Yes you can tell which manufacturer GSkill is using by the third digit from the end. In this case it ends in 20A, 2 means Hynix. 1 is Samsung, 3 is Micron, 4 is PSC (powerchip), 5 Nanya and 9 is for JHICC
Thanks for the video! I was asking you for more information on this board on reddit, and the video absolutely delivered. Very tempted to go pick one of these up and give 8000 MT/s a go.
I appreciate you uploading this video BZ. I have a 7800 CL36 XMP gskill kit on most am5 qvls that didn't work on my 7950x3d but work on the 9800x3d I got today. Was able to bump it up to 8000/2000 and use your settings and it seems totally stable so far after an hour of prime 95, vt3, and karhu each which is good enough for now as I'm still playing around with things.
About the sticker that ends in 820A, I'm pretty sure the 2 means Hynix. And a 1 there instead would mean Samsung, like 810B would be B-die. The third to last digit. Some Hynix sticks end in 821x, but the 2 is the important one.
Top notch video. I have the gskill royal neo 8000 kit (16x2) and it worked out of the box on my Asus crosshair hero x870e and 9800x3d with vsoc at 1.05. Used your secondary timings and it worked great.
Dude that perfect bec that exactly the incoming build I have coming I got thr royals and the hero and a 9800x3d did you have to change anything at all any recommendations for me I’d love it if you could discord me the setting in bios
I cant add manuel voltage in ram section ..for example if i type 1.35 v it ends up in 0.0001 ..same for soc and Ram fields..tried go enable high memory voltages and got A6 error on x870e Carbon ..
In my experience Karhu is something you want to run for a long time, like 18+ hours or so at least. Most of the time testmem5 + anta777 config catches out instability faster than Karhu and y-cruncher vt3 is also great.
Awesome video as always. Just wanted to point out. I have a 12900KS with a Z790 MSI carbon Wi-Fi motherboard and I bought A-die 7200mhz 32GB and I always heard those memory controllers were horrendous but mine have two profiles 6000mhz and 7200mhz and it booted just fine at 7200 XMP. Runs in windows just fine. I never did any real memory stress test with it but it’s never blue screened. No games have ever crashed so it’s either the A-Die has become really really good or it’s a combination of the MSI motherboard being stable. But I’m new to playing around with timings and all that and I should’ve had my windows corrupted and it hasn’t plus it was a dirty install so plus one for what you’re saying about the 8000 kit being stable
For anyone wondering. This settings are "ok" but they hammer my ram after 3min of tm5 to 52c+ at 1.5V w/o a fan 64k read 85k write 61.5k copy ~60ns latency Also they are way worse in bandwidth and latency then 8000/2200 +gdm disabled + synced phys even with worse/lose sub timings. 70.2k / 95k / 68k 59-60ns latency and way lower temps in 47C range after hours of tm5_1usmus_v3
This guy found a 12% improvement in MSFS 2020 at 1080P (which is why I am planning to try it) but it depends on the game (and probably other stuff; I'm not expecting such a difference on the 9800X3D). th-cam.com/video/bz_yA1YLCFY/w-d-xo.html
@@WillFuI This is generally not the case but in this particular scenario it is very much so. I would even go as far as to say this performs exactly like a tweaked ddr5 6000 setup. On a single ccd chip like the 9700x you will be maxing out the bandwidth of 2000FCLK at ddr5 6000 easy.
Would love to see you get hold of and tune with 9800X3D 1. Silicon Power XPOWER Storm RGB 64 GB (2 x 32 GB) DDR5-8000 CL38 2. Patriot Viper Xtreme 5 48 GB (2 x 24 GB) DDR5-8200 CL38
In 2:1 mode a single CCD has no chance of fully using this bandwidth. Infinity fabric is 32 bytes read + 16 bytes write per clock per link. So for a single CCD @ 2000MHz FCLK, the bandwidth limits are 64 GB/s read + 32 GB/s write. It is capable of read and write at the same time unlike RAM. Assuming your workload has exactly a 2:1 read:write ratio, the theoretical maximum memory bandwidth would be 96 GB/s. This just so happens to be the theoretical max bandwidth you can get from DDR5 6000. (128 bit interface * 6000 MT/s)
That is GMI (infinity fabric) limitation is definitelly from Ryzen 5000. AMD is probably using GMI-wide on Ryzen 7000 / 9000. If AIDA64 numbers can be trusted, Ryzen 7000 / 9000 variant of Infinity Fabric can do 32 bytes read + 48 bytes write per cycle.
