G.SKILL Trident Z5 DDR5-7200 CL34 2x 16 GB

[ir_cow]

G.SKILL is at it again pushing the bar and raising expectations. Using the newest DRAM revision, the G.SKILL Trident Z5 DDR5-7200 brings us an enthusiast-grade memory kit marketed towards hardcore gamers and PC enthusiasts alike. With serious high frequency overclocking, follow along as we see what G.SKILL has in store!

Show full review

 

[RogueSix]

From personal experience, I would advise caution on boards with four DIMM slots. I'm not sure how much time and effort TPU put into true^TM stability testing but the gold standard is still Prime95.

I have an ASUS ROG STRIX Z790-E mainboard, i9-13900K CPU and 32GB of Corsair Vengeance DDR5-7200.
At 7200MHz and 1.48V, I was able to get this setup to "fake stability", i.e. I was able to run the Karhu memory test to ~12000% coverage but then still got errors in Prime95 (after two to three hours of testing). The Karhu folks are saying on their website that 6400% coverage catches 99.41% of errors. Um, well, yeah...

After lots of testing, I finally settled for DDR5-6800 at 1.46V. This was the first and only setting, working my way down from 7200MHz, that I was able to get Prime95 stable (~15 hours or so).
I have read quite a few user reports with similar experiences as mine. The 7000MHz barrier is a pretty tough wall to break on a board with four DIMM slots (on a two DIMM slot ASUS APEX it might be significantly easier to accomplish these clocks).

Be wary of "fake stability". Do not trust in one test only that might seem convenient and that appears to produce fast, pleasing results.
Invest the time and effort to test stability properly. Prime95 has always been and will always be the gold standard due to its methodology. If you get errors in Prime then there is something wrong with your system. Period.
You do not want to run a system with "hidden" stability issues. Some games are very sensitive to bad (RAM) OC, so to be on the 100% safe side that you have a crash-free system, make sure that your system is Prime stable. Yes, there are people saying that Prime loads are "unrealistic" for every day usage and all that blahblah... in reality most of those people simply do not like the results of their Prime95 testing .

 

[megaclite]

From personal experience, I would advise caution on boards with four DIMM slots. I'm not sure how much time and effort TPU put into true^TM stability testing but the gold standard is still Prime95.

I have an ASUS ROG STRIX Z790-E mainboard, i9-13900K CPU and 32GB of Corsair Vengeance DDR5-7200.
At 7200MHz and 1.48V, I was able to get this setup to "fake stability", i.e. I was able to run the Karhu memory test to ~12000% coverage but then still got errors in Prime95 (after two to three hours of testing). The Karhu folks are saying on their website that 6400% coverage catches 99.41% of errors. Um, well, yeah...

After lots of testing, I finally settled for DDR5-6800 at 1.46V. This was the first and only setting, working my way down from 7200MHz, that I was able to get Prime95 stable (~15 hours or so).
I have read quite a few user reports with similar experiences as mine. The 7000MHz barrier is a pretty tough wall to break on a board with four DIMM slots (on a two DIMM slot ASUS APEX it might be significantly easier to accomplish these clocks).

Be wary of "fake stability". Do not trust in one test only that might seem convenient and that appears to produce fast, pleasing results.
Invest the time and effort to test stability properly. Prime95 has always been and will always be the gold standard due to its methodology. If you get errors in Prime then there is something wrong with your system. Period.
You do not want to run a system with "hidden" stability issues. Some games are very sensitive to bad (RAM) OC, so to be on the 100% safe side that you have a crash-free system, make sure that your system is Prime stable. Yes, there are people saying that Prime loads are "unrealistic" for every day usage and all that blahblah... in reality most of those people simply do not like the results of their Prime95 testing .

True about fake stability i can add to this list HCI MemTestPro.
150% pass MemtestPro while in TM5 extreme errors in 2 minutes or OCCT cpu avx
Best stabillity check is AVX load: prime95, OCCT, Y-cruncher VST

You cant run stable 7200 because you didn't found good VDDQ TX voltage value for your CPU
Z790-E strix for sure can do 7200

 

[ir_cow]

From personal experience, I would advise caution on boards with four DIMM slots. I'm not sure how much time and effort TPU put into true^TM stability testing but the gold standard is still Prime95.

A lot of time is put into stability checking . Prime95 can be problematic as it just cooks the CPU and IMC. Need a proper cooling solution for that. A stock Intel 13900K (PL1=4096w) can't handle Prime95 (without xmp enabled) unless you have a way to dissipate 400+ watts. Just food for thought.

