Ryzen - Return of the Jedi

[chew*]

Here is an answer to one of the big questions. PCI E gen 3 vs gen 1.

GEN 3 what is most important is game tests.


GEN 1


As you can see yes there is a slight performance drop in gen 1........you can also see that cpu performance scales with ref clock and lower mem dividers.

This is extremely consistent. I see fps drop but cpu performance gains repeatedly test after test.

Also as always the TAICHI drops its wifi/bt device whenever I run up ref clock like this. I suggest disabling it as leaving it on could damage it.

[chew*]

Quick test with flare X and latest agesa on Taichi moving on to a board known to have issues with stability @ 3200 to see if it can actually run 3200 now....

32m looks slower on latest agesa code.........

[techtrancer]

Think this is the newest One

29.1 build 15

http://www.mersenneforum.org/showthread.php?t=22141

Also Nice ist HWinfo64
It has an Error Couter
Shows errors while Prime is Running

[chew*]

I only installed it to show people who rely on software for voltage readings how wrong they are compared to socket readings and a dmm...

Like for example...my psu is so good it supplies 20+v to the 12v rail...

[techtrancer]

Sometimes i can see errors bevore a prime thread breake down
Save a Lot of Time

[chew*]

Yah i have looked at it. Ive seen no errors thus far when doing my 24hr runs. I also am not running ragged edge stabilty and have remained conservative voltage and speed wise for any of my "stability" testing that i am reporting.

Still i don't trust it...its way off actual voltages. Software is only as accurate as the calibration...and even if calibrated to the sensors right the chips are also calibrated...if they are not calibrated via actual voltage at socket...its pointless.

[AMD_OCer]

Ok a couple updates.

Someone said I should update prime 95.

I am on and have been on version 29.1 for awhile now........if there is newer feel free to link me.....

4x8gb on x370 taichi = fail 1.94a and 2.0 bios......

By fail I mean if they can not run 3200 14-14-14-34. Could be due to the fact that I am mismatched with Flare X and Trident-Z but I do not believe that to be the case....

Both have same specs although I have not looked into the SPD table in mem tweak it yet to look for differences.

2x8g pc3200 14-14-14-34 Flare X works great on taichi. Load XMP and go....that simple.

I will be reviewing it soon on multiple boards as it seems to matter more board then cpu in my testing......

did you try 3x8gb to see if it boots?

I read somewhere on newegg a reviewer having the same issue. G.Skill replied to him that he should buy the 4x8GB kit, but that kit runs much lower 2400 Mhz.

[chew*]

3 x 8 gb? the system is not designed to run dual and single channel.

@ some point it is going to take a massive performance hit.

I tested 4x8gb on taichi. No good for 3200 operation regardless of how I tried to get there......ASUS has gotten there 4x8g but cant with 2x16gb. go figure......

[AMD_OCer]

3 x 8 gb? the system is not designed to run dual and single channel.

@ some point it is going to take a massive performance hit.

I tested 4x8gb on taichi. No good for 3200 operation regardless of how I tried to get there......ASUS has gotten there 4x8g but cant with 2x16gb. go figure......

Sounds like BIOS issue to me for the Taichi

On the ASUS, you actually got 3200 with 4x8gb @CL14...?

[chew*]

Infrared who knows his stuff like me managed it he even ran prime blend with a hefty chunk of ram allocated @14-14-14 3200 4x8g on ch6.

Here is his post.

http://www.xtremesystems.org/forums/...=1#post5257046

I'm going to try to duplicate it, Basically its done the same way I did it on the taichi.

[Formula350]

Ok a couple updates.

Someone said I should update prime 95.

I am on and have been on version 29.1 for awhile now........if there is newer feel free to link me.....

4x8gb on x370 taichi = fail 1.94a and 2.0 bios......

By fail I mean if they can not run 3200 14-14-14-34. Could be due to the fact that I am mismatched with Flare X and Trident-Z but I do not believe that to be the case....

Both have same specs although I have not looked into the SPD table in mem tweak it yet to look for differences.

2x8g pc3200 14-14-14-34 Flare X works great on taichi. Load XMP and go....that simple.

I will be reviewing it soon on multiple boards as it seems to matter more board then cpu in my testing......

I should've posted the comment I was going to in regards to your Taichi praise...

Numerous people on [H] who have Taichis and Killer boards and have updated to 2.0 within the last week, all are complaining about problems with overclocks, not being able to hit 3200 or even >2667 anymore, etc.
My suspicion is that ASRock is either using the same faulty Microcode that MSI had put out in the Titanium 1.41 BIOS and caused them to pull ALL of their Beta's offline for every board... OR that it's 1.0.0.4a which even ASUS is using, and I believe I've seen a case or two with the C6H having woes after latest betas. Apparently, according to word on the MSI forum from people who flashed the latest Tomahawk, is that MSI is using 1.0.0.4b of AGESA. So it'll be interesting to see what the results from that are.

