FEATURED G.Skill Trident Z5 RGB 2x16GB DDR5-6000 CL36 - F5-6000U3636E16GX2-TZ5RK

[Woomack]

Today I have some results made on the latest G.Skill DDR5-6000 kit, which is widely available in stores under the product number F5-6000U3636E16GX2-TZ5RK. I assume it won't change, and this PN will be using Samsung IC as G.Skill already released new PNs with XMP that may easily fit Hynix IC. Hard to say right now, so I guess we will find out soon.
The tested memory kit can be found on the manufacturer's website here ~click.

Product photos:

Spoiler: more photos

The tested memory kit has one XMP profile of DDR5-6000 CL36-36-36-76 1.30V, which is supposed to work on the most popular motherboards. However, somehow Gigabyte Z690 Master still has problems with stability at DDR5-6000 or above, even though the G.Skill memory is on the QVL (both G.Skill and Gigabyte lists). The memory kit works without issues on ASUS Strix Z690I-Gaming (which isn't on the QVL list), so I will perform tests using this motherboard.

There are no problems with the XMP profile or overclocked settings up to at least DDR5-6600. However, I recommend additional airflow if we set higher voltages as DDR5 runs pretty hot under a longer high load.
All settings on the list passed more demanding and long tests but not necessarily long stability tests. There were no problems during tests at the listed settings.

Tests:

• AIDA64 Cache & Memory Benchmark
• Final Fantasy XV
• Shadow of the Tomb Raider 2560x1440 High Details
• 3DMark Time Spy
• PCMark 10 Extended

XMP #1 DDR5-6000 36-36-36-76 1.30V

Spoiler: test results

DDR5-6200 34-35-35-75 1.40V

Spoiler: test results

DDR5-6400 34-36-36-76 1.45V

Spoiler: test results

DDR5-6600 36-38-38-78 1.45V

Spoiler: test results

DDR5-6800 38-40-40-78 1.50V

Spoiler: test results

DDR5-6800 is already hard to stabilize and requires some additional timing tuning and higher voltages. DDR5-7000 is the maximum at which I could make this specific kit boot. I had one more kit that required lower voltages, but I couldn't stabilize it at DDR5-6800 or above.
Below is a screenshot of the AIDA64 benchmark. These settings are crashing in more demanding tests, but it's nice to see 110/100/100GB/s+ bandwidth.

DDR5-7000 40-40-40-78 1.50V




Higher voltages than 1.50V are not helping to stabilize any setting. It let me only boot at one step lower CL. CL32 is not working on my kit, even at DDR5-6200. My previous kit was working at CL32 so I assume it's a matter of luck and you can get something better for OC. It won't help much with the performance so I wouldn't push for the tightest timings at high voltages if we are going to use the PC for gaming. XMP results are already great and it's easy to set DDR5-6400 or 6600.


G.Skill Trident Z5 kits are widely available, not much more expensive than other brands and have faster XMP profiles than the (available in stores) competition. Other brands have some work to do as most of their kits are only on the product lists but are missing in stores.
G.Skill added the DDR5-6400 CL32 kit to the product list in the last few days, which is already expected to appear in some stores this week. It's going to be very expensive, but I bet there will be people who won't hesitate to order it.

Feel free to share your results, as I bet there are more users that have this memory kit.

 

[Blaylock]

Great results. Thank you Woomack for testing these. This is likely the kit I will buy once I make the switch. I'm trying to wait until AMD releases their 7000 series later this year to determine if I go team red or blue this time around. This should also give DDR5 some time to mature and possibly drop in price.

 

[EarthDog]

I've been waiting for GSkill to send me this kit...lol

Isn't this the third kit of that RAM? The first two failed?

 

[Woomack]

@Blaylock, I assume that in some weeks/months, there will be a bit different bins or IC so that you may get something better in some months. I already see PN differences (1 letter which went from A to J already, the tested kit is E), voltages went up from 1.30V to 1.35V, and single timings are higher. It suggests a different IC ... or worse chips, but somehow, everything I've seen in local stores (5600/6000 kits) is based on Samsung IC. The same PNs in press releases used for top OC results are based on Hynix IC.

@EarthDog, this is the second kit, the first one was working for about two weeks and one stick failed. I guess it's my luck as I had no faulty RAM for some long years. I may get a DDR5-6400 kit in about a week.


I forgot to add some photos today

 

[custom90gt]

Great post as always @Woomack, I just got the 5600U version of this kit and am going to try your settings. Hopefully I can get some reasonable speed out of them.

