Official AMD Ryzen Benchmarks, Reviews, Prices, and Discussion

[unseenmorbidity]

ht4u.net finally posted their Ryzen review, it is packed with data. They seem impressed
They included 1800X @ stock+ DDR4-3200 memory, along with 2400 and 2666Mhz results. Pretty massive review, took some time to read it through.
2666+MHZ memory really does make a difference in games with Ryzen, in some titles there was 15+% difference in 1080p (resolution they used for their tests).

https://www.ht4u.net/reviews/2017/amd_ryzen_7_1800x_im_test/index37.php

I am wondering if the gains from faster memory will be greatly reduced when the bugs are fixed.

 

[IEC]

ht4u.net finally posted their Ryzen review, it is packed with data. They seem impressed
They included 1800X @ stock+ DDR4-3200 memory, along with 2400 and 2666Mhz results. Pretty massive review, took some time to read it through.
2666+MHZ memory really does make a difference in games with Ryzen, in some titles there was 15+% difference in 1080p (resolution they used for their tests).

https://www.ht4u.net/reviews/2017/amd_ryzen_7_1800x_im_test/index37.php

This is what's holding me back from higher performance more than the clockspeed wall I am hitting past 4.15 GHz (at safe voltages).

I cannot get a divider above the 2666 to work on the C6H with the 0902 BIOS (and it's forcing me to 2T command rate). I can BCLK OC to get DDR4-3200 but I'd prefer to set all-core turbo at 41x, BLCK 100, and use the 3200 divider.

 

[looncraz]

Would be enough of a diff in clocks to see if there is an impact.

Results are somewhat finicky. At first, I thought you might be onto something... now I see that everything is just in the same general window of performance. Enabling/disabling HPET timer in windows (bcdedit /set useplatformclock true/false) seems to have a larger impact.

On an interesting side note, I'm having pretty high DPC latency - but it's the standard Microsoft driver culprits (USB, HDAUDIO, and SATA). Going to try manually forcing to use AMD drivers (the ones a fscking slipstreamed, Microsoft!). I know they were in use originally, but I guess updating to Build 1607 overrode those. Fortunately, most of my storage testing was done before hand (and it's all REALLY good).

In fact, aside from some weird scheduling issues that only occur with SMT enabled on Windows 10, Ryzen is behaving wonderfully... as long as you know that it has a crazy many-stage error correction protocol built in...

Power on -> off -> on -> off -> on -> off -> on -> off -> on -> warm reboot -> pause -> warm reboot -> pause -> warm reboot -> reset at BIOS screen -> normal boot

Kind of disconcerting when that behavior, in the past, was indicative of a stability setting you had to manually correct. The system will learn the DDR itself and try various configurations, and it will reset platform errors after only so many power cycles (hence all the power cycling). I have yet to do much testing on the exposed bits related to this logic.

For now, I'm reinstalling AMD drivers onto Windows 10 and validating a few results don't change, then doing BF1 testing all over again... VSync on is not so good for benchmarking

 

[imported_jjj]

This is what's holding me back from higher performance more than the clockspeed wall I am hitting past 4.15 GHz (at safe voltages).

I cannot get a divider above the 2666 to work on the C6H with the 0902 BIOS (and it's forcing me to 2T command rate). I can BCLK OC to get DDR4-3200 but I'd prefer to set all-core turbo at 41x, BLCK 100, and use the 3200 divider.

Was it working with an earlier BIOS?
Just in case, you know that you need an even number for CL above 2666? Just in case that's what's holding you back.
SoC voltages adjusted up could help but take it easy with those.

 

[Puffnstuff]

Don't jump into a rage on me fellas but have any of you noticed the writing on the lids differing? All say diffused in the USA but some say made in China and others say made in Malaysia. Is this difference having an effect on differing performance between the chips?

