Ryzen with 3600MHz RAM Benchmark

I got my 32GB (2x 16GB) kit running at 2666 15 CAS w/ 1.35v. Seemed to work alright, but I didn't do any intensive stress tests.
 
Very weak troll bait. Even ignoring that CPUs are only part of Intel's business, Ryzen isn't destroying anything of Intel's thus far. It's a decent competitor for the price; nothing more and nothing less.



We've known Intel's roadmap for quite some time now. I believe his post was meant fasciously as troll bait.

Troll bait to an Intel fan, valid and interesting question to an objective mind.
 
Troll bait to an Intel fan, valid and interesting question to an objective mind.

That fact that it was posted as a question was facetious at best. We've known Intel's roadmap for quite some time, so the question was answered long ago and has no merit now.

In case you somehow weren't aware.
 
Whoa I just realized what you're referring to... I hadn't noticed that there was a test size discrepancy! That's interesting and peculiar... I have no clue as to why that happened other than to suspect as a result of overall Cache Size changing due to disablement of a 4C module.

I've done my absolute best to modify the first 8C/16T to only be the 128MB worth of data and still retain the correct overall size as the 4C/8T for ease of back-and-forth comparisons.

View attachment 20739

When you get time please could you consider testing 1+1 core setup as it may provide some interesting insight to how AIDA calculates overall data set test size and also to see how the L3 cache works per CCX with cores disabled; some of it may be more of a validation follow-up test.

Thanks and much appreciated.
 
When you get time please could you consider testing 1+1 core setup as it may provide some interesting insight to how AIDA calculates overall data set test size and also to see how the L3 cache works per CCX with cores disabled; some of it may be more of a validation follow-up test.

Thanks and much appreciated.
Can do. :p
 
The Flare X seem like a surefire bet. Says made for Ryzen and it's on at least MSI's QVL list. 3200 seems to be the max (without overclocking) so I'll probably just stick with that unless the situation drastically changes.

Only thing that could be better is I had to drop to 16GB from the 32GB I'm at now. 16 is still "enough" but I'd have preferred the full 32GB if given a choice (at desired speed).

I received my Flare X 3200 today and it's running solid as a rock in my ASUS x370 Pro at 3200.
 
When you get time please could you consider testing 1+1 core setup as it may provide some interesting insight to how AIDA calculates overall data set test size and also to see how the L3 cache works per CCX with cores disabled; some of it may be more of a validation follow-up test.

Thanks and much appreciated.
So #1, virtually unchanged results still.
#2, apparently I CAN change the Stride size! By chance I just right-clicked on the graph and up popped more options :| 1024, 2048 and 4096 Bytes are also available above the default, and 256, 128 and 64 Bytes below it.
#3, theory confirmed about the max amount it will try transferring being based off how much cache is available. With both CCX available the test once more runs up to 256MB.

Note: Given the proximity of this run to the system being booted up, the oddness that occurred at the 16K/32K area MAY have been something happening in the background. Nevertheless, the results are as I mentioned, virtually the same. I presume that the bulk of this is due to there only being 4 threads and just not enough muscle to transfer the data between the Cache and the RAM, given that at 4+0 the Memory bandwidth saw a few GB/s dip. Though... it looks like that might not be true! Tests on the memory speeds are in at 1+1 (still DDR4-3200 14-15-15): Read 45.1GB/s, Write 46.4GB/s, Copy 41GB/s... However, looks like it still IS true i a sense, as the Cache scores are in and they're pretty much looking to be at 100GB/s per Core on L1/2 reads, 50GB/s per Core on L1/2 Writes, and screw math but roughly same on Copys. L3s are more math that generally equates to the same fraction of 8 lol

Anyways, here are your results at 1+1:
cachememlat_0512B-stride - Ryzen @ Stock 1+1.png

EDIT: It isn't boosting beyond the 100MHz over stock it seems, so 3500MHz, where as in 4C4T mode it's hitting 3900. Mind you I'm basing this on the FPU VP8 results indicating 3500. Monitored it while it ran and it only hit 3430.
 
