intel 13/14th gen fast RAM experiences: >7,000 MT/s is XMP stable!

[Dr. Dro]

memtest5 was easy, what's next? This is the hottest I've ever seen my ram get btw. It crested 60 C and I turned my case fans on full blast for the rest.

View attachment 325387

Oh and an addendum to my earlier post, regarding TM5: I've had the Extreme1 preset error out 1h30 into testing. I tracked it down to my tWRRD_sg and _dg settings. Tertiaries are hard to get right. But if you're interested in a reference, I've got this running at 6400 C30:

I'm already stable until proven otherwise

I see. To me, I got it to manifest the "hidden" instabilities pretty easily with some games. For example, if there's any memory instability at all, Genshin Impact loves to simply abruptly close with a silent crash to desktop. Found that out because somehow for some reason my motherboard fumbled the VDDQ/VDD2 voltage settings last time I changed something in the BIOS, and the CPU is ultra sensitive to those. My settings might not look at all impressive (especially considering I use Z690), but it's triple verified and I've followed all of the known DDR5 timing formulae I've come to learn over time.

 

[fusseli]

And to be honest I trust Intel LGA1700 110% more than I trust aftermarket contact frames. Nobody knows the level of halfassery, or even whole-assery, that went into the mechanical design of contact frames versus the cpu heat spreader and socket of LGA 1700. I guarantee you Intel has a "simple" model such as Solidworks for torsional, stress, and strain model of the socket and all OEMs have to meet specifics specs to assure performance. Yes, they can account for microns of bending in the cpu heat spreader and the socket pressure, as any competent mechanical engineers can do let alone those at a multi billion dollar global company. What's a $10 contact frame got behind it besides some CNC milling of aluminum and a few screws you can wrench down as tight as you want? We don't actually know, other than the minority that seems to have found a ram stability issue with cpu cooler / AIO / contact frame pressure.

Oh and an addendum to my earlier post, regarding TM5: I've had the Extreme1 preset error out 1h30 into testing. I tracked it down to my tWRRD_sg and _dg settings. Tertiaries are hard to get right. But if you're interested in a reference, I've got this running at 6400 C30:

View attachment 325388



I see. To me, I got it to manifest the "hidden" instabilities pretty easily with some games. For example, if there's any memory instability at all, Genshin Impact loves to simply abruptly close with a silent crash to desktop. Found that out because somehow for some reason my motherboard fumbled the VDDQ/VDD2 voltage settings last time I changed something in the BIOS, and the CPU is ultra sensitive to those. My settings might not look at all impressive (especially considering I use Z690), but it's triple verified and I've followed all of the known DDR5 timing formulae I've come to learn over time.

In a general sense I concur; any abundance of app crashes or system hangs/resets indicate underlying instability. One must vet out cpu, ram, gpu, etc. one by one to isolate root cause. I didn't have any crashes on 7200 mt/s for about a month. I just installed 7800 mt/s and it's so far so good, my intent is to intersperse reasonable stress tastes and real life use to determine a conclusion.

 

[ir_cow]

There is very strong evidence that Intel screwed up the 1700 socket design. People like der8auer has done laser scans and you can see the uneven pressure the latch side has with the CPU installed. Can't even do LN2 without a bracket because of this unevenness.

Anyways congrats on 3 passes. That means your on the right track. Lazy people like me wouldn't do 25 loops until I dialed in my voltages and sub-timings.

If your bored enough, start to adjust the SA,TX and VDD2 voltages. The lower you go, the better for the long term. As for Ram. Yeah they get hot like I said. I put a fan on my work computer. It's sitting on the video card. Direct air blast

 

[Upgrayedd]

And to be honest I trust Intel LGA1700 110% more than I trust aftermarket contact frames. Nobody knows the level of halfassery, or even whole-assery, that went into the mechanical design of contact frames versus the cpu heat spreader and socket of LGA 1700. I guarantee you Intel has a "simple" model such as Solidworks for torsional, stress, and strain model of the socket and all OEMs have to meet specifics specs to assure performance. Yes, they can account for microns of bending in the cpu heat spreader and the socket pressure, as any competent mechanical engineers can do let alone those at a multi billion dollar global company. What's a $10 contact frame got behind it besides some CNC milling of aluminum and a few screws you can wrench down as tight as you want? We don't actually know, other than the minority that seems to have found a ram stability issue with cpu cooler / AIO / contact frame pressure.