Speaking of high temps. This is also why cold boots yield the highest memory overclocks. It's not technically cheating, but some of the HWBOT top submissions is done straight from a cold boot (if not LN2).

So either you can set the CPU to 4GHz and run DDR5-8000 with Prime95, or you can run other programs like TM5 (25 cycles), MemTestPro or Karhu (2000%) + y-cruncher 2.5B (a few times in a row) instead and call it a day. No way you will pass both and not be 99% memory stability. Only thing you have to worry about is heat build up. Like say if your video card is blow hot air on the memory.

I always suggest doing your memory and CPU overclocks separately to find the maximum both can do before combining. This lets you find what each can do without adding more unneeded factors that lead you down the rabbit hole of trying to figure out what's wrong. I settled on 5.5P / 4.3 E because its a happen medium for daily use. If i'm pushing records, I'll go to 5.7 / 4.6, but that isn't good for daily use. Not without deliding the CPU or doing direct die (for me).

 

[RogueSix]

You cant run stable 7200 because you didn't found good VDDQ TX voltage value for your CPU
Z790-E strix for sure can do 7200

Well, I "accidentally" tried everything from 1.45V all the way up to 1.51V . "Accidentally" because initially I did not know how the voltages are linked together. If you enter a value for VDD voltage in the ASUS ROG STRIX Z790-E BIOS then it automatically applies the same value to the VDDQ TX voltage which is actually called IVR Transmitter Voltage in the ASUS BIOS.
You have to manually set a different value for IVR Transmitter Voltage (VDDQ TX) if you want to test it separately from the VDD voltages.

Raising VDD voltages definitely helped with stability. I received the errors in Prime95 later (after two to three or more hours) and they were fewer with only one worker quitting on me but I could not find a combination of voltages at 7200MHz and 7000MHz that achieved 100% stability.
That is with the 0703 BIOS. There is now a 0813 BIOS but I have mostly seen reports of reduced DDR5 OC compatibility so I have not upgraded my BIOS. I'm waiting for a new version before I do any further testing.

I have seen at least one other guy on a German forums thread who shared my experience. He was also only able to push his Corsair 7200MHz kit to 6800MHz. Anything at or above 7000MHz ended with errors in various testing tools.
So maybe the Corsair Vengeance memory is at fault here(?). I have not tried Teamgroup or G.Skill. From what I have read, Corsair does have a temperature advantage at least. Its heatsink is supposed to be more capable than the relatively poor G.Skill heatsink.

Anyway, I'm perfectly fine with 6800MHz at CL34. My system performs great. I want an everyday 24/7 stable build that will also work in the summer. I really do not care about 1 or 2 fps that I would get from running 7200MHz vs. 6800MHz. It would have been nice to reach the nominal clocks of the memory kit but if I can not get it 100% Prime95 stable then I just gotta admit defeat... for now (as I said, I might try again when a new BIOS is released).

 

[AnarchoPrimitiv]

a bit off topic, but I'm just curious as to why the test system OS is on a SATA SSD and not on the NVMe SSD that the "data" is on?

Oh, and if anyone is an expert at memory, can you explain why there's still DDR4 outperforming DDR5 in these benchmarks?

 

[ir_cow]

I found different motherboards have different sweet spots for SA, VDD2 and TX (also CPUs too). I personally start with SA 1.25 / VDD 1.4 / TX 1.4 and go from there. Often higher SA or TX and the system won't boot. You don't have to go high on the memory to find this out. Do like DDR5-6000. Like I know my CPU gets mad at me if I go with 1.5 on the TX. Can't pass anything. Once you hit the soft limit with whatever voltages YOU are comfortable with. NOT what the internet says. Adjust one at the time. Go down .25V on the SA, Go up .25V SA. just go all the combinations and keep a sheet.

This is old notes from 12900K and I was playing around with 1T one day. I think my problem was actually the memory kit and scared to do 1.3 SA. At this point 6600+ retail didn't exist yet.

Edit: It helps to have a pre-binned kit. For example, recently while going for DDR5-8000, I had the kit but couldn't get it stable on the GB Z690 Tachyon. After entering pretty much all the sub-timings manually and voltages, it worked. Now with this ASUS Z790 Apex, 8000 MT/s is set and go. Memory was never the problem. It was the previously motherboards inability to properly train the memory. 7800 MT/s on the Z690 Tachyon - no problem. 8000 MT/s refuses to boot without manually entering everything. Lots of factors to consider when overclocking memory, even if you are just using the XMP profile.

a bit off topic, but I'm just curious as to why the test system OS is on a SATA SSD and not on the NVMe SSD that the "data" is on?

