Discussion AMD Ryzen MegaThread! FAQ and Resources

[-Fran-]

AnandTech's Review: http://www.anandtech.com/show/11658/the-amd-ryzen-3-1300x-ryzen-3-1200-cpu-review

TechReport's Review: http://techreport.com/review/32301/amd-ryzen-3-1300x-and-ryzen-3-1200-cpus-reviewed

Interesting sets of data in both.

Specially the topic around how much power Infinity Fabric is actually consuming compared to the in-core elements. For just being an intra-connect, it is quite power hungry. I see an opportunity there for AMD in order to optimize for power consumption. It also puts in perspective how efficient the Zen core really is in the overall picture. EDIT: The other theory is the disabled cores are still getting power, but that would be really suspicious. I wonder how AMD is fusing off the parts, if it's really fusing them off. I wouldn't mind another "core unlocking" fest

Cheers!

 

[MrN1ce9uy]

I don't believe either of those numbers on Userbenchmark are overclocked. I think those are just two samples. Very good scores, as Userbenchmark isn't the end all be all, it does pretty accurately place components where they should be.

One thing that I find interesting is the single/quad core performance of the 1300x exceeds that of the R5 1400 in both OC'd and non OC'd (1300X shows 3.5/3.6GHz, 1400 shows 4.0GHz). This leads me to believe the 1300x is possibly just using one CCX instead of 2 cores on each 2 CCXs. That, or the poster of said benches has access to some new mc/bios that isn't available yet to the public. Though I find it a little odd that a 1080ti was used in conjunction with the peak OC bench if AMD posted it (or maybe not...).

I think it is just that the cores w/o SMT are faster than w/ SMT. *Edit. But not according to the Cinebench single-thread scores on techpowerup.

 

[jdwii]

Well after reading several reviews 110$ Ryzen 1200 OC to 3.9Ghz and get close to I5 6600K at stock speed performance while coming even with a I5 7400 often. Also 1400-1500x simply seem like they keep getting worse in terms of performance per dollar as a 1600 can be had for just a bit more and you aren't losing much by saving 50$ and just getting a 1200 over a 1400. 50$ in budget builds can be huge.

This makes the I3 irrelevant for future budget gaming builds but i will still recommend them for basic usage as it has a igpu. G4560 is great if you can find one at its original retail price point even at 80$ its not terrible and probably easier for me to recommend for normal usage vs the I3.

 

[-Fran-]

Until the APUs come along, jdwii.

I have a feeling, given the information they've been gathering about the power consumption of the IF that APUs are going to be way more efficient parts than CPUs. The second CCX is no more and it will get replaced with a GPU, probably connected using IF to the CCX having half the L3, maybe? Oh, but then again they would still need the IF for RAM duties for both the GPU and CPU with NUMA stuff. I need to go back and read a bit more, lol.

Cheers!

 

[Eximo]

I also find the APUs more appealing. A decent quad core with decent onboard graphics. Might pick one up and see how it does as an HTPC. I've kind of been waiting for an AMD NUC equivalent. I did have an E350 board once, and it worked for a while until the general internet exceeded its performance.

I also have one of those Intel compute sticks, and it just isn't quite enough. Probably just use that as an emulator for old video games.

 

[jdwii]

I also find the APUs more appealing. A decent quad core with decent onboard graphics. Might pick one up and see how it does as an HTPC. I've kind of been waiting for an AMD NUC equivalent. I did have an E350 board once, and it worked for a while until the general internet exceeded its performance.

I also have one of those Intel compute sticks, and it just isn't quite enough. Probably just use that as an emulator for old video games.

I expect Ryzen 4 core APU 4mb L3 cache
4+0 CCX
Most likely iGPU RX 550 like performance

This is top model and it will probably be priced around ryzen 3 and be unlocked as well.