Last edited:
So #1, virtually unchanged results still.
#2, apparently I CAN change the Stride size! By chance I just right-clicked on the graph and up popped more options :| 1024, 2048 and 4096 Bytes are also available above the default, and 256, 128 and 64 Bytes below it.
#3, theory confirmed about the max amount it will try transferring being based off how much cache is available. With both CCX available the test once more runs up to 256MB.

Note: Given the proximity of this run to the system being booted up, the oddness that occurred at the 16K/32K area MAY have been something happening in the background. Nevertheless, the results are as I mentioned, virtually the same. I presume that the bulk of this is due to there only being 4 threads and just not enough muscle to transfer the data between the Cache and the RAM, given that at 4+0 the Memory bandwidth saw a few GB/s dip. Though... it looks like that might not be true! Tests on the memory speeds are in at 1+1 (still DDR4-3200 14-15-15): Read 45.1GB/s, Write 46.4GB/s, Copy 41GB/s... However, looks like it still IS true i a sense, as the Cache scores are in and they're pretty much looking to be at 100GB/s per Core on L1/2 reads, 50GB/s per Core on L1/2 Writes, and screw math but roughly same on Copys. L3s are more math that generally equates to the same fraction of 8 lol

Anyways, here are your results at 1+1:
View attachment 20870

EDIT: It isn't boosting beyond the 100MHz over stock it seems, so 3500MHz, where as in 4C4T mode it's hitting 3900. Mind you I'm basing this on the FPU VP8 results indicating 3500. Monitored it while it ran and it only hit 3430.

Ah your the man :)
Ok that definitely confirms the shared L3 cache is maintained in its entirety even with 1+1, so fits my thoughts that any simulations reviews of say the 4C 1500 have to be taken a bit wary.
Still wierd what is happening at say the 6-8MB (depending if AIDA or the bespoke tool hardware.fr used).

From watching the tool, does it tie affinity to a particular thread and core and so CCX?
Just wondering if that is why 8MB to 16MB has same latency as system memory as it must page memory rather than use both CCX.
Or from a coherent perspective it is not truly an overall 16MB L3 cache but always just 2 separate 8MB L3 cache and then has to go to system memory after 8MB.

We know the improvements to AoTS came down to how they use cache/memory.
Thanks again.
 
Last edited:
Ah your the man :)
Ok that definitely confirms the shared L3 cache is maintained in its entirety even with 1+1, so fits my thoughts that any simulations reviews of say the 4C 1500 have to be taken a bit wary.
Still wierd what is happening at say the 6-8MB (depending if AIDA or the bespoke tool hardware.fr used).

From watching the tool, does it tie affinity to a particular thread and core and so CCX?
Just wondering if that is why 8MB to 16MB has same latency as system memory as it must page memory rather than use both CCX.
Or from a coherent perspective it is not truly an overall 16MB L3 cache but always just 2 separate 8MB L3 cache and then has to go to system memory after 8MB.

We know the improvements to AoTS came down to how they use cache/memory.
Thanks again.
Up to 16K it's pretty much just Core1-Thread1 (aka Core0 in Affinity speak) loaded to 100%. After that, Thread2 (aka Core1) starts to see some usage, maybe... 5%? Around that 1M step-up it begins to see probably 12% and around that 8M step spikes every couple seconds up to maybe 40%. Then on the last few, maybe around 32M there's some spikes that occur on Core2-Thread1 (aka Core2) of maybe 25%, and Core2-Thread4 (core3) of maybe 20%, at probably 2sec intervals. I couldn't tell if they happened to trade off or occur at the same times. This is on Balanced..
Running on HiPerf now, but honestly, am not seeing much change except that since all cores are unparked it decided to tap Core4-Thread1. Ok now it tapped into Core5-Thread2 which I've now seen that those spikes are occurring during the Full Random.

