catch up to intel on IPC

> catch up to intel on IPC
> implements crippling hardware bottleneck

Other urls found in this thread:

pugetsystems.com/labs/articles/Adobe-Photoshop-CC-2017-AMD-Ryzen-7-1700X-1800X-Performance-907/
pugetsystems.com/labs/articles/Premiere-Pro-CC-2017-AMD-Ryzen-7-1700X-1800X-Performance-909/
pugetsystems.com/labs/articles/Adobe-Lightroom-CC-2015-8-AMD-Ryzen-7-1700X-1800X-Performance-910/
pugetsystems.com/labs/articles/SOLIDWORKS-2017-AMD-Ryzen-7-1700X-1800X-Performance-908/
reddit.com/r/vmware/comments/5z25qp/any_information_about_running_vmware_in_ryzen/
cisco.com/c/en/us/products/collateral/switches/nexus-5020-switch/white_paper_c11-465436.html
twitter.com/SFWRedditGifs

This graph is useless without a legend

That's what AMD does.
Makes a powerful product on paper that has terrible efficiency in real world applications.

The furyX is a prime example of that
Fuck ton of horsepower, but 20% of the shaders were unused and sat idle most of the time.
Then combined with driver overhead, it was slower than it's competitors.

Pic unrelated

it's almost as if software that isn't optimized runs less efficiently on architectures it wasn't designed to run on

Inter-core/CCX communication is faster than intel
>KEK WILLS IT

Just AMD Things

Explain to retard

> hardware bottleneck
Lol no. Zen arch is designed for servers first you put 4 of those fuckers on a MCM and have 32 core monster.

>real world applications

And by that you mean video games.

I'm not buying a server CPU for my home computer.

>Zen arch is designed for servers

It's a compromise they made to keep power consumption and transistor count down, for highly parallel workloads where there is little inter-thread dependency the CCX design works perfectly well, and is a lot more scalable than Intel's approach of just chaining multiple ring busses together with a switch. Of course the downside to AMD's approach is clear by now, but for high core counts and enterprise applications, it's the better option.

Photoshop: pugetsystems.com/labs/articles/Adobe-Photoshop-CC-2017-AMD-Ryzen-7-1700X-1800X-Performance-907/

Premiere Pro: pugetsystems.com/labs/articles/Premiere-Pro-CC-2017-AMD-Ryzen-7-1700X-1800X-Performance-909/

Lightroom: pugetsystems.com/labs/articles/Adobe-Lightroom-CC-2015-8-AMD-Ryzen-7-1700X-1800X-Performance-910/

Solidworks: pugetsystems.com/labs/articles/SOLIDWORKS-2017-AMD-Ryzen-7-1700X-1800X-Performance-908/

>AMD won't shill its own product
No shit dumbfuck.

>Catch up to where intel was three years ago on IPC
>Just bad cause AMD
FTFY

>Merging the worlds of enthusiast, gamer and creator
>IT'S A GAYMING CPU DUDE
Kill yourself

>It's a compromise they made to keep power consumption and transistor count down

Really makes you think

Unfortunately the industry doesn't care about your personal computer anymore, the big money is in servers.

Yeah, it uses less power than Intel's octa core. If you're comparing a quad core to an octa core in terms of power usage, you are of course retarded.

DELETE

Intel cares

>8 core CPUs use more power than 4 core ones
Are you braindead? It has the lowest idle power consumption and uses less power than its 8 core Intel competitor under load.

>Ryzen 8 core uses less power than Intel's 8 core while on an inferior and less mature 14nm designed for mobile SoC, and running last it's optimal frequency. Even in a cherry picked benchmark Which shows Intel in the best light possible
It really does make you ponder

How come the 7700k is both faster and uses less power

Well lets hope there is a way for game devs to code around this. AMD is seeming to focus a lot more on game dev support for Ryzen and hopefully this will pay out soon.

(you)

because it got its clock speed cranked, its developers are familiar with its architecture.

This is without even regarding that there are a few applications and programs that the 7700k is already getting beaten in by 8 and even 6 core haswell-e's. Despite both running significantly slower clock speed wise.

