AMD Zen processors...Are you interested?

[moremana]

its quite impossible for a 4.0 or sub-4.0 Zen to get better minimums in WoW than a 6700K .. esp if the 6700K is OCed

in BF1 MP maybe

In wow I cant really tell if its better minimums, it does run wow smooth as hell though.

- - - Updated - - -

My point is that unless a 1700 is much better than what you are currently using there is no point in buying one hoping that it will be better than what you are using in the future. Then it's better to wait and spend the same amount of money for a Ryzen 2 which has been confirmed for early 2018. At that time motherboards will be fixed, the windows scheduler will be improved and AMD will have fixed some of the kinks of the Ryzen cards.

But ...... thats no fun.

[Fascinate]

On a side note, my vizio 55" 4k tv just blew half of its backlight lol:

[moremana]

oh my, didnt you just buy that?

- - - Updated - - -

Meanwhile my 5960x at 4.5ghz runs at 1.3v - That's some terrible voltages for a low clock speed.

Then again it's AMD. Always slow and always hot.

I love these kinds of posts.

SMH

[Fascinate]

Yup lol, at least its under warranty i guess.

- - - Updated - - -

oh my, didnt you just buy that?

- - - Updated - - -



I love these kinds of posts.

SMH

Whats highest you got your ram to post at? Asrock has a beta bios to do with memory speeds, hoping my 3200mhz stuff will work right away.

[moremana]

Yup lol, at least its under warranty i guess.

- - - Updated - - -



Whats highest you got your ram to post at? Asrock has a beta bios to do with memory speeds, hoping my 3200mhz stuff will work right away.

2666, I didnt try to oc memory. Im going to wait until some bios updates come out. I spent so much time on ocing today that Im burnt. Im waiting to see if NE gets this in 3200

[chronia]

It is 2 extra cores, it is going to run hotter.

If you run virtual machines then the choice is a no brainer, 2 extra cores is huge.

This really depends on workload though, in both my Homelab (where i test stuff like Windows releases, Hyper-V, vSphere and related VMware products, Exchange, MSSQL, System Center suite and more of these types of applications and at in production enviroments at either on-premise at the customer of in datacentres CPU power is almost never the limiting factor, most of the times either Memory capacity per node or the amount of available storage IOPS are the limiting factor in the number of VM's that can be deployed per node (memory) or in the total cluster (Storage IOPS) cpu power is only rarely the limiting factor.

For some situations extra cores can be a factor, but ppl generally underestimate how much you can oversubscribe cpu cores in Virtual enviroments depending on workload (VMware has had ratio's of 25:1 vCPU to pCPU without performance degradation, while this a extreme example, 10:1 to 4:1 are more common depending on light to heavy workloads) while you generally don't want to overcommit memory or storage IOPS.

[Gray_Matter]

For some situations extra cores can be a factor, but ppl generally underestimate how much you can oversubscribe cpu cores in Virtual enviroments depending on workload (VMware has had ratio's of 25:1 vCPU to pCPU without performance degradation, while this a extreme example, 10:1 to 4:1 are more common depending on light to heavy workloads) while you generally don't want to overcommit memory or storage IOPS.

We have a 50 odd core server for our testing setup and we oversubscribe 2:1. HD's we can do 5:1 and SSD's 15:1.

[chronia]

We have a 50 odd core server for our testing setup and we oversubscribe 2:1. HD's we can do 5:1 and SSD's 15:1.

What do you mean with oversubscribing HDD's and SSD's, like runnin 15 VM's on 1 SSD, and 5 VM's on 1 HDD? Because thats not what i mean (and is pretty normal).

What i mean is, that you don't want to oversubscribe the total capacity in IOPS of the whole storage layer presented to your hosts, for example if you have a SAN that can deliver 4k IOPS under normal use (lets say a pretty much standard run of the mill SAN with 24 SAS 15K RPM drives and no SSD cache or tiering). you will have full performance up to the point that you stay under 4k IOPS. Once you go over 4K IOPS the performance of the SAN wil collapse due to oversubscribing / overcommiting and you will need to either lower the density of the VM's on that SAN or expand / replace the SAN to keep performance.