Overall the HiPerf exhibited far LESS additional core loading as Balanced did. That Core4-Thread1 load was maybe 2% when it picked up at the same spot around the 16-32K area, but remaining that 2% the entire time. Those spikes were roughly the same % as Balanced.

For the luls I tried to force the test to run on Affinity Core 6 and 12 only, but in typical AIDA fashion it does what it wants and is once again doing the exact same thing on the exact same cores. I can now say though that the Full Random spike first occurs at....... oook. It did not present itself this time?? Weird. :|

At any rate, now that I found out last night how to run a different Stide size, here it is at 1KB (and unless something totally interesting stands out to you, I dunno if I'll bother changing which CCX becomes active given the similarity between runs):
Added both that I had just now ran at HiPer Profile, 512b | 1024b
cachememlat_0512B+1024B-stride - Ryzen @ Stock 4+4 HiPofile.png
 
I always thought that it is the APU that needs faster DDR Memory. Does faster memory helps a Ryzen CPU ?
 
Yes, but I have you read my remark ?
What he was implying, was that the answer to your question is contained somewhere within the 200 previous posts of this thread. :)

However, as a favor: Yes, Ryzen currently loves RAM speed, but loves lower Subtimings (which we do not currently have access to) at high speeds even more.
This is because certain parts of the Infinity Fabric (the interconnecting circuits inside Ryzen that link various elements) run at 1/2 DRAM speed. Therefore the higher the DRAM, the higher that specific circuit can operate at. :)
 
Hi, I'm new here but found this thread very useful, I'm just buying my first new pc build in quite a while and I decided to go with Ryzen because of the multi-threading for 3d and video editing (plus it just feels far more exciting than intel at the moment). Can anyone recommend a good 3600 DDR4 ram set? I was looking at the Corsair Vengeance LPX, any experience with that one? (http://www.corsair.com/en-gb/vengea...600mhz-c16-memory-kit-red-cmk32gx4m4b3600c16r) Is there a benefit for keeping it at 2x8, or would 4x8 still work fine? Thx

Just to add to the other recommendation/ If you want to be absolutely sure you get high ram speed on Ryzen order Ryzen certified memory. The only one offering high speed Ryzen certified memory now is G.Skill Flare X brand. You have two high speed choices with Flare X. One is ddr4 3200 cas 14-14-14 or DDR4 3466 CL 16-16-16. This will guarantee a 2 dimm kit willoperate at rated speed. 4 dimm kitss will not give high speeds right now. Perhaps after May when the AMD agesa memory fix is encapsulated into new bioses that situation may improve.
 
Hi guys,

So would it be better to get 24GB or 32GB if I am trying to achieve 3000+ mhz RAM on Ryzen? Have not got a MB yet, so any suggestions would be welcome. What is currently the highest timing and frequency achieved for either 24GB or 32GB RAM??
 
You are better off with 2 sticks rather than 4. However, 8GB sticks seem to clock higher than 16GB.

If you really need the memory, then 16GB x 2 is preferred, though 8GB x 2 may have better luck with the speed.

However, each RAM stick has different properties. Find a few motherboards and look on their QVL list on their website and see if there are any sticks that hit the speed you want and buy that specific model.

For example, with my mobo it is only confirmed to reach 2667 with 16GB on a few specific ADATA sticks. But I was able to overclock my previous G.Skill TridentZ sticks (16GB x 2) to 2667 after tweaking voltage and timings.

That said, if you would rather have the speed, than the 16GB Flare X 3200 kit is the best option right now.
 
You are better off with 2 sticks rather than 4. However, 8GB sticks seem to clock higher than 16GB.

If you really need the memory, then 16GB x 2 is preferred, though 8GB x 2 may have better luck with the speed.

However, each RAM stick has different properties. Find a few motherboards and look on their QVL list on their website and see if there are any sticks that hit the speed you want and buy that specific model.

For example, with my mobo it is only confirmed to reach 2667 with 16GB on a few specific ADATA sticks. But I was able to overclock my previous G.Skill TridentZ sticks (16GB x 2) to 2667 after tweaking voltage and timings.

