AMD Zen “RYZEN” CPUs Detailed – 8 Cores, 3.4Ghz+ & Auto Overclocking With “XFR”

[LTC8K6]

We really need some ST benches to settle this, all this speculation isnt going to get us anywhere.

It does not sit well that they were avoiding ST like the black death though with this initial press release. However i can see that they look to be still refining the turbo and final clocks so i can see their point in waiting until that is finalized, before giving us the ST benches.

There's no reason they couldn't lock a BW-E chip and RYZEN at 3.0ghz and single core and run a bench, though.
Final clocks and turbo don't really matter for that.

We didn't need to see anything but that to give us an idea of IPC.

 

[LTC8K6]

Why does RYZEN only have dual channel ram? Did anyone clarify?

 

[raghu78]

Why does RYZEN only have dual channel ram? Did anyone clarify?

Because Ryzen uses the AM4 socket which is dual channel. AMD already said that Zen will use the same socket as current Bristol Ridge APUs.

http://www.anandtech.com/show/10705...e-and-am4-analysis-a12-9800-b350-a320-chipset

 

[sirmo]

Why does RYZEN only have dual channel ram? Did anyone clarify?

Dual channel.. cheaper motherboards though, less layers needed.

 

[LTC8K6]

Dual channel.. cheaper motherboards though, less layers needed.

It does come with some ram limits, though. It seems odd to limit what looks like a very good chip.

 

[TheELF]

To me it's interesting that months ago they showed us a fairly close performance in Blender only to show us a pretty hefty lead in Handbrake only a few days ago. If they were cherry picking performance you'd think they would have showed Handbrake first. I think they could be holding back on the performance rather than the lack of.

And next time they might show us 7zip or fritz benchmark and we will end up thinking that zen will have 20% more IPC then sky/kaby...
I'm not talking about cherry picking as such,but avoiding any ST test should be a red flag for anyone.

If they really are that close in ST then boosts and such would be irrelevant they would show a test, without numbers just like the blender test, just to brag about it.

 

[sirmo]

And next time they might show us 7zip or fritz benchmark and we will end up thinking that zen will have 20% more IPC then sky/kaby...
I'm not talking about cherry picking as such,but avoiding any ST test should be a red flag for anyone.

If they really are that close in ST then boosts and such would be irrelevant they would show a test, without numbers just like the blender test, just to brag about it.

If you're building a fast car, and say you just got it to start, and drive around, but you done no tuning on it. Do you run out and show your 0-60 results then, or once you're done tuning it?

You wait till you're done for maximum impact/best result.

Also think about this product and what it means to AMD. 4+ years in development, you want to show it in the best light possible. This is true whether the end result impresses or underwhelms.

 

[cytg111]

I wouldnt call it a red flag, just a mental note.. wont be surprised if st does not stack up as well as mt indicates so far. is what it is.

 

[krumme]

And next time they might show us 7zip or fritz benchmark and we will end up thinking that zen will have 20% more IPC then sky/kaby...
I'm not talking about cherry picking as such,but avoiding any ST test should be a red flag for anyone.

If they really are that close in ST then boosts and such would be irrelevant they would show a test, without numbers just like the blender test, just to brag about it.

Well if they showed 8c Zen beeing 20% faster in zzip and fritz at same freq as bwe 8c i would not be so worried about st perf. And neither should you then i would start to worry about price...
But hey i would just buy a cheap 300 usd 8c bwe then. Fine for me.

 

[krumme]

I wouldnt call it a red flag, just a mental note.. wont be surprised if st does not stack up as well as mt indicates so far. is what it is.

Naa dont worry to much. This processor will be plenty fast and attractive vs i7. The problem arives if its to efficient. Then servermarket will take the capacity from us and prices will stay high. So lets hope for a little sloppy efficiency.

 

[cytg111]

Well if they showed 8c Zen beeing 20% faster in zzip and fritz at same freq as bwe 8c i would not be so worried about st perf. And neither should you then i would start to worry about price...
But hey i would just buy a cheap 300 usd 8c bwe then. Fine for me.

That will be the real downside to all this if zen stacks up, cause 6-8 cores will be mainstream real fast and all us with recent quad rigs, even i7s is sorta left behind. I thought this 4770 was going to be a ten year rig...

 

[sirmo]

That will be the real downside to all this if zen stacks up, cause 6-8 cores will be mainstream real fast and all us with recent quad rigs, even i7s is sorta left behind. I thought this 4770 was going to be a ten year rig...

At the same time though it will be nice to shed the single thread bottleneck, once multithreaded (Vulkan/DX12) becomes a norm.

 

[krumme]

At the same time though it will be nice to shed the single thread bottleneck, once multithreaded (Vulkan/DX12) becomes a norm.

Yep. It was like all games looked like doom 3 on dx9 for 6 years. Then dx11 for like another 6. Now it aparently takes 6 years for dx12 vulcan to get there too so we can use the cores and some rts that is different than Dune 2 from 1995. I thought those consoles would speed up the process but it seems it takes forever as usual.

@CyTG Worried about 4c i7. Wait until ps5 with 8c zen arives that should wake some pc programmers

Edit. Until then i recon an i7 4770 will be fine. 3 more years.

 

[sirmo]

Yep. It was like all games looked like doom 3 on dx9 for 6 years. Then dx11 for like another 6. Now it aparently takes 6 years for dx12 vulcan to get there too so we can use the cores and some rts that is different than Dune 2 from 1995. I thought those consoles would speed up the process but it seems it takes forever as usual.

@CyTG Worried about 4c i7. Wait until ps5 with 8c zen arives that should wake some pc programmers

Edit. Until then i recon an i7 4770 will be fine. 3 more years.

