AMD RYZEN Builders Thread

[CuriousMike]

I use you as my BIOS update checker.
I'm not sure why you say latency is better; the prior AIDA screenshot you posted shows a faster memory latency by 1ns.
The L3 cache numbers look improved on this compared to your last screenshot.

ASUS Released the BIOS 0606 for Prime B350 Plus. Thats 3 BIOSes in less than a week. Wow.!

Probably AIDA is still reporting something wrong. but the Latencies are drastically improved.

 

[guachi]

It's useful for quick and dirty, noob level overclocking, especially with all of the early BIOS wonkiness. It just works for that use case. You can OC and then close it. Outside of that....you are most definitely spot on.

I had to use it to OC because my Gaming 3 is pretty terrible. Open, set a few things, apply (reboot, if necessary), close. Done!

It just works.

 

[Markfw]

Chrome has a built-in translator, works very well.

I would love to use chrome, its becoming my primary. But on this computer, I have so many remembered passwords, and bookmarks, that I just put up with it,.

 

[Malogeek]

I would love to use chrome, its becoming my primary. But on this computer, I have so many remembered passwords, and bookmarks, that I just put up with it,.

I believe they can all be imported. And with your Google account linked into Chrome, you can login to any Chrome anywhere (including your phone), and all your passwords, bookmarks and history will be there. Can't live without that feature.

 

[Timur Born]

Ryzen Mastery just applies a dumb -20C to Tctl and "Sense Skew" on Asus boards practically applies another -20C when temps are low. So RM most of the time is not correct in its temperature readings (full idle with Sense Skew disabled is when it is correct).

Since AMD unfortunately failed to provide a proper explanation of how temperature readings work I had to analyze the behavior of CPU temp (Tctl) and the respective offsets (plural!) myself. Here is in short what I found out using a 1800X on an Asus Crosshair Hero VI board. The "Sense Skew" BIOS option of the Asus CH6 makes this even more confusing and I advice to turn it off and change your fan-curves accordingly.

The problem is that the offset is not fixed, but changes dynamically, seemingly based on the CPU instruction set that us currently used by software. There are three offsets: +0C (base), +10C and +20C.

Which offset is used does not seem to depend on current temperature or even CPU load in terms of percentage, but really on what CPU instructions are currently used. The software "Statuscore" can be used to demonstrate this, because it seems to use +10C inducing instruction sets when even (2, 4..) cores are stressed, but +20C inducing instructions when odd cores (1, 3..) are stressed.

Most of the time offsets jump up in one go, but decrease only gradually. If the current temp reading is in between base and an offset then it might increase gradually instead of jumping.

At full idle no offset (+0C) is applied and your readings with "Sense Skew" enabled might show something weird, like being 10-20C below your room's ambient temperature. You will get a +0C offset mostly only while using the "Power Safer" and maybe "Balanced" plan.

Often you might see your temps jump up by +10C regularly, just to have it drop down gradually and then jump up by +10C again. This happens because some software/driver/service in the background induced the +10C offset shortly and then released it again.

Most of the common stress tests induce the +20C offset, but "Heavyload" is a good example of a stress test that only induces the +10C offset, even though it draw the very same current/power as some other stress tests that induce the +20C offset.

To make things more complicated, once Tctl hits 95C any offset is gradually decreased to match 95C, at the same time soft throttling (down to about x30) increases gradually. This continues until the offset hits +0C and the real CPU temp hits 95C. At this point hard throttling to x0.5 (0.5 GHz!) happens and Tctl is allowed to increase over 95C. At 115C Tctl a thermal shutdown happens.

One word of warning: If your motherboard uses "Sense Skew" by default then it artificially lower Tctl to a point where your CPU might hardly reach its soft throttling points, never reaches its hard throttling point and especially never does a thermal shutdown. Instead your CPU will likely crash (Code 8 on the CH6) and still get some voltage applied (1 V on the CH6).

 

[Crumpet]

Well, just had my first attempt at streaming with the Ryzen rig, playing the new Yooka Laylee game.

On previous pc, i5 6600 (same RAM and GPU) streaming meant stuttery gameplay, low FPS and frankly a shit stream. All pixelated.

First experience with Ryzen.. 130fps and a stream that looks almost exactly like i'm seeing it.

Can. Not. Believe. The. Difference.