Because the data drive can be used on all the computers. Each SATA has the OS for that platform (AMD / Intel). Also copies that I can corrupt and not worry about. Much easier to swap SATA or unplug it instead of removing M.2 heatsinks all day long.

 

[GhostMotley]

When you tested the kit and it hit 69c without a fan, using the XMP profile, did it actually produce errors at that temp?

 

[Vario]

Pretty well written review and I liked your explanation of GPU bound scenarios.

 

[fevgatos]

A lot of time is put into stability checking . Prime95 can be problematic as it just cooks the CPU and IMC. Need a proper cooling solution for that. A stock Intel 13900K (PL1=4096w) can't handle Prime95 (without xmp enabled) unless you have a way to dissipate 400+ watts. Just food for thought.

Speaking of high temps. This is also why cold boots yield the highest memory overclocks. It's not technically cheating, but some of the HWBOT top submissions is done straight from a cold boot (if not LN2).

So either you can set the CPU to 4GHz and run DDR5-8000 with Prime95, or you can run other programs like TM5 (25 cycles), MemTestPro or Karhu (2000%) + y-cruncher 2.5B (a few times in a row) instead and call it a day. No way you will pass both and not be 99% memory stability. Only thing you have to worry about is heat build up. Like say if your video card is blow hot air on the memory.

I always suggest doing your memory and CPU overclocks separately to find the maximum both can do before combining. This lets you find what each can do without adding more unneeded factors that lead you down the rabbit hole of trying to figure out what's wrong. I settled on 5.5P / 4.3 E because its a happen medium for daily use. If i'm pushing records, I'll go to 5.7 / 4.6, but that isn't good for daily use. Not without deliding the CPU or doing direct die (for me).

That. Ycruncher and tm5 works for me, never had any issues when i pass these 2.

As you said in the review, trefi is the most demanding timming, i never go above 65535 cause I don't have active cooling on the ram.

Curious, did you manage to boot these with cl30? My 7600c36 die fails miserably even at 7000c30

 

[ir_cow]

When you tested the kit and it hit 69c without a fan, using the XMP profile, did it actually produce errors at that temp?

I have it set to stop on any errors. It did at 71c and froze the system a few seconds later. I noted in the review, that the memory is basically touching with 2-DIMM motherboards. Using a 4-Slot that has spacing the temp was 10c lower. Not uncommon to have a system freeze if the temps get out of control. Though, its one of many things that can be the issue.

Curious, did you manage to boot these with cl30? My 7600c36 die fails miserably even at 7000c30

Yeah it works with 7800. Just copied G.SKILL 7800 memory kit settings (minus voltage). Don't forget to adjust tRFC too That one had me. Completely forgot about it at one point. tRDRD_sg and tRRD_sg are big ones too. Try something like 25 and lower if later. Some MBs seriously have issues going below 20. Idk why some have this set really low by default. I guess because it gives better perf, but at the cost of not being bootable.

 

[Minus Infinity]

Honestly this memory seems pointless. The gains are minimal even at 1080p, you get to run higher voltages and use more power and it performs worse that several slower DDR5 kits. Still impressed how well the DDR4 3600 CL14 did.

 

[GhostMotley]

I have it set to stop on any errors. It did at 71c and froze the system a few seconds later. I noted in the review, that the memory is basically touching with 2-DIMM motherboards. Using a 4-Slot that has spacing the temp was 10c lower. Not uncommon to have a system freeze if the temps get out of control. Though, its one of many things that can be the issue.

Thanks for the reply, did you record any 1% and 0.1% lows results?

 

[ir_cow]

Thanks for the reply, did you record any 1% and 0.1% lows results?

Some to make sure I'm not making stuff up But really waiting for when I switch to a RTX 4090 and add more games. Probably going to do 1080 low and get rid of medium in the future. Don't want to start what is going to take hundreds of hours just to do it again here soon.

Honestly this memory seems pointless. The gains are minimal even at 1080p, you get to run higher voltages and use more power and it performs worse that several slower DDR5 kits. Still impressed how well the DDR4 3600 CL14 did.

Gains are nearly none-existent if you only look at the average FPS. Being GPU bound gives the illusion these are all the same.