In a general sense I concur; any abundance of app crashes or system hangs/resets indicate underlying instability. One must vet out cpu, ram, gpu, etc. one by one to isolate root cause. I didn't have any crashes on 7200 mt/s for about a month. I just installed 7800 mt/s and it's so far so good, my intent is to intersperse reasonable stress tastes and real life use to determine a conclusion.

I'm still not sure about the large differences in our reported cache speeds. Everything about yours was reporting slower. You did finish the test about 5 min faster than me though.

Edit: do you know what your board scores you get for silicon prediction? Mine says 75

 

[tabascosauz]

@fusseli what does your CPU-Z Memory tab page look like? Might shed some light as to Gear 2 vs 4

Gear 4 is a lot like half UCLK speed on AM5. All show and no go

 

[ir_cow]

@fusseli what does your CPU-Z Memory tab page look like? Might shed some light as to Gear 2 vs 4

Gear 4 is a lot like half UCLK speed on AM5. All show and no go

Has anyone ever gotten Gear 4 to work? I don't keep tabs on the latest developments. So far I just get a boot loop if I switch to Gear 4.

Edit: do you know what your board scores you get for silicon prediction? Mine says 75

Gigabyte (Biscut score) and ASUS SP score are not 100% comparable. I was told by Gigabyte that 103 is like a WR sample. ASUS need to be in the 107-115 range.

 

[tabascosauz]

Has anyone ever gotten Gear 4 to work? I don't keep tabs on the latest developments. So far I just get a boot loop if I switch to Gear 4.

Gigabyte (Biscut score) and ASUS SP score are not 100% comparable. I was told by Gigabyte that 103 is like a WR sample. ASUS need to be in the 107-115 range.

Think I've seen a few Gear 4 validations but they are super scarce. Probably goes without saying nobody runs them on the regular.

I'm just following up on a previous reply about it. CPU-Z would clear up any doubts. 14th gen Gear 2 stability at 7800 wouldn't surprise me terribly; 13th gen did what, 7200 pretty easy?

 

[Upgrayedd]

Think I've seen a few Gear 4 validations but they are super scarce. Probably goes without saying nobody runs them on the regular.

I'm just following up on a previous reply about it. CPU-Z would clear up any doubts. 14th gen Gear 2 stability at 7800 wouldn't surprise me terribly; 13th gen did what, 7200 pretty easy?

I just ran a 7200 xmp profile tuned down to 6800 on a 14700K with MCE off and still had an error. Might just return everything and start fresh. Last thing I'll try is the stock ILM and if it still errors its all going back.

 

[fusseli]

I just ran a 7200 xmp profile tuned down to 6800 on a 14700K with MCE off and still had an error. Might just return everything and start fresh. Last thing I'll try is the stock ILM and if it still errors its all going back.

Interesting. The z790 Apex should have been a pretty decent board.


Here are my cpuz tabs. In bios I have Gigabyte set to "enhance settings," and I have energy efficient turbo disabled, otherwise 100% stock on the 14700k and XMP settings. I suppose I could boot back into memtest with different bios enhancement settings to check my cache speeds. It's possible Gigabyte is messing with them somehow.

Has anyone ever gotten Gear 4 to work? I don't keep tabs on the latest developments. So far I just get a boot loop if I switch to Gear 4.


Gigabyte (Biscut score) and ASUS SP score are not 100% comparable. I was told by Gigabyte that 103 is like a WR sample. ASUS need to be in the 107-115 range.

I can't find jack on cpu bisuits, mine is 92.272 or something. On the contrary over on the ROG forums there is plenty of info on where people are at with asus SP/IMC scores.


----------------------------------------------------------------


Here's an overnight run of XTU Memory Stress Test. I think it beats the max dram temp from your grueling TestMem5

Curiously my 14700k is power throttling with stock settings. This and the cache speed differences compared to the 14700k on the z790 Apex is fairly intriguing.

 

[fusseli]

Overnight run of OCCT memtest, no errors. I might try OCing this ram farther than XMP!