 

[imported_jjj]

Results are somewhat finicky. At first, I thought you might be onto something... now I see that everything is just in the same general window of performance. Enabling/disabling HPET timer in windows (bcdedit /set useplatformclock true/false) seems to have a larger impact.

On an interesting side note, I'm having pretty high DPC latency - but it's the standard Microsoft driver culprits (USB, HDAUDIO, and SATA). Going to try manually forcing to use AMD drivers (the ones a fscking slipstreamed, Microsoft!). I know they were in use originally, but I guess updating to Build 1607 overrode those. Fortunately, most of my storage testing was done before hand (and it's all REALLY good).

In fact, aside from some weird scheduling issues that only occur with SMT enabled on Windows 10, Ryzen is behaving wonderfully... as long as you know that it has a crazy many-stage error correction protocol built in...

Power on -> off -> on -> off -> on -> off -> on -> off -> on -> warm reboot -> pause -> warm reboot -> pause -> warm reboot -> reset at BIOS screen -> normal boot

Kind of disconcerting when that behavior, in the past, was indicative of a stability setting you had to manually correct. The system will learn the DDR itself and try various configurations, and it will reset platform errors after only so many power cycles (hence all the power cycling). I have yet to do much testing on the exposed bits related to this logic.

For now, I'm reinstalling AMD drivers onto Windows 10 and validating a few results don't change, then doing BF1 testing all over again... VSync on is not so good for benchmarking

You are having all the fun!
Have you tried OC with fewer cores yet?

 

[imported_jjj]

Do jump into a rage on me fellas but have any of you noticed the writing on the lids differing? All say diffused in the USA but some say made in China and others say made in Malaysia. Is this difference having an effect on differing performance between the chips?

The difference should be where the packaging is done, all should be made by GloFo in the US so no material differences.

 

[looncraz]

You are having all the fun!
Have you tried OC with fewer cores yet?

I only set 3.8Ghz on all eight cores. I don't have any airflow on this thing other than the CPU fan (Cooler Master TX2), so the VRMs are already warm even at 3.8Ghz.

Doing 6C/12T testing now. This will take an hour.

 

[imported_jjj]

I only set 3.8Ghz on all eight cores. I don't have any airflow on this thing other than the CPU fan (Cooler Master TX2), so the VRMs are already warm even at 3.8Ghz.

Doing 6C/12T testing now. This will take an hour.

Memory scaling could be interesting with 4C but you can't push memory. With 4C it would be easier to see where the sweet spot is for BW/core.

EDIT: New thought, does disabling SMT improve memory BW and/or latency? Only latency actually as BW is very close to theoretical.

 

[looncraz]

Memory scaling could be interesting with 4C but you can't push memory. With 4C it would be easier to see where the sweet spot is for BW/core.

EDIT: New thought, does disabling SMT improve memory BW and/or latency? Only latency actually as BW is very close to theoretical.

Disabling SMT had no impact on bandwidth or latency (my very first thought - and test - this morning ).

Anyway, I'm convinced my gaming benchmarks on Windows 10 are worthless. I get far too much variation. I'm going to try to get Windows 7 on it, it will take a few hours to get it setup for testing, so I'm done for now. Snack time, watch some Big Bang Theory, cook up some onions and garlic for my 15-bean soup, watch Star Trek TNG, eat some 15-bean soup, then store the rest of my 15-bean soup and clean up my mess, then back to Ryzen.

 

[IEC]

Just an FYI the power delivery on the ASUS ROG Crosshair VI Hero is seriously overbuilt and they use the pricier TI combined high side/low side chip. VRMs and current delivery are definitely NOT a hindrance with the C6H.

The voltage curve past 4.1GHz (or less, depending on your sample) is what prevents the higher OCs at the moment. Perhaps updates will yield slightly better OCs in time but I think 4.1GHz+ is a very good OC for the time being.

 

[unseenmorbidity]

Disabling SMT had no impact on bandwidth or latency (my very first thought - and test - this morning ).