While with vCPU to pCPU ratio you can easily in most cases assign between 16-40 vCPU's if you have a 4 core pCPU host (unless you are running really funky workloads, then 2:1 might apply)

And generally in a virtual cluster its way more common to run into either Memory limits or the IOPS limits of the storage layer than that you run into the limits of the vCPU to pCPU ratio (again unless you run really cpu intensive workloads then i can be otherwise, but that is not the norm).

For example alot of the vSphere machines i manage have less than 70% CPU usage at any given time, but i cannot run more VM's on them because they are at their memory limit (384GB for HP DL380 Gen7 machines out of my head and 768 for HP DL380 Gen8 and 9 out of my head (Gen9 could be 1TB though, not 100% sure).

[Gray_Matter]

What do you mean with oversubscribing HDD's and SSD's, like runnin 15 VM's on 1 SSD, and 5 VM's on 1 HDD? Because thats not what i mean (and is pretty normal).

What i mean is, that you don't want to oversubscribe the total capacity in IOPS of the whole storage layer presented to your hosts, for example if you have a SAN that can deliver 4k IOPS under normal use (lets say a pretty much standard run of the mill SAN with 24 SAS 15K RPM drives and no SSD cache or tiering). you will have full performance up to the point that you stay under 4k IOPS. Once you go over 4K IOPS the performance of the SAN wil collapse due to oversubscribing / overcommiting and you will need to either lower the density of the VM's on that SAN or expand / replace the SAN to keep performance.

We can't use a SAN for our testing because it would kill the SAN. It's not really practical to look at it from a IOPS point of view because load isn't uniform. We have periods of very high load and then periods of no load at all.

While with vCPU to pCPU ratio you can easily in most cases assign between 16-40 vCPU's if you have a 4 core pCPU host (unless you are running really funky workloads, then 2:1 might apply)

And generally in a virtual cluster its way more common to run into either Memory limits or the IOPS limits of the storage layer than that you run into the limits of the vCPU to pCPU ratio (again unless you run really cpu intensive workloads then i can be otherwise, but that is not the norm).

For example alot of the vSphere machines i manage have less than 70% CPU usage at any given time, but i cannot run more VM's on them because they are at their memory limit (384GB for HP DL380 Gen7 machines out of my head and 768 for HP DL380 Gen8 and 9 out of my head (Gen9 could be 1TB though, not 100% sure).

We barely get away with 2:1 at times. We have a lot of heavy multit-hreaded code and the testing hammers the CPU. You are right, though, memory is the biggest limiting factor.

[Life-Binder]

http://www.zolkorn.com/en/amd-ryzen-...-by-core-en/3/
http://www.zolkorn.com/en/amd-ryzen-...-by-core-en/5/

4c/8t 4.0 Zen vs 4c/8t 4.0 Kaby

Core by Core, MHz by MHz with AMD RYZEN 7 1800X vs Intel Core i7-7700K [English]
Mar 8, 2017

Final Thought

In the end, we get the answer is more clearly from results as we have seen at all. Even that does not mean it can be confirm in 100%, but it is enough for some guidelines which allow us to assess what has been from what we have or from what we can do right now. From what we can do the test with the MHz by MHz measured in term of face to face as a result, it would make it easier to us for assess what performance or power per MHz of RYZEN’s IPC by estimate

What we can know today from all results we have seen ? Firstly the power or performance of RYZEN’s IPC It will still be behind the Intel 7th Generation or Kaby Lake approximately 10-15% if we have refer from what we have done today and if we do directly proportional to the extrapolation relative in a simple term with previous version of Intel’s Haswell which that we will know that it is effective between Kaby Lake is about 5-10% or maximum around 15%, so maybe we can say for a simple way the performance of RYZEN that close to Haswell CPUs from Intel or comparable to 4th Generation because today we have seen RYZEN is behind Kaby Lake about 10-15% also. If asking to me it was okay ? I would say it’s OKay or good enough for AMD with a good start even it still can not beat or won in competition with Intel but one thing we continue to see the result is the same working of the Multi-Thread in the form of an increase in the proportion of single RYZEN core of AMD still do better than Intel.

