You probably wont notice CL 17 unless you bench test. AMD is suppose to be coming out with a new memory update soon. Link: http://wccftech.com/amd-ryzen-memory-support-to-be-expanded-with-upcoming-microcode-update/
-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
Which RAM to use with Ryzen?
- Thread starter Thronius
- Start date
I feel like I should post back what I ended up with. After company not shipping me items, I was forced to refund and I had to order from another one where they have very limited options to pick from. Luckily enough, I purchased for 189€ HyperX Predator CL16 3200MHz (HX432C16PB3K2/16). Currently I am running it with XMP 2 on 2933 and I am waiting for a more stable bios to have it finally 3200, maybe I'll even overclock then to higher, I guess Frequency > CL17 would be worth?
Here is single sided Hynix IC
It's not about Samsung or other IC but memory build and its profile ( more like single sub timings and how BIOS can read them ).
Did you try at 3200 ? You may try higher VDIMM and DDR VTT voltages to make it work on Gigabyte board. On Samsungs it was required to run stable but I had no Hynix 3200+ kits to check it since I got ryzen. I had only one lower binned kit which couldn't pass 2800 on Intel and made exactly the same on ryzen. You can also try manual timings like 16-16-16 ( or 14-16-16 at higher voltage ).
I see that people are creating weird theories before they actually check all options. Other forums are flooded by stupid comments about memory for ryzen.
For me on Gigabyte B350 and ASUS X370 all single rank memory kits could run the same as on Intel or at 3200. All dual rank kits could make stable 2666 ( 2x8GB and 2x16GB ). I guess that new BIOS will improve max memory frequency.
All are saying that single rank Samsung B is the best for Ryzen ... just because there is almost no other single rank Samsung IC sold in 8GB modules. Hynix AFR/MFR is also single rank in 8GB modules so if manufacturer made profiles right then it will work on Ryzen.
- Thread Starter
- #43
I've only tried XMP profiles as overclocking memory manually looks really complicated, I mean, I've never tried it so I feel like I would mess it up. Same way I don't get it how could I get lower CL with 3200 if the default one is already at CL16?
The XMP1 (3200) profile won't work for now, the new BIOS should come this week with new AGESA so I hope it will change it a bit. Also the BIOS software is quite buggy, I had cold boot, no boot at all, bios errors and such, even bios not saving properties.
I will try to play with VDIMM and DDR VTT once the new bios rolls out. My true goal would be 3600MHz on this RAM if possible, with having CPU at 3.7-3.8, I guess that should give me plenty of room for any title.
You won't be able to set 3600 on this memory kit or it will be hard to stabilize. I've reviewed similar memory kit some time ago. Most Hynix based kits won't make more than ~3466. The best Hynix based 2x8GB kit which I had a chance to test made 3600 17-19-19 at lower voltages ( 1.40V or something near ). Some other from new series couldn't pass 3200 or 3333.
- Joined
- May 4, 2011
Yep. the difference between 2600 and 3200 for gaming is 1 to 3 fps, if that.
In most cases difference won't be big but on Ryzen everything is linked to one bus so higher memory performance is affecting everything else. Depends on used software and additional hardware you may see higher performance gains while using higher memory frequency. On the other hand tighter timings are not helping much. The same frequency perform about the same regardless if memory runs at CL14 or CL16 while there are couple of FPS difference between 2666 and 3200. I think on TPU was larger memory test in games with these results.
So memory frequency counts when you see lower FPS but totally won't make any difference when you already have 60FPS+.
So memory frequency counts when you see lower FPS but totally won't make any difference when you already have 60FPS+.
With my memory i ended up at march BIOS with 3200CL13 or 3600CL18 on my x370 taichi board. Can say that 3200 gives big boost in performance above 2133. 3600 gives boost but not very noticeable above 3200. Still hope they will release new updates for faster RAM clocks and lower latency.
- Thread Starter
- #49
There are recently posts popping up where they claim Dual Ranked RAM gives extra performance compared to Single Ranked RAM (Like, 2666 Dual Ranked gives performance of 3200 Single Ranked RAM).
What do you think about those claims? I really hope thats temporary and new AGESA 1006 will bring better performance.
http://www.pcgameshardware.de/Ryzen-5-1600X-CPU-265842/Tests/R5-1500X-Review-Mainstream-1225280/3/
On the side note, I've managed to change my RAM latency from 17-17-17-39 to 14-15-15-30, I am hoping that with new AGESA I will be able to reach 3200MHZ and possibly overclock to 3333MMHz as sadly Hynix doesent OC that well.
I already said that while testing memory on Intel that dual rank is faster than single rank but there is a limit and not all applications can use that additional performance. On AMD I haven't seen so big difference. I was testing 2x16GB 2666 memory on Ryzen and it was performing worse than 2x8GB 3200. The same dual rank 2666 kit on Z270 board was performing between 2666 and 3200 single rank ( depends on used test ). There are test results in my memory reviews. Almost all results were made on the same CPU/cache speed so it's easier to compare.
New dual rank memory generally can't OC above 3200. Some Samsung B kits can make more but not many. Single rank memory can make up to 4133+. In the end in applications which can use additional ranks, 3200 dual rank will be about as fast as 3733 single rank and in applications that can't use it, 3200 dual rank can be even a bit slower than single rank because of more relaxed sub timings ( not always, depends on IC and programmed profiles ).
Problem is that new dual rank = 16GB modules, single rank = 8GB so it's up to you if you care about a bit higher performance in single applications for about 2x higher price.
I hope we will see higher memory ratios with new AGESA. I'm actually surprised that there are no new BIOS releases with 1006 as it was out some time ago and for some boards we got 1004a instead of 1006.