My x870 Tomahawk was freezing on code 15 on boot after I updated the BIOS but didn't load defaults afterwards. Resetting the BIOS made the first boot work and then the next one would freeze on 15 again. Did this on two memory kids and carried on for about a week until I decided to load the damn defaults. Worked without a hitch since then.
@@juice7661 I use the 40 Gbps USB-C for my audio interface. On top of that I generally invest a bit more in my pc. I wouldn't say the HDV m.2 is better but I agree it's a solid choice for it's price for most people. If someone asks me what to buy and they have no particular preferences. I generally point them toward the B650M HDV m.2 as well.
@@tkgg Outside from very obscure stuff like audio and people who need atleast 20usb ports because their must juggle their data between all their usbs to keep them active, there is really no reason not to buy a B650M HDV m.2. Supports DDR5 11000+ and has been tested up to 400W so it can handle any Ryzen CPU out there and not throtle.
@@juice7661 If that's what you enjoy, that's what you get. I wanted the Tomahawk so I got the Tomahawk. It was within my budget and had the features I wanted.
@@tkgg Bros trying to say I got the asrock board because I'm broke🤣. The reason I got it its because like I've said, killer performance and the next tier up after it is the ASUS ROG CROSSHAIR X670E GENE which costs over 5 times as much and is basically discontinued everywhere. People on ebay want $683 for the board used, no thank you. I'm already at IMC limit using $100 board anyways.
I got the same motherboard running with Patriot Viper Xtreme 5 32GB (2X16GB) 8200MT/s CL38 kits. Can you try your 8000MT/s Patriot kits on this motherboard please? I am curious because mines are not stable with EXPO (profile 1) and I don't know what timings to tweak to make it stable.
Hello Buildzoid.. On x870e hero motherboard tphyrdl sync when you set ARdPtrInitVal P0 to 1. This can vary between different brand of motherboards. Imo tphyrdl sync when value set to 1. On some brand boards they sync when set to 0 or 2. I have also enabled robust training mode along with nitro mode 1 3 1 8x 8x Can you share insights on this..
I have a kit like this but 24GB (I haven't installed it yet). So hoping it "just works" and also hoping one of your other kits is like mine and you do a video on it :)
I cant enter vull numerical values for RAM ..for example cant type in 1.4V it always eds up as 0.0001 v .. tried enabling high voltage ..got A6 error ..on msi x870e Carbon with 9700x
Are there chances that Asrock x670e pg lightning will support a 8000 mhz overclock with ddr5? I know there's no 8000 mhz kits in qvls, but it's pretty much the same chipset as an x870. I'm asking this to learn if I should bother getting a 8000cl38 kit instead of the cheaper one and just go with 6200-6600 1:1. Of course assuming that I would get very lucky with a cpu controller. 9800x3d cpu. I'd appreciate if you'd help me with that.
Would love to see you get hold of and tune with 9700X and 9800X3D 1. Silicon Power XPOWER Storm RGB 64 GB (2 x 32 GB) DDR5-8000 CL38 2. Patriot Viper Xtreme 5 48 GB (2 x 24 GB) DDR5-8200 CL38
any word if any memory manufacturer is going to come out with single rank 32GB sicks of ram in the future? 32 gigs total aint enough for some games these days.
You can get single rank 24GB sticks that uses Hynix M-die which is very similar to A-die, that'll give you 48GB total Ram which is much better than 32GB. I also want 32GB single rank sticks (my dual rank sticks doesn't get me far enough). They're supposedly coming but I don't know when.