Did some brief COD testing to check the 1%, 5% and avg. You can see in the chart that the CPU can render even more frames the faster the memory is. Though need a video card to output those frames.

 

[dgianstefani]

Some to make sure I'm not making stuff up But really waiting for when I switch to a RTX 4090 and add more games. Probably going to do 1080 low and get rid of medium in the future. Don't want to start what is going to take hundreds of hours just to do it again here soon.


Gains are nearly none-existent if you only look at the average FPS. Being GPU bound gives the illusion these are all the same.

Did some brief COD testing to check the 1%, 5% and avg. You can see in the chart that the CPU can render even more frames the faster the memory is. Though need a video card to output those frames.

View attachment 282681

Need a better GPU than the 3080ti if you're going to really showcase the performance reserves of leading edge memory.

 

[ir_cow]

Need a better GPU than the 3080ti if you're going to really showcase the performance reserves of leading edge memory.

Give me your RTX 5090000 Ti

 

[dgianstefani]

I'll be tempted by the 4090ti on release.

 

[I_Want_Answers]

I keep seeing these RAM reviews and asking myself "who in the world is this memory for?"

You can get 32GB of the Z5's 6000MHz C34 version for around $160. Every single review I've seen of high-frequency RAM, including this one, demonstrates that gains are marginal at best. There's little gain in practical scenarios and even less in games. I'm willing to be educated if I'm wrong, but here's my reasoning:

Who gives a flying fart that it's better at memory read and write? It doesn't have any impact on render times either, which interests me personally since I dabble in Blender. The only productivity scenarios I've seen where faster RAM makes sense seem to be file (de)compression and Adobe Photoshop.

And gaming? The results are all over the place there. Some games might benefit more from the improved speed than others. I don't play it personally, but supposedly Tarkov scales well with better memory. For almost everything else, you see, what, a 3-5 fps improvement when the game is already pushing past 200? How does anyone benefit from this?

So again, who is this memory for? The only people I can think of are folks who have to have the best of the best for some reason. That excludes 95%+ of the PC-buying populace. I know that RAM is among the cheaper investments if you're building a high-end PC these days, so someone might not be bothered to spend another $150 when they're already $2,000+ in the hole. However, after removing everything else from the equation you have to ask yourself whether a 100%+ price increase is worth a 5% performance gain at best.

It's weird how people are up in arms when this happens in the GPU space but no one seems to mind when it's about RAM.

 

[dgianstefani]

I keep seeing these RAM reviews and asking myself "who in the world is this memory for?"

You can get 32GB of the Z5's 6000MHz C34 version for around $160. Every single review I've seen of high-frequency RAM, including this one, demonstrates that gains are marginal at best. There's little gain in practical scenarios and even less in games. I'm willing to be educated if I'm wrong, but here's my reasoning:

Who gives a flying fart that it's better at memory read and write? It doesn't have any impact on render times either, which interests me personally since I dabble in Blender. The only productivity scenarios I've seen where faster RAM makes sense seem to be file (de)compression and Adobe Photoshop.

And gaming? The results are all over the place there. Some games might benefit more from the improved speed than others. I don't play it personally, but supposedly Tarkov scales well with better memory. For almost everything else, you see, what, a 3-5 fps improvement when the game is already pushing past 200? How does anyone benefit from this?

So again, who is this memory for? The only people I can think of are folks who have to have the best of the best for some reason. That excludes 95%+ of the PC-buying populace. I know that RAM is among the cheaper investments if you're building a high-end PC these days, so someone might not be bothered to spend another $150 when they're already $2,000+ in the hole. However, after removing everything else from the equation you have to ask yourself whether a 100%+ price increase is worth a 5% performance gain at best.

It's weird how people are up in arms when this happens in the GPU space but no one seems to mind when it's about RAM.

It's more than 5%.

In this case the games are gpu limited since 13900k + 7200 ram is extremely fast cpu.

 

[Genten]

It's been more than a year and G.Skill still refuses to put a thermal pad on their PMICs? Shame on them

 

[Micko]

@ ir_cow : One question, soon I will be upgrading to a new system with 13700k / Noctua NH-D15s and Kingston 7200Mhz ram (not decided on the motherboard yet). For tweaking the ram to 7600+ i will most likely need a separate fan to blow on it. I have never used a separate fan for ram cooling before. Where would be the optimal position to place it ? Is placing it perpendicular over the ram best, or at an angle ? I'm not sure how much room will i have for setting it up. Could you please link a picture of how it is supposed to be mounted ideally ? Thanks.