I'm still running Titanium BIOS v1.10 that my board shipped with! Been so good to me I haven't felt a reason to update. Not until a new release based on 1.41 comes out, since that has BCLK!

In other news... Thermaltake is being naughty with their AM4 brackets and listing support for Water AIO models. They claim they ONLY work on Water3.0 products. I called their bluff a couple weeks back and just now was able to confirm that Water2.0 WORKS FINE with that bracket. So much so, that I used the Corsair bracket on my Tt Water2.0 Pro! lol

[chew*]

it ends in 111 the prime pro is on the same microcode.

The main problem is the shotgun approach does not work on ryzen regardless of motherboard and if it does or XMP loads its not stable in some instances.

The other issue is AMD seems to have slacked off timings in an effort to appease people however its doing absolutely nothing but hurting performance. ( compare my screenshots and you will see TRC 63 1.94a, TRC 73 2.0 and that's what we can see without even digging )

Funny thing in my youtube video some guy claimed latency was better in latest......umm survey says XXXXXXXXXXXXX wrong.

The same baby step no shutgun approach applies and now performance has gone down.....

The biggest kicker is the new win 10 update that everyone thinks is the shiznit with gaming mode.......just makes things even worse.

Latest agesas are losing 6-12 secs in 32m pi.......new win 10 build 1703 ( gaming mode ) losing another 30-40 seconds in 32m pi......no love lost really RTC makes 2d benching win 10 pointless anyway.......

Now i'm not a software expert but if we were to run a slew of tests i'm sure we could find something real world that it is impacting........just like PI 32m.

Oh heres a really good kicker......reviewers actually used this new build 1703 on R5 reviews.......I guess reviewers job is just to assume performance is on par nowadays and not actually verify it......

I find it pretty funny that some average guy like me can spot this immediately after updating......but a guy that does this for a living for a review site and gets paid to do it can't.

[techtrancer]

The other issue is AMD seems to have slacked off timings in an effort to appease people however its doing absolutely nothing but hurting performance. ( compare my screenshots and you will see TRC 63 1.94a, TRC 73 2.0 and that's what we can see without even digging )

Does the higher tRC change anything with Voltage use?
Because the FX & PII Need less v-Core with higher tRC

[thigobr]

Is there any easy measure point for vcore? I am using an Asus Prime X370-PRO

About the new AGESA, at least using AIDA64, memory latency went down (about 5ns) but as so the bandwidth (~500MB/s)... Unfortunately I did not ran other tests using the old BIOS.

[AMD_OCer]

Infrared who knows his stuff like me managed it he even ran prime blend with a hefty chunk of ram allocated @14-14-14 3200 4x8g on ch6.

Here is his post.

http://www.xtremesystems.org/forums/...=1#post5257046

I'm going to try to duplicate it, Basically its done the same way I did it on the taichi.

Your memory is 3200 @CL14 whereas Infrared used the 3733@CL17, meaning am kind of not surprised he was able to get 3200@14 (down clocking his timings and frequency)

Am really curious as to what timing were you able to get 3200 for 4x8gb?

[chew*]

Is there any easy measure point for vcore? I am using an Asus Prime X370-PRO

About the new AGESA, at least using AIDA64, memory latency went down (about 5ns) but as so the bandwidth (~500MB/s)... Unfortunately I did not ran other tests using the old BIOS.

http://www.xtremesystems.org/forums/...=1#post5256260

Or you can follow the you tube video on advice on how to set voltages so set = real.

[chew*]

Your memory is 3200 @CL14 whereas Infrared used the 3733@CL17, meaning am kind of not surprised he was able to get 3200@14 (down clocking his timings and frequency)

Am really curious as to what timing were you able to get 3200 for 4x8gb?

If i cant do 14-14-14 i probably wont bother.

Right now im back on the prime pro trying to figure out what its deal is.

Memory, cpu are all proven yet board can not pass prime blends over 2933 even slacked timings..

[AMD_OCer]

If i cant do 14-14-14 i probably wont bother.

Right now im back on the prime pro trying to figure out what its deal is.

Memory, cpu are all proven yet board can not pass prime blends over 2933 even slacked timings..

I would be happy with a 4x8gb at 2933 and 14 timing! If you get there, please post some pics.

For the gaming crowd, 16gb is proven rock solid 3200@CL14, but for those who need 32GB, we need more tests....