Take the 1200 and minus like 10% more performance at its stock speed currently and add a 550 and that is what i expect. TDP of a 550 is 50 watts minus GDDR5 probably shave off a few more watts and a 4 core 1 CCX 4mb of L3 cache Ryzen chip won't use much power at all guessing 35-45 watts max. Add that together and you get 95 watt TDP or less.

Memory performance won't be a massive concern but still a issue and i'm hoping that 2400mhz max limit isn't true for the APU's as the bandwidth in a 550 is 112GB vs like 38GB max on DDR4 2400 even 3200mhz is only 51gb and like always the CPU and GPU will be fighting for bandwidth.

 

[juanrga]

I also find the APUs more appealing. A decent quad core with decent onboard graphics. Might pick one up and see how it does as an HTPC. I've kind of been waiting for an AMD NUC equivalent. I did have an E350 board once, and it worked for a while until the general internet exceeded its performance.

I also have one of those Intel compute sticks, and it just isn't quite enough. Probably just use that as an emulator for old video games.

I expect Ryzen 4 core APU 4mb L3 cache
4+0 CCX
Most likely iGPU RX 550 like performance

This is top model and it will probably be priced around ryzen 3 and be unlocked as well.

Take the 1200 and minus like 10% more performance at its stock speed currently and add a 550 and that is what i expect. TDP of a 550 is 50 watts minus GDDR5 probably shave off a few more watts and a 4 core 1 CCX 4mb of L3 cache Ryzen chip won't use much power at all guessing 35-45 watts max. Add that together and you get 95 watt TDP or less.

Memory performance won't be a massive concern but still a issue and i'm hoping that 2400mhz max limit isn't true for the APU's as the bandwidth in a 550 is 112GB vs like 38GB max on DDR4 2400 even 3200mhz is only 51gb and like always the CPU and GPU will be fighting for bandwidth.

If the memory bandwidth is about one third of that in the RX 550, how could the APU have similar performance?

 

[jdwii]

I also find the APUs more appealing. A decent quad core with decent onboard graphics. Might pick one up and see how it does as an HTPC. I've kind of been waiting for an AMD NUC equivalent. I did have an E350 board once, and it worked for a while until the general internet exceeded its performance.

I also have one of those Intel compute sticks, and it just isn't quite enough. Probably just use that as an emulator for old video games.

I expect Ryzen 4 core APU 4mb L3 cache
4+0 CCX
Most likely iGPU RX 550 like performance

This is top model and it will probably be priced around ryzen 3 and be unlocked as well.

Take the 1200 and minus like 10% more performance at its stock speed currently and add a 550 and that is what i expect. TDP of a 550 is 50 watts minus GDDR5 probably shave off a few more watts and a 4 core 1 CCX 4mb of L3 cache Ryzen chip won't use much power at all guessing 35-45 watts max. Add that together and you get 95 watt TDP or less.

Memory performance won't be a massive concern but still a issue and i'm hoping that 2400mhz max limit isn't true for the APU's as the bandwidth in a 550 is 112GB vs like 38GB max on DDR4 2400 even 3200mhz is only 51gb and like always the CPU and GPU will be fighting for bandwidth.

If the memory bandwidth is about one third of that in the RX 550, how could the APU have similar performance?

Good point which is why i'm saying memory performance is going to determine everything with the APU's.

Once one lowers resolution and texture settings performance should start to improve. If only we got HBM on the APU's as a cache. Remember the TurboCache with nvidia 6200 series haha.

 

[juanrga]

I also find the APUs more appealing. A decent quad core with decent onboard graphics. Might pick one up and see how it does as an HTPC. I've kind of been waiting for an AMD NUC equivalent. I did have an E350 board once, and it worked for a while until the general internet exceeded its performance.

I also have one of those Intel compute sticks, and it just isn't quite enough. Probably just use that as an emulator for old video games.

I expect Ryzen 4 core APU 4mb L3 cache
4+0 CCX
Most likely iGPU RX 550 like performance

This is top model and it will probably be priced around ryzen 3 and be unlocked as well.