I just ran a 7200 xmp profile tuned down to 6800 on a 14700K with MCE off and still had an error. Might just return everything and start fresh. Last thing I'll try is the stock ILM and if it still errors its all going back.

Any status updates?

I notice AIDA64 gives cache speeds too, here's what I am getting on 7800-c3. I don't think memtest86 is giving good figures compared to what we had at:
mine: L1 459 GB/s : L2 87 GB/s : L3 38 GB/s : mem 24 GB/s
yours: L1 700 GB/s : L2 132 GB/s : L3 46 GB/s : mem 26.8 GB/s

 

[tabascosauz]

Is latency a bit high? Swear I've seen 6200-6400CL30 results in the low 50ns range

 

[fusseli]

Is latency a bit high? Swear I've seen 6200-6400CL30 results in the low 50ns range

I don't know, I'm looking around for aida64 scores to compare to and so far my speeds look pretty good/monstrous such as compared to this 7800x3d guy on 6200-CL28 https://www.techpowerup.com/forums/...y-benchmark-here.186338/page-109#post-5157104

 

[tabascosauz]

I don't know, I'm looking around for aida64 scores to compare to and so far my speeds look pretty good/monstrous such as compared to this 7800x3d guy https://www.techpowerup.com/forums/...y-benchmark-here.186338/page-109#post-5157104

Compare intel to intel, Ryzen is not comparable, although that one does look to be a very good result for 7800X3D

It's hard to tell, I still have to keep going back through the OCN thread because everybody is experimenting in the 8000-9000 region now with Hynix. I seen a few 8000-8200CL38/40 results in at 60ns or just below. One 6400CL30 I have saved on my phone at 54ns. Think I saw 8400CL38 at 52ns.

*Official* Intel DDR5 OC and 24/7 daily Memory Stability...

www.overclock.net

 

[fusseli]

Compare intel to intel, Ryzen is not comparable, although that one does look to be a very good result for 7800X3D

It's hard to tell, I still have to keep going back through the OCN thread because everybody is experimenting in the 8000-9000 region now with Hynix. I seen a few 8000-8200CL38/40 results in at 60ns or just below. One 6400CL30 I have saved on my phone at 54ns. Think I saw 8400CL38 at 52ns.

*Official* Intel DDR5 OC and 24/7 daily Memory Stability...

www.overclock.net

I didn't close all background apps, I just ran it casually looking for basis of comparison.

 

[tabascosauz]

I didn't close all background apps, I just ran it casually looking for basis of comparison.

Fair enough. You can double click individual boxes in AIDA to get it just to run that number. ie. run latency repeatedly to see what you can get. Point still stands, comparisons to AM5 can't be properly made, and I don't think your background apps on a new build should have that much of an impact.

 

[fusseli]

Fair enough. You can double click individual boxes in AIDA to get it just to run that number. ie. run latency repeatedly to see what you can get. Point still stands, comparisons to AM5 can't be properly made, and I don't think your background apps on a new build should have that much of an impact.

Sure but it's a benchmark, is it not? On what basis "can't they be compared"?

 

[tabascosauz]

Sure but it's a benchmark, is it not? On what basis "can't they be compared"?

Literally different platform? idk the exact difference for LGA1700 vs AM5, but LGA1155/1200 vs AM4 was on the order of 10-20ns penalty for AMD depending on exact CPU family in the comparison

If you don't compare like for like, how do you get a sense of whether you are getting expected performance for 13th/14th gen?

 

[Dr. Dro]

Yeah those numbers look pretty bad for a DDR5-7800 setup. Here's mine:

 

[ir_cow]

This is no adjustments XMP. Change the tRFC and tREFI it will go down.

Also double clicking the box will retest just that one. AiDA64 likes to go 2-3ns higher on the first run. Not always though.

 

[fusseli]

Literally different platform? idk the exact difference for LGA1700 vs AM5, but LGA1155/1200 vs AM4 was on the order of 10-20ns penalty for AMD depending on exact CPU family in the comparison

If you don't compare like for like, how do you get a sense of whether you are getting expected performance for 13th/14th gen?

Sure but that doesn't mean they can't be compared, metric for metric of course they can be compared.

Yeah those numbers look pretty bad for a DDR5-7800 setup. Here's mine:

View attachment 325946

Huh? Besides latency what are you talking about besides your clocked higher cpu with faster cache and your slower memory than mine?

