FinalWire Unleashes AIDA64 v5.90，Optimized 64-bit benchmarks for AMD Ryzen “Summit Ridge” processors

[csbin]

https://www.aida64.com/downloads/ZjljMTEzZDU=

Release notes:

• AVX2 and FMA accelerated 64-bit benchmarks for AMD Ryzen Summit Ridge processors
• Microsoft Windows 10 Creators Update support
• Optimized 64-bit benchmarks for Intel Apollo Lake SoC
• Improved support for Intel Cannonlake, Coffee Lake, Denverton, Kaby Lake-X, Skylake-X CPUs
• Preliminary support for AMD Zen server processors
• Preliminary support for Intel Gemini Lake SoC and Knights Mill HPC CPU
• NZXT Kraken X52 sensor support
• Socket AM4 motherboards support
• Improved support for Intel B250, H270, Q270 and Z270 chipset based motherboards
• EastRising ER-OLEDM032 (SSD1322) OLED support
• SMBIOS 3.1.1 support
• Crucial M600, Crucial MX300, Intel Pro 5400s, SanDisk Plus, WD Blue SSD support
• Improved support for Samsung NVMe SSDs
• Advanced support for HighPoint RocketRAID 27xx RAID controllers
• GPU details for nVIDIA GeForce GTX 1080 Ti, Quadro GP100, Tesla P6

 

[Carfax83]

Well now that AIDA64 has specific enhancements for Ryzen, AMD can't say that the reason why there memory latency score is so terrible is due to the application itself being buggy or unoptimized. The terrible memory latency probably partially explains why Ryzen gets "killed" by the Core i7 in games which use a single thread for rendering like Witcher 3 and Far Cry Primal.

 

[IEC]

From reddit posters:

DDR4 @ 2933 CL16:

DDR4 @ 2962 CL14:
http://imgur.com/a/bU1wM

With any luck I will have results for DDR4 @ 3200+ CL14 by this weekend.

 

[Carfax83]

^^ That's not the Ryzen optimized version that was just released. Those numbers don't seem accurate at all. The Ryzen optimized version would show version 5.90 at the bottom, like this:

*Edit* I saw you post a 5.90 screenshot in the imgur link. Now the latency is looking a lot more reasonable!

 

[csbin]

1800X@4.05G+DDR4 3600

 

[PhonakV30]

1800X@4.05G+DDR4 3600

Whoa that result! compare to 6900K , only Memory and L1 cache.but if Ryzen goes quad channel , I predict Memory read will get at least 80K or above.

 

[Atari2600]

1800X@4.05G+DDR4 3600

Are you doing anything fancy with the memory and is it stable?

 

[unseenmorbidity]

Well now that AIDA64 has specific enhancements for Ryzen, AMD can't say that the reason why there memory latency score is so terrible is due to the application itself being buggy or unoptimized. The terrible memory latency probably partially explains why Ryzen gets "killed" by the Core i7 in games which use a single thread for rendering like Witcher 3 and Far Cry Primal.

I wish everything could always be fixed with a single update. Let's hope Aida64 did it.

 

[Atari2600]

Well now that AIDA64 has specific enhancements for Ryzen, AMD can't say that the reason why there memory latency score is so terrible is due to the application itself being buggy or unoptimized. The terrible memory latency probably partially explains why Ryzen gets "killed" by the Core i7 in games which use a single thread for rendering like Witcher 3 and Far Cry Primal.

I think you'd look at the new results and say "what terrible memory latency"...?

 

[unseenmorbidity]

These are my results at 2133 memory. Perhaps this might be relevant as a baseline.

1800X@4.05G+DDR4 3600

What changes did you make in the uefi for that Bclk OC?

 

[Carfax83]

I think you'd look at the new results and say "what terrible memory latency"...?

It is a big improvement, but still well below (or above I should say) what Intel is getting. This will be a bigger problem in servers, HPC and databases where extremely large datasets are more common.

 

[IEC]

Consumer Ryzen is revision ZP-B1. We know from ES leaks that the HEDT chips (and likely, server chips) are already revision B2. So we cannot extrapolate to server performance just yet.

 

[Atari2600]

This will be a bigger problem in servers, HPC and databases where extremely large datasets are more common.

Large datasets are more likely to be bandwidth starved than killed by latency.

 

[guskline]

Here are my results with a 1800x stock but memory upped to 2933

 

[Carfax83]

Large datasets are more likely to be bandwidth starved than killed by latency.

Perhaps. But the point is that Ryzen's ODMC is unbalanced in its performance profile versus that of Intel's. In fact, Ryzen's memory latency to me is reminiscent of the off die memory controllers from the P4 and Athlon XP days that had much higher latencies compared to the on die memory controllers that we're now familiar with.

 

[IEC]

DDR4-3200 on my $79.99 motherboard with R7 1700 @ 4GHz:

Both 2.1 and 2.2 BIOSes were literally one setting clicked on (A-XMP) and done.

 

[imported_jjj]

Perhaps. But the point is that Ryzen's ODMC is unbalanced in its performance profile versus that of Intel's. In fact, Ryzen's memory latency to me is reminiscent of the off die memory controllers from the P4 and Athlon XP days that had much higher latencies compared to the on die memory controllers that we're now familiar with.

You just don't stop with the nonsense. 4 posts (25% of the thread) saying the same thing.
Ryzen is 10-15ns behind Broadwell-E in memory latency and that's right now when the secondary timings are locked and likely not tuned. Before you try to argue with that, go check Broadwell-E memory scaling and don't just post the best result you can find.
That's not terrible or catastrophic just not great.
Ryzen does scale above 3200MHz unlike Intel so if they manage those speeds, that's a plus. Bandwidth is great though.
What is poor is memory OC support and remains to be seen how much better that gets.