Anyway, I'm convinced my gaming benchmarks on Windows 10 are worthless. I get far too much variation. I'm going to try to get Windows 7 on it, it will take a few hours to get it setup for testing, so I'm done for now. Snack time, watch some Big Bang Theory, cook up some onions and garlic for my 15-bean soup, watch Star Trek TNG, eat some 15-bean soup, then store the rest of my 15-bean soup and clean up my mess, then back to Ryzen.

That might be interesting. Recording the same run in Windows 10 on Ryzen three times, and then syncing them together in the same video to see the variance in performance.

 

[looncraz]

That might be interesting. Recording the same run in Windows 10 on Ryzen three times, and then syncing them together in the same video to see the variance in performance.

Dang it, why didn't I think about that?

I guess I have an excuse to try out ReLive!

Just don't laugh at my BF4 skills...

 

[Shamrock]

ht4u.net finally posted their Ryzen review, it is packed with data. They seem impressed
They included 1800X @ stock+ DDR4-3200 memory, along with 2400 and 2666Mhz results. Pretty massive review, took some time to read it through.
2666+MHZ memory really does make a difference in games with Ryzen, in some titles there was 15+% difference in 1080p (resolution they used for their tests).

https://www.ht4u.net/reviews/2017/amd_ryzen_7_1800x_im_test/index37.php

Did anyone else notice he had a Ryzen 5 1600x ?

 

[Mockingbird]

One of the thing that is so dumbfounded is that AMD went with PGA again instead of LGA with AM4.

AMD has a chance to start over with AM4 and instead chose PGA again.

 

[piesquared]

Did anyone else notice he had a Ryzen 5 1600x ?

Wow that is a really good showing for both the 8 and 6 core Ryzen. He might have disabled 2 cores to get the R5.

 

[imported_jjj]

Dang it, why didn't I think about that?

I guess I have an excuse to try out ReLive!

Just don't laugh at my BF4 skills...

Doing resolution scaling in gaming 8 vs 4 cores? Reviewers seem to need a push to understand what they are doing. Aside from Zen suffering in high FPS testing due to mem latency and maybe more.
Since you don't have a GTX 1080 or above it would be best to start as low as possible , well below 1080p.

 

[looncraz]

One of the thing that is so dumbfounded is that AMD went with PGA again instead of LGA with AM4.

AMD has a chance to start over with AM4 and instead chose PGA again.

I **LOVE** PGA. LGA is so problematic when you switch CPUs!

Did you know that the motherboard LGA pins (I mean... springs) can get weak with thermal cycling and start to lose connectivity? Or that installing a different CPU might not always work because the indentions on the pads cause a weak connection? Or that there's pretty much no worthwhile means of fixing a broken pin?

With PGA, the only issue is bent or broken pins. Both issues which can usually be resolved with relative ease.

Ryzen pins are crazy small, though... I'd hate to have to try and fix one. But if a pin were to break, you just put it back in its place in the socket and install the CPU again

I've never seen a CPU become unusable because of its pins (I know it happens, just never to me in >20 years). I've seen at least a half dozen LGA motherboards destroyed because of someone who was careless and two boards which became useless because the pins lost their springiness.

 

[looncraz]

Doing resolution scaling in gaming 8 vs 4 cores? Reviewers seem to need a push to understand what they are doing. Aside from Zen suffering in high FPS testing due to mem latency and maybe more.
Since you don't have a GTX 1080 or above it would be best to start as low as possible , well below 1080p.

I am personally only interested in 1080p results, with maxed graphical settings. My trick is to change CPU frequency to find the bottlenecks. I'm testing at 3Ghz, stock, and 3.8Ghz. I would test at 4Ghz (heck, I can get 4.05Ghz stable - 4.1 even works for a while... without touching voltage), but I'm uncomfortable with the fact that every single piece of software that monitors voltage is giving me a different answer... and the BIOS gives me another answer once more...