Because it has only 4 cores, but with better singlethreaded performance. And it's only faster in poorly threaded workloads, as expected.

>HOW COME THE 7700K IS FASTER THAN THE 6900K AND USES LESS POWER? LMAO INTEL FINISHED XDXDXDXD

The Zeppelin die is CLEARLY designed for HPC and server workloads.

Profit margins are MUCH higher in the enterprise market.

AMD cannot currently afford to put out separate dies for the home and enterprise market and the enterprise market is FAR more important to to their survival as a company than the home market.

Small dies + many core scalability is KING in enterprise.

Selling to home users is just icing on the profit cake. It is however not their primary focus.

4 cores

>Nobody should advertise to a market unless they're the absolute best of the best in that market
wew

Does this mean the Ryzen 6 cores are a complete joke?

It means Microsoft needs to fix there scheduler to prioritise keeping workloads on the same CCX.

...

Take the graph above and take off cores 0, 1, 14 and 15.

That's the 6/12 ryzen

The 6 core versions won't have any L3 cache disabled, so having more L3 cache per core should let offset some of the performance losses.

ping within CCX is faster than Intel. keeping threads mostly within CCXs is a software problem.

AMD confirmed there is zero problems with the scheduler

Layman here

Is ryzen basically two cpus on one die and the issue is latency when the two cpus have to communicate?

They confirmed the issue isn't with how the scheduler is handling SMT. The issue is with how it is handling 2 separate CCXs.

Pretty much.

Yes, though it has less latency than two seperate cpu's would have.
Well, the same report that op's pic is from confirmed that the scheduler does actually try and keep threads within the same ccx as much as possible

So what will you fall back on when they confirm that is not an issue either?

Basicly yes.
It isn't optimal in a lot of end user applications but for enterprise the design is brilliant.

This isn't a architectural issue. AMD could make a single CCX 8 core if they wanted to. The issue is AMD cannot currently afford to make separate dies for the home and enterprise markets. Thus the less profitable of the two (home) is a secondary concern.

AMD wants there yields high and there margins large as they retake the HPC and server market.

>adobe
invalid
>adobe
invalid
>adobe
invalid

Adobe has notoriously poor multithreading support in many of their products, these benchmarks mean nothing. Look at results for well designed workloads, like h.264 encoding. All your bench's show is that Adobe don't have good support for multithreading.

It is obviously an issue. Inter CCX communication is causing latency spikes. Communication between cores inside the CCX is actually faster than Intel's.

Ok yeah cool
So back to my original question, what excuse should we come up with when they confirm it's not an issue?

delusional amd cucks, everyone

It is a known issue.

>but what if it isn't? XD

Kill yourself m8.

Why would the one 8 core cpu solution be worse for enterprise?

Oh yeah sure it is *wink* *wink*
But we really should get our story together when they find out it isn't, should we blame the software again or stick to blaming microsoft cause I think that will work better.

The 6900k is worse than the 7700k in those same programs. You are clearly delusional if you think Adobe software is highly parallel.

wew lads. now ryzen isn't even a chip for content creators, a at-home workstation chip, but a server chip. wew lads. you make me feel like not waiting for r3.

anyways, ryzen 2 will probably have this fixed, which is cool. better hardware incoming. and ryzen 4 cores (basically one ccx off) are gonna be cheap and great, seemingly.

We go with architectural decision and they are willing to take the latency penalty.

>Videogames and all adobe applications foesnt matter for a gpu
The fuck would you use an amd gpu for? Certainly not cuda developmrnt, lmao. Kys promptly

>cancerous avatarfag
>amdpajeet
like pottery

lower yields and and performance per watt.

>The Zeppelin die is CLEARLY designed for HPC and server workloads.
Which is why it still doesn't work with VMware?

reddit.com/r/vmware/comments/5z25qp/any_information_about_running_vmware_in_ryzen/

So we should say that AMD is so advanced that software can't keep up?
Just too smart to win? I feel like that's a good excuse, cause the whole memory thing didn't work well last time and I feel like this excuse will last longer.

Install gentoo, if you're not using a server os you are at the wrong board.

>Esxi 6.5 purple screens on install because it cannot identify CPU.
>So its going to require an update.