Need to test with some of my fast paced racing stuff, but colour me impressed so far.

 

[Crono]

I believe they can all be imported. And with your Google account linked into Chrome, you can login to any Chrome anywhere (including your phone), and all your passwords, bookmarks and history will be there. Can't live without that feature.

Firefox Sync does the same.

I use Firefox (Waterfox on desktop), Chrome, Chromium, and Edge to varying degrees. Chrome still has odd behavior from time to time on certain websites (like on these forums, as reported by different users). Plus there may or may not be privacy concerns with Chrome for those who care*.

* If you don't, that's fine... not trying to start a privacy debate here, plus nearly everything does data collection to some degree

 

[guachi]

I think sponsoring streamers in some way or another would be a great way to advertise the multi-core power of Ryzen. Because the value of that doesn't just apply to streaming. It basically applies to any circumstance where you game + do something else.

I'd love to see more benchmarks where someone could set up a test that mimics something someone might actually do while gaming and benchmark that. For me, I have some encoding programs (not Handbrake) that take up about 15% of the CPU on my R7 and now, unlike before, I can do that encoding and still game. Two encodes is only about 1/3 of my CPU power and that leaves about 11 threads to do other stuff.

And since games seem to have a hard time using piles of cores it's like having an entire other computer available to do stuff! It's pretty cool. Oh, sure, I could have had this with an Intel chip. But that would have cost hundreds more.

 

[hotstocks]

I am like you and stuck on Firefox. I have years of ordered bookmarks and it is the only browser that properly displays bookmarks in the left column for easy clicking of hundreds of bookmarks. I've tried every browser even with add ons, but they just don't do the left bookmark column correct and persistant. That said, I have tried both 32 and 64 bit firefox and can't stand that there is a memory leak they won't fix. I have to close and restart firefox about 3x a day when it starts to used over 4gb ram and slows to a halt. Doesn't matter if I have 16 or 32gb ram either. I am sure it is due to me having about 25-100 tabs open, but I think it is Twitter and Facebook feeds that use java and Firefox just is too stupid to flush its memory once over 1 or 2 gb.

 

[Agent-47]

just finished the 1600 build.

MSI B350 Mortar, Bios rev 1.2 10th April (i donot think its the new microcode)
G Skill 3000C14 running at 2933@14-14-14-14-34, 1.35v. AIDA latency of 76 ns.
SoC@1.056v, as assigned by A-XMP
CPU@3.7Ghz@1.27v 2hr stress-test stable (1.25v failed after 30 minutes). Max temp 58 C. not pushing any higher till easter weekend as I have no idea what my VRM temps are. HWMonitor does not load, probably because of the RAID issue i mentioned (see below)

OC went smooth, A-XMP worked first try. did try to boot at 3200C16, did not stay stable. will try more SoC volt later.

Only annoying issue: could not get the bloody RAID working at all. the RAID prompt comes up asking to press Ctrl+R to enter setup, but it does not do anything other than prolong the boot time by 5 sec. I think its a BIOS issue. will try downgrading BIOS over the weekend.

Minor issue: BIOS post takes a good 5 seconds. Not as bad as some people with 25 sec post delay!

EDIT: anyone know where the HPET setting is located in MSI BIOS? Can not seem to find it.

 

[JimmiG]

Ryzen Mastery just applies a dumb -20C to Tctl and "Sense Skew" on Asus boards practically applies another -20C when temps are low. So RM most of the time is not correct in its temperature readings (full idle with Sense Skew disabled is when it is correct).

Since AMD unfortunately failed to provide a proper explanation of how temperature readings work I had to analyze the behavior of CPU temp (Tctl) and the respective offsets (plural!) myself. Here is in short what I found out using a 1800X on an Asus Crosshair Hero VI board. The "Sense Skew" BIOS option of the Asus CH6 makes this even more confusing and I advice to turn it off and change your fan-curves accordingly.

The problem is that the offset is not fixed, but changes dynamically, seemingly based on the CPU instruction set that us currently used by software. There are three offsets: +0C (base), +10C and +20C.

Which offset is used does not seem to depend on current temperature or even CPU load in terms of percentage, but really on what CPU instructions are currently used. The software "Statuscore" can be used to demonstrate this, because it seems to use +10C inducing instruction sets when even (2, 4..) cores are stressed, but +20C inducing instructions when odd cores (1, 3..) are stressed.