Unless you believe me when I say I was dead-stable with 0.7V @ 4Ghz...

To be fair, my CPU would barely even get warm during testing... and I was just using a TX-2 cooler. The fan never sped up once during all my testing, temps always below ~50C. VRMs only mildly warm. No matter what, I expect this chip to do well on the Asus C6H and my water-cooling setup I'm going for a 24/7 4Ghz overclock.

 

[imported_jjj]

I am personally only interested in 1080p results, with maxed graphical settings. My trick is to change CPU frequency to find the bottlenecks. I'm testing at 3Ghz, stock, and 3.8Ghz. I would test at 4Ghz (heck, I can get 4.05Ghz stable - 4.1 even works for a while... without touching voltage), but I'm uncomfortable with the fact that every single piece of software that monitors voltage is giving me a different answer... and the BIOS gives me another answer once more...

Unless you believe me when I say I was dead-stable with 0.7V @ 4Ghz...

To be fair, my CPU would barely even get warm during testing... and I was just using a TX-2 cooler. The fan never sped up once during all my testing, temps always below ~50C. VRMs only mildly warm. No matter what, I expect this chip to do well on the Asus C6H and my water-cooling setup I'm going for a 24/7 4Ghz overclock.

Reviewers need to understand the difference in scaling with res (or FPS) between 4 cores and 8 cores SKUs and nobody seems to be looking at that.Granted, with 2xCCX Ryzen likely scales less than it should, at least for now.
Sounds like you have a good chip if you can get 4GHz at stock voltages and that's on a B350 mobo.Unless the mobo's default V are set high?

 

[inf64]

Did anyone else notice he had a Ryzen 5 1600x ?

The author emulated Ryzen 5 via motherboard BIOS downcore option . There is a disclaimer in test setup section where he states that results may be different with retail Ryzen 5 SKUs due to possible cash configuration being different.

 

[imported_jjj]

The author emulated Ryzen 5 via motherboard BIOS downcore option . There is a disclaimer in test setup section where he states that results may be different with retail Ryzen 5 SKUs due to possible cash configuration being different.

Google Translate suggests that the RAM was at 2400MHz no idea what CL for the hexa, maybe you can help clarify in case you speak German?

 

[inf64]

Google Translate suggests that the RAM was at 2400MHz no idea what CL for the hexa, maybe you can help clarify in case you speak German?

Unfortunately I don't speak German. From what I understand the default 1800X scores are done with DDR4 2400Mhz, the rest are noted in the brackets (DDR4 2666-3200). OCed score was not included in the charts. Ryzen 5 1600X emulated scores were also achieved with DDR4 2400Mhz.

 

[CatMerc]

I am personally only interested in 1080p results, with maxed graphical settings. My trick is to change CPU frequency to find the bottlenecks. I'm testing at 3Ghz, stock, and 3.8Ghz. I would test at 4Ghz (heck, I can get 4.05Ghz stable - 4.1 even works for a while... without touching voltage), but I'm uncomfortable with the fact that every single piece of software that monitors voltage is giving me a different answer... and the BIOS gives me another answer once more...

Unless you believe me when I say I was dead-stable with 0.7V @ 4Ghz...

To be fair, my CPU would barely even get warm during testing... and I was just using a TX-2 cooler. The fan never sped up once during all my testing, temps always below ~50C. VRMs only mildly warm. No matter what, I expect this chip to do well on the Asus C6H and my water-cooling setup I'm going for a 24/7 4Ghz overclock.

inb4 you won the silicon lottery and your sample IS stable at 0.7V @ 4GHz.

 

[inf64]

A bit tongue in cheek video, still very amusing

In this video, I benchmarked Quake 2 running at 320x240 on a Ryzen CPU and a Kabylake CPU. As can you see, AMD's new Ryzen processor is a terrible choice for playing games that are 20 years old.