 

[fevgatos]

@ ir_cow : One question, soon I will be upgrading to a new system with 13700k / Noctua NH-D15s and Kingston 7200Mhz ram (not decided on the motherboard yet). For tweaking the ram to 7600+ i will most likely need a separate fan to blow on it. I have never used a separate fan for ram cooling before. Where would be the optimal position to place it ? Is placing it perpendicular over the ram best, or at an angle ? I'm not sure how much room will i have for setting it up. Could you please link a picture of how it is supposed to be mounted ideally ? Thanks.

As long as voltage stays at 1.45 or below and you dont go crazy with trefi you dont really need a fan. Was running 7600c34 at 1.45v, no problem at all.

 

[Sabishii Hito]

I keep seeing these RAM reviews and asking myself "who in the world is this memory for?"
.

Not you? Tweakers and OCers love mem kits like this, and not to run at XMP either. Better binned ICs = better OC potential.

 

[ir_cow]

@Micko and @fevgatos gather around as we put the tin-foil hats on.

While I agree mostly with fevgatos that a direct fan is unnecessary for most scenarios. Don't go crazy on the voltage and have decent case-airflow and you will probably be alright. After all, most likely the highest temps recorded will come from these memory stress test programs. It takes a while for the heat to build up. Normal daily stuff isn't going to get hot enough to error out.

Now for a bit of conspiracy derived from observation during many many hours of memory overclocking.

1: tRFC and tREFi are highly temperature sensitive. These are both have to do with refreshing the memory data.

2: Memory in a 2-slot MB will not have enough spacing for passive / case airflow between them, leading to temperature related errors.

3: NVIDIA 30 / 40 FE cards blow hot air directly onto the memory and cause problems. When I reviewed the RTX 3080 (not for TPU), I had randomly crashes. After some poking around for a few days figured out my 32GB 3600 DDR4 kit was overheating after a few hours of game benchmarks. This was in a highly vented computer case mind you (not open bench). Contacted my NVIDIA PR rep who just told me I have bad "airflow" and its basically just me... Some motherboards DIMM slots line up better, but imagine pumping 300+ watts of heat over your ram for hours. My solution in a computer case has been to put a slip of cardboard between the two, if not a direct fan on the memory. I mentioned this in the video card review of course, but never gained traction. I guess it really is just a "me" problem...across multiple FE cards and computers...

 

[fevgatos]

@Micko and @fevgatos gather around as we put the tin-foil hats on.

While I agree mostly with fevgatos that a direct fan is unnecessary for most scenarios. Don't go crazy on the voltage and have decent case-airflow and you will probably be alright. After all, most likely the highest temps recorded will come from these memory stress test programs. It takes a while for the heat to build up. Normal daily stuff isn't going to get hot enough to error out.

Now for a bit of conspiracy derived from observation during many many hours of memory overclocking.

1: tRFC and tREFi are highly temperature sensitive. These are both have to do with refreshing the memory data.

2: Memory in a 2-slot MB will not have enough spacing for passive / case airflow between them, leading to temperature related errors.

3: NVIDIA 30 / 40 FE cards blow hot air directly onto the memory and cause problems. When I reviewed the RTX 3080 (not for TPU), I had randomly crashes. After some poking around for a few days figured out my 32GB 3600 DDR4 kit was overheating after a few hours of game benchmarks. This was in a highly vented computer case mind you (not open bench). Contacted my NVIDIA PR rep who just told me I have bad "airflow" and its basically just me... Some motherboards DIMM slots line up better, but imagine pumping 300+ watts of heat over your ram for hours. My solution in a computer case has been to put a slip of cardboard between the two, if not a direct fan on the memory. I mentioned this in the video card review of course, but never gained traction. I guess it really is just a "me" problem...across multiple FE cards and computers...

View attachment 282751

It's quite common actually to error out in games, ive seen it happen to a lot of people, even though they passed all kinds of stress tests. Poorly ventilated cases are pretty terrible for memory ocing. I remember during my 3090 custom bios days on an o11 dynamic, memory was hitting 68c during gaming, while it sat at 55 during memory stress tests

I have the fractal torrent, which is basically front to back airflow, so memory doesnt give a damn about what the gpu is doing. Im also going very conservative with timings, I never touch trefi above 65535. That's already pushing it for passive cooling ram.