Most of the time offsets jump up in one go, but decrease only gradually. If the current temp reading is in between base and an offset then it might increase gradually instead of jumping.

At full idle no offset (+0C) is applied and your readings with "Sense Skew" enabled might show something weird, like being 10-20C below your room's ambient temperature. You will get a +0C offset mostly only while using the "Power Safer" and maybe "Balanced" plan.

Often you might see your temps jump up by +10C regularly, just to have it drop down gradually and then jump up by +10C again. This happens because some software/driver/service in the background induced the +10C offset shortly and then released it again.

Most of the common stress tests induce the +20C offset, but "Heavyload" is a good example of a stress test that only induces the +10C offset, even though it draw the very same current/power as some other stress tests that induce the +20C offset.

To make things more complicated, once Tctl hits 95C any offset is gradually decreased to match 95C, at the same time soft throttling (down to about x30) increases gradually. This continues until the offset hits +0C and the real CPU temp hits 95C. At this point hard throttling to x0.5 (0.5 GHz!) happens and Tctl is allowed to increase over 95C. At 115C Tctl a thermal shutdown happens.

One word of warning: If your motherboard uses "Sense Skew" by default then it artificially lower Tctl to a point where your CPU might hardly reach its soft throttling points, never reaches its hard throttling point and especially never does a thermal shutdown. Instead your CPU will likely crash (Code 8 on the CH6) and still get some voltage applied (1 V on the CH6).

Wow, I didn't know any of that, but it explains a lot. I've seen a few reports of 15C idle temps in a 25C room etc..
This is the first time I've heard of the variable skew, but it makes sense if AMD were trying to prevent CPUs from overheating with XFR enabled by kicking the fans into high gear at high CPU loads while at the same time keeping the RPM low at idle.

Where AMD messed up was when they didn't tell anyone about this, including motherboard manufacturers. Clearly Asus had no idea about any of this. They suddenly introduced a fixed Sense Skew with the 515 BIOS. With earlier BIOS versions, CPU temp running Prime95 was about 75C for me, now it's about 62C. What's the real temp? Who knows. Probably somewhere between 50C and 85C.

 

[Dresdenboy]

For those that don't speak german, can you summarize ?

krumme already mentioned the 6% min fps advantage. I'll add more points:
DDR4-2667 Single-Rank vs. DDR4-2667 Dual-Rank (improvement for the latter):
Games: +6.2 % min fps, +5.3 % avg. fps
Applications: +2.3% (7-zip: +12.2% Lightroom +3.7% x264 +2.4% were the biggest gainers)

DDR4-2667 Single-Rank vs. DDR4-3200 Single-Rank (improvement for the latter):
Games: +3.7 % min fps, +3.5 % avg. fps
Applications: +1.2%

I hope that bodes well for my 2x16GB DDR4-3200 CL15 TridentZ kit - DDR4-3200 Dual-Rank: Full of win! (even at more likely 2933)

 

[imported_jjj]

Can we see that in aida64 in bandwith?

On page 120-121 in this thread there was a bit of a conversation on the single ran vs dual rank perf topic.

PCGamesHardware has lots of results with single rank vs. dual rank memory:
http://www.pcgameshardware.de/Ryzen-5-1600X-CPU-265842/Tests/R5-1500X-Review-Mainstream-1225280/3/

Thanks for this , I miss some of theses as I don't speak any german.
With both this review and HardwareFR, it's becoming pretty conclusive.
Would be interesting to look at equal clocks but more relaxed timings for dual rank as that's what we would see in practice.Shouldn't be too bad when you factor in the data fabric.

.

 

[Justinbaileyman]

Well I dont know what they all did with the F6 Bios for the Gigabyte gaming 3 motherboard but it fixed all most all my issues. The cpu now turbo's up to 4.0Ghz correctly for single threaded apps and 3.7Ghz for multi thread. Memory now runs at rated speeds at the stock 1.2v instead of 1.35v.Heck I can even go up to 2666 with all 4 sticks if I wanted to. Cpu temps dropped down from 52c idle to 32c and from 68c-72c under load to 52c-60c.Only issue I am having now is with windows 10 it self running a couple of my programs. AutoGK craps out on 90% of my media encodes with a out of bound memory error.Been using the same settings on autoGK for 12 years now so there is no reason it should be crashing. Same thing is happening with MakeMKV. I had to change things up and started using WinFF, Handbrake, and Adobe Premiere Pro. No problems with those programs.I even did the Windows Update today but it didn't help.I guess one day at a time till they sort out the rest of the bugs.