 

[Carfax83]

Ryzen is 10-15ms behind Broadwell-E in memory latency and that's right now when the secondary timings are locked and likely not tuned. Before you try to argue with that, go check Broadwell-E memory scaling and don't just post the best result you can find.

Dude, I'm sure you can count so why bother with underestimating Broadwell-E's lead, which is over 20ns in this particular example if you compare my score against IECs, using the same RAM speed with the same primary timings. There are no timings you can tweak, primary or secondary that can make up such a large gap. And if you think that gap is inconsequential, then think again because an extra 21ns of memory latency will be noticeable.

When it comes to faster RAM, it's mostly the reduction in latency and not the increase in bandwidth that manifests as greater performance for consumer applications like games.

That's not terrible or catastrophic just not great.

Well to me it's terrible. Broadwell-E's quad channel memory controller typically has higher latency than the equivalent dual channel memory controller in Skylake and Kabylake (by about 10+ns), presumably due to greater complexity. So to see my 6900K's quad channel memory latency score so much better than Ryzen's which has a dual channel memory controller, is fairly alarming and demonstrates that it's highly inefficient.

Ryzen does scale above 3200MHz unlike Intel so if they manage those speed, that's a plus. Bandwidth is great though.

Who said that Intel does not scale beyond 3200Mhz?

 

[imported_jjj]

Dude, I'm sure you can count so why bother with underestimating Broadwell-E's lead, which is over 20ns in this particular example if you compare my score against IECs, using the same RAM speed with the same primary timings. There are no timings you can tweak, primary or secondary that can make up such a large gap. And if you think that gap is inconsequential, then think again because an extra 21ns of memory latency will be noticeable.

When it comes to faster RAM, it's mostly the reduction in latency and not the increase in bandwidth that manifests as greater performance for consumer applications like games.



Well to me it's terrible. Broadwell-E's quad channel memory controller typically has higher latency than the equivalent dual channel memory controller in Skylake and Kabylake (by about 10+ns), presumably due to greater complexity. So to see my 6900K's quad channel memory latency score so much better than Ryzen's which has a dual channel memory controller, is fairly alarming and demonstrates that it's highly inefficient.



Who said that Intel does not scale beyond 3200Mhz?

You did exactly what i warned you not to do, take the best score you could find and one that is not representative (or credible) and made the false claim that there is a 20ns gap.The gap is 10-15ns as there is a large variation between mobos with Ryzen at this stage.
As for secondary timings, i have never made the claim that they make a huge difference but they do make a difference.

These results are almost all with the 0902 BIOS for the C6H so there is room for better

 

[IEC]

3433:

 

[Carfax83]

You did exactly what i warned you not to do, take the best score you could find and one that is not representative (or credible) and made the false claim that there is a 20ns gap.

What do you mean not representative or credible? That score came from my own system, and IEC is probably using the same exact RAM modules that I am using or better; G. Skill TridentZ DDR4 3200 CL14-14-14-34 (made by Samsung if I'm not mistaken). The only difference is that his is dual channel and mine is quad, and he has 16GB and I have 32GB. So all in all, it's a very good comparison.

The fact that he is using dual channel and has less populated DIMMs than I do but is still getting significantly higher latency scores, tells me a lot.

The gap is 10-15ns as there is a large variation between mobos with Ryzen at this stage.

I seriously doubt motherboard variation and even UEFI BIOS inefficiency can account for such a large gap. 10-15ns (much less 20ns) is like the difference between DDR3-1600 and DDR4-3600 or something..

As for secondary timings, i have never made the claim that they make a huge difference but they do make a difference.

Yes they do make a difference, I know this personally because I've tweaked my RAM before. I actually killed one of my sticks in fact using timings that were too aggressive. Which is why I am telling you, that secondary timings can't make up such a large gap.

The only things that can explain such a large gap, would be a massive frequency difference (obviously not the case here), or a design flaw. Since IEC and I are running at the same frequency and timings, then it must be a design flaw of some sort. A third possibility is that the benchmark is fundamentally inaccurate, but since he was using version 5.90 which is supposedly Ryzen optimized, then I suppose we can rule that one out.

These results are almost all with the 0902 BIOS for the C6H so there is room for better

Done with version 5.80, so it's not really admissible.

 

[IEC]

IMC is fine to 3600+, and that's with early AGESA and BIOS. I suspect a few more AGESA and BIOS updates will mean we will break the 60ns barrier soon enough (or maybe already broken by Elmor or other extreme OCer).

 

[guskline]

Here is an Aida64 Cache shot I just ran of my 5960x @4.4 Ghz and ram at 2133:

 

[unseenmorbidity]

My memory at 3200 and loose timings

 

[Gideon]

I seriously doubt motherboard variation and even UEFI BIOS inefficiency can account for such a large gap. 10-15ns (much less 20ns) is like the difference between DDR3-1600 and DDR4-3600 or something..

While probably correct, considerable improvements can be made:

We will soon be distributing AGESA point release 1.0.0.4 to our motherboard partners. We expect BIOSes based on this AGESA to start hitting the public in early April, though specific dates will depend on the schedules and QA practices of your motherboard vendor.
BIOSes based on this new code will have four important improvements for you

1. We have reduced DRAM latency by approximately 6ns. This can result in higher performance for latency-sensitive applications.
2. We resolved a condition where an unusual FMA3 code sequence could cause a system hang.
3. We resolved the “overclock sleep bug” where an incorrect CPU frequency could be reported after resuming from S3 sleep.
4. AMD Ryzen™ Master no longer requires the High-Precision Event Timer (HPET).

We will continue to update you on future AGESA releases when they’re complete, and we’re already working hard to bring you a May release that focuses on overclocked DDR4 memory.

sauce