Take the 1200 and minus like 10% more performance at its stock speed currently and add a 550 and that is what i expect. TDP of a 550 is 50 watts minus GDDR5 probably shave off a few more watts and a 4 core 1 CCX 4mb of L3 cache Ryzen chip won't use much power at all guessing 35-45 watts max. Add that together and you get 95 watt TDP or less.

Memory performance won't be a massive concern but still a issue and i'm hoping that 2400mhz max limit isn't true for the APU's as the bandwidth in a 550 is 112GB vs like 38GB max on DDR4 2400 even 3200mhz is only 51gb and like always the CPU and GPU will be fighting for bandwidth.

If the memory bandwidth is about one third of that in the RX 550, how could the APU have similar performance?

Good point which is why i'm saying memory performance is going to determine everything with the APU's.

Once one lowers resolution and texture settings performance should start to improve. If only we got HBM on the APU's as a cache. Remember the TurboCache with nvidia 6200 series haha.

The only I can see can help to improve performance is the presence of L3 cache and the iGPU accessing the L3 on the CCX, but this L3 is shared with the iCPU.

 

[Rookie_MIB]

I just wanted to give some people a bit of information on something that I was digging up and trying to sort out with the Wraith Spire and Wraith Max fans and their RGB lighting. I know that some of the motherboards don't come with the RGB headers (such as the Gigabyte AB350M-A), and so our pretty ring lights don't seem to work.

And that makes me sad.

So - did a little testing. The AMD light on the fan lights up whenever you put power to the fan. So - how is that set up? There's a white LED right under it, and that's where the header for the RGB control wire goes. Soooooo. looking at the fan header, you see -two- red wires coming off the header, along with two black. At the fan, those two wires split off the cable and head to the input connection on the RGB header.

I took a multimeter, did a back voltage test, and lo and behold, from the wire coming off the fan, there are only three connections. I surmised those are R,G,B, as there is already a 12v positive going in from the fan header. Backtracing shows that there's a voltage differential coming off that between those three wires and ground.

Direct connecting each one of those RGB wires lit the appropriate ring color.

In summation - if you know which color you want, and you don't have an RGB controller on your Ryzen board but have an AMD Ryzen RGB cooler, you can hardwire the appropriate R/G/B wires directly to any ground and they'll light up. You'll have to experiment with the color.

Me? I've got some Arduino boards with some PWM controllable outputs. I think I'm going to write some programs for it and see what we can do.

 

[jimmyEatWord]

Does anyone here knows if captain 360 ex cooler by deepcool supports Ryzen ? i asked deepcool themselves but i thought i doublecheck on this forum.

 

[jdwii]

Does anyone here knows if captain 360 ex cooler by deepcool supports Ryzen ? i asked deepcool themselves but i thought i doublecheck on this forum.

Well i mean buying new you would need a bracket i doubt that cooler comes with one out of the box. I did read that they will give out AM4 brackets. If it was me i'd buy a AIO that was for sure compatible out of the box unless you want to run on the stock cooler for awhile waiting(which is what i had to do)

 

[Deadparty37]

I have a coolermaster masterliquid pro 120, it wasn't compatible out of the box, but it took a week and it was worth it in my opinion. I have a Ryzen 5 1600 but upgrading to a Ryzen 7 1700x within 2 weeks.

 

[jimmyEatWord]

Does anyone here knows if captain 360 ex cooler by deepcool supports Ryzen ? i asked deepcool themselves but i thought i doublecheck on this forum.

Well i mean buying new you would need a bracket i doubt that cooler comes with one out of the box. I did read that they will give out AM4 brackets. If it was me i'd buy a AIO that was for sure compatible out of the box unless you want to run on the stock cooler for awhile waiting(which is what i had to do)

i emailed deepcool they said the 360 captain with an "EX" is compatible , but i'm simply not sure since the cooler came before Ryzen . on the users manual it says AMD bracket nothing about AM3+ , so DOES ANYONE KNOW ? i just want to make sure before i buy any ryzen

 

[Embra]

Does anyone here knows if captain 360 ex cooler by deepcool supports Ryzen ? i asked deepcool themselves but i thought i doublecheck on this forum.