[chew*]

I would be happy with a 4x8gb at 2933 and 14 timing! If you get there, please post some pics.

For the gaming crowd, 16gb is proven rock solid 3200@CL14, but for those who need 32GB, we need more tests....

I should be able to do that with my eyes closed. I will update as soon as I get to that.

Right now i'm trying to figure out what makes this prime pro tick as its proving to be a very difficult board to stabilize @ 3200. I'm going to figure it out eventually....

[chew*]

Does the higher tRC change anything with Voltage use?
Because the FX & PII Need less v-Core with higher tRC

These chips are not like BD or Deneb...old tricks do not apply.

If your lucky and you kill smt on an 8 core you can gain 25mhz with less voltage.....that is not a significant gain for the loss......

There is no magic that is going to make them clock up higher...

what they are doing is for compatibility only and in some boards that are picky its still not stable @ 3200......and it lost a lot of performance....pointless when it is all loss no gain for 24/7 users......

[Formula350]

I find it pretty funny that some average guy like me can spot this immediately after updating......but a guy that does this for a living for a review site and gets paid to do it can't.

I completely agree with you there! Reviewers also shouldn't be changing their procedure all the time either by introducing software variables, as it totally destroys their comparison database. That'd be like taking a 12V light bulb and feeding it 11V or 13V, where the change in voltage is like an update to Windows or their benchmarks... TOTALLY skews the results. Until you've first vetted the new software and compared it in a closed environment (basically, just on one system) to determine that moving over will cause no adverse effects, which being the "Creators Update" is a very BIG thing and thus is definitely going to have adverse effects, it means you can't just jump on it to be "trendy" :\ You need to then go back and re-test every other system in your comparison database in order for your results to have an ounce of validity.

TL;DR - 100% correct, Chew. Crazy to think about.

Is there any easy measure point for vcore? I am using an Asus Prime X370-PRO

About the new AGESA, at least using AIDA64, memory latency went down (about 5ns) but as so the bandwidth (~500MB/s)... Unfortunately I did not ran other tests using the old BIOS.

That's a valid point. A member on [H] had posted their AIDA64 Cachemem results. His tests were at 3200 14-14-14, but compared to my own at 14-15-15 (haven't worked out what's needed to get my kit happy w/ all 14s, since it's a 15-15-15 kit), they're the same on all results EXCEPT for the Latency, which his was ~10ns lower than mine.

As I've said over and over, I'm running the shipping version so I have the absolute oldest AGESA, and his was in all likelihood 1.0.0.4a (or at worse, RC1). Which on the note of the AGESA versions, it's silly they don't just flat out say the version number in our BIOS, instead using that cryptic hex-or-whatever numbering that they do.


Still, chew, offer remains that I'll tinker with your Prime Pro's BIOS, I just need you to dump it through the Windows tool I had linked in that PM.

[thigobr]

http://www.xtremesystems.org/forums/...=1#post5256260

Or you can follow the you tube video on advice on how to set voltages so set = real.

Really good @chew*!!! Thanks a lot! So LLC5 does the trick for manual vcore definition. Did you play with offset on this board?

My system is working very stable with 2x8GB Trident Z CL14 Samsung B-die, using VSOC @1.0V. It was stable the same with the old stock BIOS 0511...

[chew*]

Nah offsets are a pain especially when I go ln2 trying to do math for target voltage.

Ok so IMC stability testing is a painstaking process so I decided to take a break....

I tossed in the crappiest of crappy bin chips because I am grass roots and love to tinker with the lower end stuff and make it fly......this is how overclocking was born.

I have no clue if this is good bad or average for a R5 1400 as I have not checked reviews or hwbot databases but I will tell you its fun to play with.

Kicked that 4770K @ 4.4g compare results azz into next week.......and that was with a stock RGB cooler......

This was like 15 minutes worth of play time.

[StAndrew]

Given the infinity fabric clocks 1:1 with memory, would you suggest biasing memory overclock to speed vice timing? What's the highest memory clock you can get generally, with timing set liberally?

Would be curious to see a comparison between overclock focused on tighter timings and overclock focused on high memory speeds.

[chew*]

Given the infinity fabric clocks 1:1 with memory, would you suggest biasing memory overclock to speed vice timing? What's the highest memory clock you can get generally, with timing set liberally?

Would be curious to see a comparison between overclock focused on tighter timings and overclock focused on high memory speeds.

I did it with dual rank 2 different ways. 17-16-16 3400 and 14-14-14 3200 it was +/- in various benchmarks. Pretty much whatever is easier to do on your hardware is the best choice.