 

[Raduque]

OK, so I think I have my component list down. I'm going for a wintery-type theme with this.

AMD Ryzen 1600
MSI B350 Tomahawk Artic
16gb (8x2) Geil Super Luce (white w/blue LED) ddr4-3000
Asus GTX1060 Dual 6gb
WD Blue m.2 500gb SSD
Seasonic M12II 750 full modular
Corsair 460x RGB

Around $1200, and should be perfect for 1080p gaming. Hopefully I can get around $800 for my current laptop, so I won't be in the hole the full $1200.

 

[unseenmorbidity]

Ryzen + amd gpu is a safer bet

 

[IEC]

Ryzen + amd gpu is a safer bet

He specifically wants a white-colored video card to match his theme. The Asus Dual series for Radeon are pretty crappy/poor QA so in this instance, the 1060 is the safer bet.

If he didn't care about color it'd be very difficult to beat a deal on a RX470 or RX480, however.

 

[Crono]

He specifically wants a white-colored video card to match his theme. The Asus Dual series for Radeon are pretty crappy/poor QA so in this instance, the 1060 is the safer bet.

If he didn't care about color it'd be very difficult to beat a deal on a RX470 or RX480, however.

Poor QA hasnt translated into bad reviews, though, for ASUS' RX480. I'd risk it.
https://www.newegg.com/Product/Product.aspx?Item=N82E16814126135

 

[kn51]

Well I dont know what they all did with the F6 Bios for the Gigabyte gaming 3 motherboard but it fixed all most all my issues.

Yeah, I was somewhat amazed...they knocked this one out of the park. We were all waiting for more frequent releases and it looked liked they were dragging their heels. Had the beta bios for an AB350M Gaming dated 3/17. But the official F4 for my board works like a champ. Told it to use XMP and RAM running at 3200 without a single reboot or hiccup.

Granted, haven't really beat on it yet...but impressed.

 

[IEC]

Yeah, I was somewhat amazed...they knocked this one out of the park. We were all waiting for more frequent releases and it looked liked they were dragging their heels. Had the beta bios for an AB350M Gaming dated 3/17. But the official F4 for my board works like a champ. Told it to use XMP and RAM running at 3200 without a single reboot or hiccup.

Granted, haven't really beat on it yet...but impressed.

It's because those of us running the betas and beating on the platform with a hammer have been letting mobo makers know what things work/don't work/need fixin'

 

[Rifter]

Well, the new 604 bios for the Asus x370-Pro is not working out so well for me.

I had memory stable at 3000 before with 515 bios but now i cant get it to not boot loop with even 2400mhz now, will only boot windows at 2133. Even CL18 timings and 1.35v with SOC@1.1V it will not boot windows @2400. FML.

Going to flash back to 515 now.

 

[guachi]

OK, so I think I have my component list down. I'm going for a wintery-type theme with this.

AMD Ryzen 1600
MSI B350 Tomahawk Artic
16gb (8x2) Geil Super Luce (white w/blue LED) ddr4-3000
Asus GTX1060 Dual 6gb
WD Blue m.2 500gb SSD
Seasonic M12II 750 full modular
Corsair 460x RGB

Around $1200, and should be perfect for 1080p gaming. Hopefully I can get around $800 for my current laptop, so I won't be in the hole the full $1200.

I'll second or third the recommendation for a 480. It seems like the safer bet and with Freesync and a Ryzen chip you'll have the smoothest gaming experience at 1080 that you could ever have.

 

[IEC]

Poor QA hasnt translated into bad reviews, though, for ASUS' RX480. I'd risk it.
https://www.newegg.com/Product/Product.aspx?Item=N82E16814126135

Either that or if he's handy he could plastidip a better card white...

 

[Markfw]

Either that or if he's handy he could plastidip a better card white...

spray paint works good too.

 

[Rifter]

Either that or if he's handy he could plastidip a better card white...

or paint, the shrouds are easy to remove.