Well i mean buying new you would need a bracket i doubt that cooler comes with one out of the box. I did read that they will give out AM4 brackets. If it was me i'd buy a AIO that was for sure compatible out of the box unless you want to run on the stock cooler for awhile waiting(which is what i had to do)

i emailed deepcool they said the 360 captain with an "EX" is compatible , but i'm simply not sure since the cooler came before Ryzen . on the users manual it says AMD bracket nothing about AM3+ , so DOES ANYONE KNOW ? i just want to make sure before i buy any ryzen

My guess is that you might need to request a bracket. The only way to be sure is there is a AM4 is to open it and see must likely.

 

[randomizer]

More testing around the hardware bug some people think they found.

http://www.phoronix.com/scan.php?page=news_item&px=Ryzen-Test-Stress-Run

 

[-Fran-]

More testing around the hardware bug some people think they found.

http://www.phoronix.com/scan.php?page=news_item&px=Ryzen-Test-Stress-Run

https://svnweb.freebsd.org/base?view=revision&revision=321899

EDIT: Also this: https://www.reddit.com/r/Amd/comments/6rmq6q/epyc_7551_mining_performance/

Cheers!

 

[juanrga]

Phoronix test stress run can report false positives. The best option is to run the kill_Ryzen script; this will confirm if the bug is present in the hardware or not.

AMD is already RMAing CPUs to affected RyZen users.

 

[randomizer]

Yes, most of the segfaults were false positives. In the follow up article Michael did mention that the original test exposed some segfaults in Clang, which is a separate issue to the ones people pointed out as "normal".

 

[-Fran-]

And, more importantly, AMD has confirmed those issues are not present in EPYC. Or maybe they're lying to save face and patch them before delivering them to clients! Take whatever reading on the situation makes you happy.

Oh, before I forget. That reddit link I posted has updates stating the tests were failing in that EPYC CPU due to the PHP binaries that caused other issues elsewhere (other CPUs)... Or so I understood of the whole ordeal.

Smells like it can be fixed with uCode, but I would expect another revision just in case for any person looking to do a Linux Ryzen build for heavy duty.

Cheers!

 

[randomizer]

Yes, it's not an issue with EPYC or TR, only Ryzen, according to AMD. That's good, because those CPUs are the most likely candidates for the loads required to reproduce the bug in the wild.

 

[juanrga]

Although it is good that AMD finally admits the existence of the problem, I would point that the official communication is not correct when characterizes it "as a performance marginality problem exclusive to certain workloads on Linux."

The problem is not one of performance, but one of data corruption. Moreover, the problem has been reproduced under linux, windows, and BSD.

 

[-Fran-]

Segmentation Faults by default are "data corruption" problems. Depending on what data sets you're running at the time it happens and if there were any committed memory changes into perpetual state, you can have that. It's a big *depends* in my book, so I wouldn't give AMD a hard time for being broad on this particular problem. It's not the same nature as the TLB bug, for instance, or the Intel issue had back in the day when a single operation was causing a problem.

And it would be weird to not be able to reproduce it in Windows, TBH, but who does heavy work in Windows machines anyway

Cheers!

 

[randomizer]

Although it is good that AMD finally admits the existence of the problem, I would point that the official communication is not correct when characterizes it "as a performance marginality problem exclusive to certain workloads on Linux."

The problem is not one of performance, but one of data corruption. Moreover, the problem has been reproduced under linux, windows, and BSD.

Marketing translation

 

[TechyInAZ]

FYI: AMD Ryzen Threadripper MegaThread is now up and running! (Will be updated with more info on launch day.)

http://www.tomshardware.com/forum/id-3491295/amd-ryzen-threadripper-x399-megathread-faq-resources.html