What's it like, the life of someone who can't read?

>posting pictures with your post is avatar faging

You don't know the meaning of the word.

No, we say install gentoo.

why dont you tell me?

>With AMD popularity right now it will take a while.

The most popular hypervisor doesn't support it, even though engineering samples have been around for a long time. That has to tell you something.

>So we should say that AMD is so advanced that software can't keep up?
No
What i am saying is that isn't the workload it was built for. Latency is unimportant in server/HPC workloads.

So we should let everyone know they are just using the wrong workloads with their 500 dollar processor? I feel like that one will last a while, just poke around and find something that performs decently and say that's what the R7 was designed for. Anything else and your using it wrong.

Engineering samples being around a while doesn't mean they had access to them. Not even the motherboard manufactures had final silicon until a month before launch.

>B b b but the Intel is bad too!
AMD is worse though.

Amd is in a financial ruin and has to sell server CPUs on the regular consumer market

>amdfags are literally "holding it wrong" tier

>Latency is unimportant in server/HPC workloads.
Which is why ESXi has a latency sensitivity option for VMs? Pic related. And why there are two types of ethernet switches, cut through and store and forward, the former being for the purpose of reducing latency for HPC workloads:

cisco.com/c/en/us/products/collateral/switches/nexus-5020-switch/white_paper_c11-465436.html
> Furthermore, cut-through switches are more appropriate for extremely demanding high-performance computing (HPC) applications that require process-to-process latencies of 10 microseconds or less.

>largest virtualization company
>wouldn't have access to engineering samples
>for a chip you tards are saying is targeted towards enterprise class workloads.

OR maybe you should look at the benchmarks and purchase the correct software for your workload instead of being retarded.

>cpu benchmarks, about amd's new ryzen cpu, tested against intel cpu's
>"hurr durr but gpu's though"

but amdfags were tellling me that the desktop market doesn't matter and are going to make a killing in the server
why would they market it to gaymen then

>>for a chip you tards are saying is targeted towards enterprise class workloads.
R7 is not Naples. It was explicitly intended to target the enthusiast and content creator market and compete with Intel's i7s which are in the same tier.

Exactly. Chip level latency is insignificant when network level latency is involved.

YES!
THATS THE ONE!
Tell them they bought it wrong

It's irrelevant.
That's like trying to say a rugby ball is better than a football for playing vollyball.
Who cares?
You shouldn't be using either of those balls to play volleyball in the first place..

They have to sell something to the average joe consumers, they don't have money to design CPUs that best suit the average joe so they just sell the server CPUs for every market

>latency doesnt matter guise
>its not targeted for enterprise class workloads even though we were saying it was earlier
>its for gaymers and content creators
>which is why we're even having this discussion of how the task scheduler is so fucked up in the first place

If someone is dumb enough to try to eat steak with a spoon i am going to tell them they are using the wrong tool for the job.
Is a flashlight a bad tool because it is bad at turning bolts?

Yes

>It's faildozer 2.0

So it almost seems like it's operating like a dual quadcore system rather than an octacore system.

Tell me again how IBM's Power9 is bad at gaming and therefor a bad chip.

Yes? Somebody should make an 8 core vs 4 cores disabled benchmark for gayyming to see how much the task scheduler is fucked and if AMD is right for the tradeoff.

sauce now

But does latency actually matter? Do you have any real world examples where cache is bottlenecking the system?

Looks like you have ton of Meymey arrows and no actual proofs

Power 9 is not sold to gamers and average joes

No. What are you, retarded?

>But does latency actually matter?
Yes, why I have latency sensitivity set to high for my RDS VM. Otherwise video and vidya stutters on playback/playing.

literally kill yourself
I don't even know where to begin explaining how retarded you are.

He is right though

>vidya gaymen

Wow! Tell us more!

If he is right, that means Intel has four single cores

another thread where Sup Forums tells us how important vidya games are to the server/workstation market

No

Ryzen is basically two quad cores on a single die

With both quad cores having 8 threads

More like how Intel C2Q was two dual cores in a single chip.

This. Hilariously c2q were better than amd "true" quad core phenom1/2

can I make one ccx Linux and the other windows?