[mrgreenthump]

http://www.zolkorn.com/en/amd-ryzen-...-by-core-en/3/
http://www.zolkorn.com/en/amd-ryzen-...-by-core-en/5/

4c/8t 4.0 Zen vs 4c/8t 4.0 Kaby

Not to poke at their results, but they clearly don't know what IPC means. ;D

Anywho, yes.. If their 4c/8t will be at 4 GHz, it will be around 15% slower in gaming than a 7700K for ~60% of the price plus AM4 is cheaper than Z270 boards. And that is probably the worst case scenario, because we might see some optimization for Ryzen in gaming by the time Ryzen 3 and 5 drop. Not to mention Microsoft is finally pushing their Ryzen patch through.

[Life-Binder]

15% slower in gaming

4.1+ GHz vs 5.0 GHz + Zen having up to ~10% worse IPC in games than Kaby + actual R3 Zen having 8MB L3 cache instead of 16MB as this simulated 4c/8t

thats gonna come up to more than 15% difference unless they manage to get R3 to ~4.4-4.5 OC

[mrgreenthump]

4.1+ GHz vs 5.0 GHz + Zen having up to ~10% worse IPC in games than Kaby + actual R3 Zen having 8MB L3 cache instead of 16MB as this simulated 4c/8t

thats gonna come up to more than 15% difference unless they manage to get R3 to ~4.4-4.5 OC

Was comparing stock clocks.. Because frankly, we overclockers are the minority.

Also didn't notice the L3 cache thing, because the benchmarks I saw before had 8 MB disabled, according to them. But if said cache was not disabled, these results can just be thrown out, due to the scheduler problem.

[Evildeffy]

4.1+ GHz vs 5.0 GHz + Zen having up to ~10% worse IPC in games than Kaby + actual R3 Zen having 8MB L3 cache instead of 16MB as this simulated 4c/8t

thats gonna come up to more than 15% difference unless they manage to get R3 to ~4.4-4.5 OC

Except the vast majority with those chips do not overclock and thus it'll be 15% tops.

If you push average overclocks it'll be 4,9GHz vs. 4,0 since the top model should be clocked that.

That said unlike what I believe the 6C/12T will be (should have identical OCing potential to 8C/16T due to likely disabled cores in the CCX-es) the 4C/8T CPU could be a different die as the APUs should be taking that.
And if it's a different die (in essence the complete absence of 1 CCX) then the OCing potential could be far higher.

Majority of the world you compare stock vs. stock.. not overclocked vs. overclocked.
Many people just want the best possible stock performance you can get.

[Deleted]

Was comparing stock clocks.. Because frankly, we overclockers are the minority.

Also didn't notice the L3 cache thing, because the benchmarks I saw before had 8 MB disabled, according to them. But if said cache was not disabled, these results can just be thrown out, due to the scheduler problem.

indeed, also those many that overclock don't push the clocks to the limits, mean my 5820K is on 4.3 Ghz, thats not its limit, thats my limit, its running that speed at 1.15 volts, I can easily choose to try 4.5 or heck, 4.6 - 4.7 Ghz if I wanted to sacrifice noise but I won't be doing that, to get 5 Ghz on Kabylake, you need to invest in a very strong cooler and set fans higher then what I would be comfortable with, I always keep my fans on the lowest settings yet my chip gets around 60-65 degrees C at full load.

[Deleted]

4.1+ GHz vs 5.0 GHz + Zen having up to ~10% worse IPC in games than Kaby + actual R3 Zen having 8MB L3 cache instead of 16MB as this simulated 4c/8t

thats gonna come up to more than 15% difference unless they manage to get R3 to ~4.4-4.5 OC

These tests were down on windows 10 prior ryzen patch, so first question is should we trust them?
Also I don't think anybody expected amd to outright win against intel in pure single threaded performance. They did achieve what they set out to do and that is provide good performance with good power consumption for good dollar. Mission accomplished.

[Artorius]

Actual IPC benchmarks from someone who has a clue or two:



Individual tests

[Fascinate]

Still a decent upgrade in IPC over my 2500k, and 4x the threads to boot.

[Fascinate]

My entire build was under a grand lol. I could have dropped that even lower but put a bit of an emphasis on aesthetics this time around.

[Artorius]

I'd expect the 6/12 and 4/8 SKUs (all Summit Ridge) to be manufactured on the same 14LPP process as full Zeppelin, but maybe they shift it to 10LPE/LPP with Raven Ridge or perhaps even TSMC's 16FF+. Would give us a measurable Fmax increase of more or less ~300MHz if I had to guess it.