That said, if you would rather have the speed, than the 16GB Flare X 3200 kit is the best option right now.


Flare X also has a kit running at 3466mhz.
 
You are better off with 2 sticks rather than 4. However, 8GB sticks seem to clock higher than 16GB.

If you really need the memory, then 16GB x 2 is preferred, though 8GB x 2 may have better luck with the speed.

However, each RAM stick has different properties. Find a few motherboards and look on their QVL list on their website and see if there are any sticks that hit the speed you want and buy that specific model.

For example, with my mobo it is only confirmed to reach 2667 with 16GB on a few specific ADATA sticks. But I was able to overclock my previous G.Skill TridentZ sticks (16GB x 2) to 2667 after tweaking voltage and timings.

That said, if you would rather have the speed, than the 16GB Flare X 3200 kit is the best option right now.

Thanks for your reply. For MBs, it would be one of these: Asus Crosshair VI, the ASrock x330 Taichi or the MSI x370 Titanium / Carbon.

As for RAM, I would need a min of 24GB or 32GB, 16 is just not enough for my non-gaming needs.

What is the best 16x2 kit right now with 3000-3600 Mhz and good timing (CL15 or CL16)?
 
The QVL is listing RAM kits that they've tested to run at rated speed. Supposedly no dual rank (16GB sticks) can run at 3200 right now.
 
Looks like the Asus is the only one confirmed to be able to hit 3200 on 32GB.
http://dlcdnet.asus.com/pub/ASUS/mb...0316.pdf?_ga=1.190964906.120583173.1489863893

Specifically with these kits:
F4-3200C16Q-32GFX
F4-3200C16Q-32GFXR
Though I can't seem to find either of those sold anywhere (at least in the US).

On ASrock TAichi
http://www.xtremesystems.org/forums...-of-the-Jedi&p=5257198&viewfull=1#post5257198

2x 16g Dual Rank PC 3200 @ 14-14-14-34

Some other folks getting results on crosshair VI hero of 3600 with newer beta bios.
 
Ok so from what I can gather so far reading many threads on different sites, the following are the most promising:

F4-3400C16D-32GTZ - 16-18-18-36 - Trident Z

F4-3200C15D-32GVK - 15-15-15-35 - G.Skill Ripjaws 5

F4-3733C17D-8GTZA - 17-17-17-37 - 4x8GB TridentZ

If someone can add or provide more details to the list, please do so
 
I wanna create a list and keep it up to date so buyers save tons of hours researching here and there everywhere. I know Ryzen is still new and many tests are needed, but nothing better than finding all the info in one place
 
I wanna create a list and keep it up to date so buyers save tons of hours researching here and there everywhere. I know Ryzen is still new and many tests are needed, but nothing better than finding all the info in one place

Someone put a list on Reddit.

Don't know how accurate it is but you might be able to add or atleast confirm which sticks work best on Ryzen with HardOCP members adding in their results.
 
Nice list, though they only have 1x8GB sticks.

Seems the 4x8GB rated 3600+ are getting 3600 and 2x16GB rated 3400 are getting 3400... I think this is where it would make most sense to buy ATM
 
First time I've seen it broken down like this



If you check what he says from this point on it gets entertaining. Suggesting that a BIOS might fix problems for overclocking lower latency memory kits (he rambles on about how things are not officially supported as well clearly stating that it might work or not and if not to bad) , then he goes on that the performance can be so little and only for certain benchmarks that I'm wondering why start this endeavour down the road of futility?

And as a counterpoint he throws in his record of firestrike where he achieved a memory frequency of 3600 but that was with an expensive piece of ram with way better timings.

Talk about mixed messages .....
 
Last edited:
If you check what he says from this point on it gets entertaining. Suggesting that a BIOS might fix problems for overclocking lower latency memory kits (he rambles on about how things are not officially supported as well clearly stating that it might work or not and if not to bad) , then he goes on that the performance can be so little and only for certain benchmarks that I'm wondering why start this endeavour down the road of futility?