 

[Woomack]

Thanks. Hopefully DDR5-6400 C32 kit arrives soon so I will post a short forum review about it too. It is supposed to be tomorrow but hard to say as it was shipped from Denmark and tracking ends somewhere on the German border.
Btw. if I'm right then your kit is on Samsung IC too so should OC about the same. My first 6000 kit was running at tighter timings (CL32 instead of CL34) and required lower voltages to boot at 6800+ but was harder to stabilize at 6600.
I'm not suggesting any sub-timings as on ASUS and Gigabyte it works differently. Not so many sub-timings are worth tightening. I'm at work right now and can't even check anything. I will try to add some more comments in my next tests.

 

[custom90gt]

One odd thing is that at similar timings and speeds my memory has significantly slower read/writes as well as higher latency. I'm totally new to DDR5 so I'm sure it's just a learning process.

 

[Woomack]

The XMP 6000 result was at all auto timings so it should be about the same. The only difference can be motherboard training. It sometimes gives +/- 2ns latency and 1-2GB/s bandwidth at all auto settings. The same, AIDA64 itself gives about the same difference when you rerun the benchmark a couple of times.
You will see a significant difference for both, bandwidth and latency, setting tREFI at about 100k. On my ASUS it doesn't work without errors above that and on Gigabyte, the limit is about 64k. At auto, it's at 5k or something near. You can also play with other timings and set them lower by 1-2 in pairs when something seems the same but one is W and one is R, then just adjust both of them at the same time, 1-2 lower, and check if it boots and passes more demanding tests, not necessarily long stability test. This generally doesn't help much, unless you can significantly drop the value.
Most timings give about nothing so are not worth adjusting, especially that affect stability. Some timings were reacting better on DDR4.
In the end, you won't see more than 3-4GB/s bandwidth and 3-5ns latency improvement, when at 6000-6200 and auto/XMP timings it can be already up to 97GB/s+ and ~62ns. In 3D benchmarks, the difference between auto and manual at 1440p is 0-1FPS (when is already 100FPS+).

 

[custom90gt]

Yeah that's where I'm still confused, I'll show you my results. Basically I loaded XMP and tried 5600, 6000, 6200, and 6400 and ran AIDA. Even multiple runs don't give me any higher results.

Spoiler: Results

5600mhz @ 36-36-36-76 @1.2v



6000mhz @ 36-36-36-76 @1.3v


6200mhz @ 36-36-36-76 @1.35v


6400mhz @ 36-36-36-76 @1.35v

Sadly these sticks are not even looking too hot when it comes to running an AIDA64 stress test on them. Anything about 6000mhz seems to fail after ~20 minutes regardless of voltage, but again I need to look at other supporting voltages and such. I may try to pick up some hynix memory to try as well.


*On edit*
My results with tREFI look better, thanks for the tip there, my latency is still pretty far off but again will play around with it. Now if only I could get these stabilized at ~6400mhz I'd be happy with them.

Spoiler: tREFI 100k results

6200mhz @ 36-36-36-76 @1.375v tREFI 100k

 

[Woomack]

Maybe it's a matter of motherboard or what it sets under auto timings. These results are a bit too low. I have better results at auto on Gigabyte mobo which has problems with OC. I would play some with sub-timings as I guess something runs at too low value.

I couldn't fully stabilize anything above 6600. 6800 was sometimes passing 2-3h stability tests and sometimes there were single errors in memtest. On the other hand, I have a 2-slot motherboard which is supposed to be a bit better for memory OC. It's not the best mobo but I guess that many others are worse.

I don't know if it's worth spending money on an additional Hynix kit. You may have the same results as on Samsung. I mean at not so high voltages like 1.30-1.40V and 6000-6400, Samsung runs at something like 36-36-36 when Hynix runs at 32-38-38. When you bump voltages to 1.5V+ then you can go lower with CL, to about 32 on Samsung and 28 on Hynix. Still, most other timings won't change and +/- bandwidth and latency won't be much different on these two IC. To see the difference, you will need a top OC motherboard that seems too expensive to get 0-1% better performance.

I see over 80°C at 1.5V if I won't use additional airflow. I just don't feel comfortable with so high temps. Even if it's still safe for RAM then I may lose warranty when stickers fall off after a while or change colors.

 

[custom90gt]

@Woomack, sadly I am struggling to figure out which subtimings are killing my latency. Is there anyway you could post up your subtimings? Not sure if Asrock Timing Configurator works or not to do so.
Thanks

 

[Woomack]

I'm not sure if it's only a matter of sub-timings as I also have ~65ns at XMP 6200 on Hynix IC with more relaxed timings.
I will post 3 more DDR5 tests soon. I'm finishing Hynix kit tests right now and next I will back to Samsung.