New dual rank memory generally can't OC above 3200. Some Samsung B kits can make more but not many. Single rank memory can make up to 4133+. In the end in applications which can use additional ranks, 3200 dual rank will be about as fast as 3733 single rank and in applications that can't use it, 3200 dual rank can be even a bit slower than single rank because of more relaxed sub timings ( not always, depends on IC and programmed profiles ).
Problem is that new dual rank = 16GB modules, single rank = 8GB so it's up to you if you care about a bit higher performance in single applications for about 2x higher price.
I hope we will see higher memory ratios with new AGESA. I'm actually surprised that there are no new BIOS releases with 1006 as it was out some time ago and for some boards we got 1004a instead of 1006.
- Joined
- Dec 18, 2000
I'm not surprised, just disappointed with the motherboard vendors. Having played with ASRock, Asus and MSI AM4 motherboards, I just don't see that any of them are making much of an effort on this, mostly waiting on AMD. But AMD is new to, and clueless with DDR4, and the motherboard manufacturers are experts on DDR4 having used it for years in Intel motherboards. My current ASRock mATX B350 board lists different memory on it's QVL than the two ATX B350 boards they sell. So the Corsair RAM I have now, which ran fine on the ATX version at 2933, is stuck at 2666 on the mATX variant. The most recent BIOS incorporating the 1004a AGESA did nothing to change that. One would think you would work with the same RAM on all motherboards using the same chipset but they don't. Contacting ASRock support with info on BIOS issues was pointless, as they just stopped communicating apparently not wanting actually fix anything. Their answer to several issues, none of which were catastrophic, was to return the motherboard and get another brand.
In 1.0.0.6 update they added new ratios, so now possible to choose 3066 3333 3466 3600 3733 3866 and 4000. Also 1.0.0.6 update should start to hit consumers next week (for asus and asrock at least). I hope that with stable 1.0.0.6 BIOS on my taichi i'll get 3600 or 3733 with CL16 without bclk.
oh... u found some very crappy RAM, cos i got 3200 even on cheapest OEM sticks. And u have ch6 which is much better for OC.
btw can some1 explain me how to tune ryzen for better power consumption? I kinda hate it already, when 65W TDP rated CPU without OC draining 120W under just 5-15% load... Need to fix their stupid turboboost, which is turn vcore to 1.4V from 4-5% CPU load. If i OC CPU i have less power consumption on low CPU load, but much more on idle
Examples: no OC = 10W idle, 120W low load, about 140W full load. OC to 3.8 = 30W idle, 60W low load, 200W full load. When OCed vcore dont drop below 1.1V, when no OC its drop to 0.9V and at the same time while OCed below 4.0GHz it never hits 1.4V vcore (unlike none OC mode).oh... u found some very crappy RAM, cos i got 3200 even on cheapest OEM sticks. And u have ch6 which is much better for OC.
btw can some1 explain me how to tune ryzen for better power consumption? I kinda hate it already, when 65W TDP rated CPU without OC draining 120W under just 5-15% load... Need to fix their stupid turboboost, which is turn vcore to 1.4V from 4-5% CPU load. If i OC CPU i have less power consumption on low CPU load, but much more on idle
It run higher on Intel, but on Ryzen it dosen't I think I can hit around 288XMhz forgot what number but I don't think is stable at all.
From my previous post:
Did u try agesa 1.0.0.6? On taichi i run my RAM 3600 without bclk.
I tried whatever bios is currently on asus crosshair vi hero am4 support site. No luck yet with it.
- Joined
- Feb 25, 2017
- Location
- Lurking over a keyboard
Samsung B die is best for Ryzen even with new AGESA 1.0.0.6. This view is formed a) from my own rig b) members shares on various Ryzen threads with results.
I wouldn't pay the premium for FlareX vs Trident Z, the author of Thaiphoon Burner has highlighted they are identical, link. I have also seen where some Ryzen owners have Trident Z & FlareX their experience was the Trident Z was better for their OC exploits.
I wouldn't pay the premium for FlareX vs Trident Z, the author of Thaiphoon Burner has highlighted they are identical, link. I have also seen where some Ryzen owners have Trident Z & FlareX their experience was the Trident Z was better for their OC exploits.
It's not a surprise when Hynix IC can't run at so tight timings as Samsung and in mass sales there are no new Hynix kits above 3333. Micron can't run at much more than 3000 so is not counting at all unless you want high capacity kit in good price and don't really need high frequency.
All memory kits with the same IC are about the same. The only issue which can happen can be error in memory profile or memory profile designed for much higher frequency that motherboard can't translate to correct settings within timing table. Since all G.Skill memory kits were designed in the same way then all have similar profiles and generally work the same. However more problems are with 4000+ memory kits than ~3200 based on the same IC ( well, Samsung as there is no other IC that can make 4000+ ).
So far the best working kits are Trident Z 3200-3600. FlareX is the same as Ripjaws V or Trident Z on the same IC, just different heatsinks. I also have exactly the same experience with Samsung based kits from other brands.
All memory kits with the same IC are about the same. The only issue which can happen can be error in memory profile or memory profile designed for much higher frequency that motherboard can't translate to correct settings within timing table. Since all G.Skill memory kits were designed in the same way then all have similar profiles and generally work the same. However more problems are with 4000+ memory kits than ~3200 based on the same IC ( well, Samsung as there is no other IC that can make 4000+ ).
So far the best working kits are Trident Z 3200-3600. FlareX is the same as Ripjaws V or Trident Z on the same IC, just different heatsinks. I also have exactly the same experience with Samsung based kits from other brands.
Similar threads