And as a counterpoint he throws in his record of firestrike where he achieved a memory frequency of 3600 but that was with an expensive piece of ram with way better timings.

Talk about mixed messages .....

Ryzen itself is a mixed fucking message, what do you want.
 
Nice to see that you can squeeze some more performance with better RAM, and it's similar to what I've seen after getting the Flare X kit.

Sadly, it does not appear to be reaching 7700K level gaming performance, and I'm starting to doubt it ever will. In any case, I am still happy with the purchase and I am getting decent performance so I'm not complaining.
 
If you check what he says from this point on it gets entertaining. Suggesting that a BIOS might fix problems for overclocking lower latency memory kits (he rambles on about how things are not officially supported as well clearly stating that it might work or not and if not to bad) , then he goes on that the performance can be so little and only for certain benchmarks that I'm wondering why start this endeavour down the road of futility?

And as a counterpoint he throws in his record of firestrike where he achieved a memory frequency of 3600 but that was with an expensive piece of ram with way better timings.

Talk about mixed messages .....

Exactly. What that video needs is a benchmark with the DDR4 @2400, and another benchmark with DDR4 @3000 or 3200. All I see is increasing the timing from 17 to 22, and while that might get him from 2400 to 3200, slower timing might just mean the same or even poorer performance....!

And like you said, the real benchmark was actually done with a better RAM above 3000Mhz, so ya.
 
Nice to see that you can squeeze some more performance with better RAM, and it's similar to what I've seen after getting the Flare X kit.

Sadly, it does not appear to be reaching 7700K level gaming performance, and I'm starting to doubt it ever will. In any case, I am still happy with the purchase and I am getting decent performance so I'm not complaining.

Well, I think its safe to say that with the game optimizations done so far and the better memory support you probably will get close enough to the 7700 that it wont matter anymore, meaning, your GPU will be an issue before Ryzen 7 is. Its those 4 extra cores (8 extra threads) that would matter the most, and thats something the 7700 will never be able to achieve.

For pure hardcore gaming nothing will beat the 7700K because it is purpose built for that need.

Everything else, nothing will beat the 1700, even for 90% of the games, and to me thats a big win for Ryzen 7!
 
Morrrrrrrrrrrrrrrrrrrrrrrre Ryzen mem Benches, this time is a more narrow look at 2933 vs 3200, there is gains but as you would expect they are minimal in most cases as 2933 is still pretty dam good as is.
 
LOL would not the FlareX be the same exact specs as the G.Skill Trident Z 3200 MHz CL14 ?? It sure does looks so!

Anyway what I would like to see is how the FlareX 3200 CL14 compares to the F4-3400C16D-32GTZ - 16-18-18-36 and the F4-3733C17D-8GTZA - 17-17-17-37, but my guess here is that it would be almost identical.
 
Last edited:
The 2x8GB FlareX 3200 CL 14 is selling at $183 USD on newegg, and the same kit for the Trident Z 3200 CL 14 is selling for $180 on special (original is $200 USD)
 
LOL would not the FlareX be the same exact specs as the G.Skill Trident Z 3200 MHz CL14 ?? It sure does looks so!

Anyway what I would like to see is how the FlareX 3200 CL14 compares to the F4-3400C16D-32GTZ - 16-18-18-36 and the F4-3733C17D-8GTZA - 17-17-17-37, but my guess here is that it would be almost identical.

The difference is, the Flare-X is guaranteed to run at those speeds on a Ryzen mobo out of the box, the G.Skill Trident Z kit isn't.
 
Last edited:
Yeah I would like someone to make sense on youtube not just talking for the sake of talking and saying nothing .....

It makes perfect sense if you don't have an AMD branded erection lasting longer than 4 hours.

He shows the math and tells you some programs respond better to infinity fabric clock, and some to memory latency. Not rocket surgery.
 
Last edited:
Back
Top