socket.

That explains why it was first named FM3.

 

FM3 was dropped A LONG time ago. All new desktop parts will be using AM4. backwards compatibility is not yet revealed, nor is it that likely to support older CPUs (due to DDR4 being in there).

Remember, the reason we have a transition from DDR3 to DDR4 like we currently do (skylake and ZEN supporting both) is because the price of DDR4 is still a tad too high. And the majority of people willing to upgrade sit on DDR3 RAM already...

Yes I kinda expect it to be like AM2+ to AM3 that only AM3 works in AM2+ because AM3 supported DDR2 and DDR3.  (AM2+ had no DDR3 compatibility)

[themaniac]

I was close with the abbrieviation  of their own version of hyperthreading, i thought it would be AMT but its SMT, only 1 letter wrong

[Prysin]

IPC is always per core. Parallel scaling depends on IPC, but there are other factors as well including bandwidth, cache size & speed, and the structure of the workload itself.

can also be affected by instruction set

[Lauen]

Yep, or i may just keep the 8320 to compare Vishera to Zen in performance wise.

My goal=4.3GHZ at least.

 

All i got is an old school Hyper 212 Evo. This thing served every Phenom CPU i've ever owned. I trust it to give my 8320 a good sendoff. 

 

 

I wish I had some better cooling to see what I could do with my 8320, running 4.4 Ghz at 1.284 volts right now with a be quiet! dark rock pro 2 and I get about 62-63 degrees with Prime95.