 

[ir_cow]

This is 8000 tweaked to nearly the max.

 

[tabascosauz]

This is 8000 tweaked to nearly the max.

The green XMP result is more along the lines of what I was expecting for a high end Hynix kit.

Sure but that doesn't mean they can't be compared, metric for metric of course they can be compared.

Of course you can compare them, but it doesn't tell you any useful info whatsoever mem-OC wise lol.........except I guess feeling some strange sense of superiority? AM5 might have a specific latency penalty in AIDA, but it has its own range of good/bad memory performance to measure against, just like 13th/14th gen. Would you use 7800X3D Cinebench results to tell you whether your 14700K results are good or not?

 

[fusseli]

The green XMP result is more along the lines of what I was expecting for a high end Hynix kit.



Of course you can compare them, but it doesn't tell you any useful info whatsoever mem-OC wise lol.........except I guess feeling some strange sense of superiority? AM5 might have a specific latency penalty in AIDA, but it has its own range of good/bad memory performance to measure against, just like 13th/14th gen. Would you use 7800X3D Cinebench results to tell you whether your 14700K results are good or not?

Thank you, that makes a lot more sense, I also wasn't realizing mem latency was mostly what you're referring to. I don't have time to read all of @ir_cow memory reviews

I’m starting to mess with some timing tweaks. Gear 4 is a no boot, so I’m leaving it on Auto. Do t see the point of forcing Gear 2.

This refresh Z790 Gigabyte has a ton of new timing enhancements compared to my old Z490 Ultra. So far “high bandwidth” and “low latency” had to be disabled for me to get XMP stable in all aforementioned stress tests. I’m now trying to see if I can get a “low latency” and “memory enhancement” stable and get eyes on what changes those are actually doing.

This is no adjustments XMP. Change the tRFC and tREFI it will go down.

Also double clicking the box will retest just that one. AiDA64 likes to go 2-3ns higher on the first run. Not always though.

View attachment 325948

My tRFC is 624 and tREFI is 7605 with XMP on and all else disabled.

 

[ir_cow]

... I don't have time to read all of @ir_cow memory reviews

No one does and you aren't missing much.

This refresh Z790 Gigabyte has a ton of new timing enhancements compared to my old Z490 Ultra. So far “high bandwidth” and “low latency” had to be disabled for me to get XMP stable in all aforementioned stress tests. I’m now trying to see if I can get a “low latency” and “memory enhancement” stable and get eyes on what changes those are actually doing.

Contrary to the name, "High Bandwidth" isn't to be enabled with high speed memory. What it does is change the tRFC and tREFi values, among a few others. The higher you go in freq, the less likely those auto MB adjustments will work. tREFI 5k-8K is normal defaults. 65K is tweaked (put a fan on me) and 255K is just stupid and pointless in my opinion.

tRFC is the cycle time to refresh the banks and tREFi is the "time" between refresh (in cycles). Lower tRFC means less time refreshing and higher tREFI means more time between refreshing. To low on the first one and the banks won't be done in time causing data loss. Go to high on the second and the banks drain of power and data is lost. Temperatures play a big roll in this area.

While tRFC is going all command are stopped and nothing can be accessed. DDR5 introduction single bank refresh and that does... stuff lol. I think the rest can still get commands if only one bank is refreshed... don't quote me on that one. Long day at work.

 

[Upgrayedd]

Overnight run of OCCT memtest, no errors. I might try OCing this ram farther than XMP!

View attachment 325863


Any status updates?

I notice AIDA64 gives cache speeds too, here's what I am getting on 7800-c3. I don't think memtest86 is giving good figures compared to what we had at:
mine: L1 459 GB/s : L2 87 GB/s : L3 38 GB/s : mem 24 GB/s
yours: L1 700 GB/s : L2 132 GB/s : L3 46 GB/s : mem 26.8 GB/s
View attachment 325944

I think my board was bad. I bought an open box Apex Encore at microcenter and have some 8000 2x24 patriot kit coming in the mail. This new board ran the current 7200 2x16 memory fine. I could only clock the memory to 7400 with 7200xmp set. We'll see what happens with the new kit.
Board says MC score is 80 but that doesn't mean anything. I'll run aida64 here after a while and see what it says.