It's the engine makers.. they need to put the effort in. And they are, it's just taking time. Unity, probably one of the most popular 3d party engines, just released their Vulkan beta, with some really nice gains.

 

[krumme]

It's the engine makers.. they need to put the effort in. And they are, it's just taking time. Unity, probably one of the most popular 3d party engines, just released their Vulkan beta, with some really nice gains.

Yeaa saw that. Good news. Still games is then 2 years out. And look at bf1. Bad usage imo vs consoles. Dx12 is crap 2 years after mantle bf4. Its sloppy. But when ps5 get there - and i asume its 8c zen like - pc games is damn forced to be improved all over. But its still 3 years out. Zzzz.

 

[TheELF]

Yeaa saw that. Good news. Still games is then 2 years out. And look at bf1. Bad usage imo vs consoles. Dx12 is crap 2 years after mantle bf4. Its sloppy. But when ps5 get there - and i asume its 8c zen like - pc games is damn forced to be improved all over. But its still 3 years out. Zzzz.

Why will it be forced to improve?
If the zen cores really are as fast as a lot of people claim then coders will be less forced to use good multithreading,they will go back to being (even more) lazy ba*rds and start doing more things in single threaded again.
(Unless this is what you meant. )

 

[sirmo]

.. (oops please delete, wrong thread)

 

[Dresdenboy]

I somewhat agree with TheELF here, either way you put it if we put relative performance of a 8c/16t zen equal to a 8c/16t broadwell, then it goes to reason that for the given workload, if HT threads on the Zen platform pulled a larger utilization level than the equal HT onthe Intel platform, then Zen ST performance will be lower.

In some scenarios it is that simple, and it could be expressed in a simple formula. But since Zen's SMT yield in Blender lies somewhere between -100% and +100%, there is a wide field for speculation about ST performance.

BTW with not too many memory references (AGU+L/S contention), FP code can see nice SMT yields AND good ST performance for the same code. Explanation: FP ops have multi cycle latencies. If there are only one or few dependency chains in the code, these latencies offer enough opportunities for another FP thread to slip in. Another source of opportunities are cache misses, causing long stalls. And power constraints in MT mode may hide the true SMT yield. Example with CB R15:
http://www.pcworld.com/article/3039...es-you-really-need-for-directx-12-gaming.html
2C -> 2C HT: SMT yield 37%
8C -> 8C HT: SMT yield 32%

Does 37% yield indicate bad ST performance for Haswell?

Even SB with all cores had some nice SMT yields for FP heavy stuff. Sometimes also just 0% -> no correlation with ST performance.

 

[Dresdenboy]

I never said it where CMT I said it was like CMT.

No it's not like HT,in HT the "partitioning" happens by the threads themselves (it's not even partitioning every core just takes what it can)and the cores are build for speed.

In amds version, senseMI will be trying to do the partitioning in the best way possible for the best performance based on the workload and the cores are build for throughput and not speed.

At least that's how I see it.

There surely will be some adaption to workloads. What exactly reminds you of CMT? Separate Int/AGU and FPU clusters? This is the old coprocessor model. Based on these clusters, Zen has roughly 70% less separated resources than XV.

 

[Dresdenboy]

AMD may define it as they want, but if Blender produces a freaking 94W worth of power delta, 95W is not anywhere near max power draw of Zen. Unless they have gutted vectorization on it.

Did you remove DIMM power and PSU/VRM inefficiency?

 

[Tuna-Fish]

It does come with some ram limits, though. It seems odd to limit what looks like a very good chip.

Increasing the width of the memory bus directly and nonlinearly increases the manufacturing cost of the system. For mainstream products, dual channel is the only sane choice.

They do have an 8-channel server platform in addition to the dual-channel mainstream one, I would be surprised if they didn't release some chips branded for workstation use for those who need ridiculous amounts of ram. Alternatively, they could just buy Opterons...

 

[DrMrLordX]

Naa dont worry to much. This processor will be plenty fast and attractive vs i7. The problem arives if its to efficient. Then servermarket will take the capacity from us and prices will stay high. So lets hope for a little sloppy efficiency.

I suspect that at least the top SKUs will be relatively high-leakage parts. Not 9590-class stuff but . . .

 

[Ajay]

It does come with some ram limits, though. It seems odd to limit what looks like a very good chip.

Now that we have 16GB modules, the limit is 32GB for two slot config and 64BG with 4 slots (and some extra circuitry for higher end boards) - not much of a limit for desktop users.

 

[deasd]

I somewhat agree with TheELF here, either way you put it if we put relative performance of a 8c/16t zen equal to a 8c/16t broadwell, then it goes to reason that for the given workload, if HT threads on the Zen platform pulled a larger utilization level than the equal HT onthe Intel platform, then Zen ST performance will be lower.

It's good question, does SMT in Zen have higher ultilization?

IF this is the case, Zen's power consumption might be much higher when SMT is on, thus it would have potencially less efficiency compared to Intel's. For example, BD/PD/SR has significantly higher consumption when go from 2C to 2M4C.

But AMD's own demo tell us, Zen's power consumption is not high. This COULD imply a low utilization of AMD's SMT.

Just a thought, I could be wrong.

 

[Ajay]

Yup it does not,but AMD avoiding ST tests like the pest although this would be their biggest and best argument for people to switch to zen makes me personally think that ST will be confined.

I do have to agree with that - I definitely find it odd. If ST is no where close to BW-E, then I don't expect we'll find out first from AMD, it will probably take a leaked benchmark.
There is much more we need to know.