Thoughts on MSi's BIOS? I'm debating to settle on MSI or ASrock for my brand choice on X870E. Asus no longer consideration given your previous info of expensive but poor quality components and Gigabyte's BIOS lags
Can i ask you something, on a x670e steel legend, dominator platinun and 7800x3d i can do 8000mhz with cl 36 46, but my question is about the tertiary timmings, i cannot run TRDRDSC at default 1, isnt just working, i trwoes error at min 1 3, if i raise it to 8 it works for hours and i can guarantee its stable, what can be the cause of this?.
Well yes but not like this. Synced is when your ram speed is 3 times your FCLK. For example at 2000fclk your latency will be lowest at ddr5 6000. Only way you manage to beat a 2000fclk/6000 ram setup is if you run the 6000 with looser timings than the other profile, but the latency running 2000fclk 6000 ram at like cl24 maxed trefi and all the other good stuff is unmatchable outside that ratio even with 8000 cl32.
@@juice7661 It depends on what is synced and what isn't. There is FCLK, UCLK(memory controller) and MCLK (DDR Speed) A typical DDR5-8000 setup has FCLK and UCLK synced at 2000 and the MCLK is asynchronous by a rather lovely 1:2. A typical DDR5-6000 setup has UCLK and MCLK synced at 3000 and the FCLK is asynchronous. With FCLK of 2000 you can get a rather nice 2:3 asynchronity that has the a rather small latency benefit compared to other factors. But the latency in total get's even less with higher FCLKs. So a FCLK of 2200 with DDR5-6000 should have a lower latency despite not being 2:3 asynchronous. But you rather go for 6200 or 6400. Especially with 6400 you can go for a good 2:3 FCLK of 2133. People always just talk about 1:1 and 1:2 but that's mostly only UCLK:MCLK. I rather go for FCLK:UCLK:MCLK. By that a DDR5-6400 should be 2:3:3 and a DDR5-8000 should be 1:1:2.
@@juice7661This is just not true. 2000uclk (8000mt/s with 2:1) with 2000 fclk is synced. 3000 uclk does not sync with 2000 fclk (how would it). Afaik you have some benefits in how the data retrieval can be optimized at a 2:3 clock ratio, but that's about it.
Sure, but 2:3 does offer some latency benefits. Which are more work it I'd say than 2:1 especially with ddr5. The bandwidth you lose at 8000 by running your fclk at 2000 as opposed to 2200 is alot.
Sorry for the probably stupid question but I have 0 experience with AMD. 2200FCLK at ddr5 6600 could be a good solution? and if so should it go in 3 to 1 mode?
Depends on size of the kit as in the more ram u got in total the longer it takes, which is why 2x16 is the lowest which is around 18 hours to be 100% sure
when i try to memtest my 2x16 8000 a die kit at 8000 it doesnt even error, instead it crashes the whole pc to blackscreen and powers off the mainboard lol my 9800x3d is so dogshit that it can't even do more than 6200 in 1to1, for some reason it can go down to c26 36 36 tho which is absolutely useless tho
I have same kit, but X870E godlike. With memory try it its running fine at 8000CL36 with 63.4 latency in AIDA64. Is it possible to tweak yours kit to 6400MHz CL28? That would he ideal for 9000 series IMO
I'd love to see Corsair Dominator Titanium 96GB 6600MT/s CMP96GX5M2B6600C32 tested on this motherboard. The RAM kit has two profiles one for 6600MT/s and one for 6800MT/s. According to MSI both profiles are supported on 9000 series CPUs.
Are 48GB A-Die sticks a thing yet? Asking for me. Cuz I'm crazy. lol Question: On a scale of 1-10, how good do you think it is to use OCCT Memory Test as a stress test to look in to Memory Instability? I ask because I've used it a few times recently, and it seems like if the settings I'm testing are unstable it errors within the first 20-30 minutes, otherwise it just chugs along... but then again I've yet to run it for 24hrs lol
Of course they are. But there's still only 16/24gbit dies, so they're dual rank. Good luck getting anything above 7000mhz. Fine if you're going for 1:1 mode, though.
guys, i bought a 9800x3d and kingbank 6000 CL28 A die, and i cant get pass VT3 8 mins without errors on expo all default settings, what do i do, do i return the ram kit??
I had the same issue, for whatever reason the EXPO profile is 28-*34-34*-74, when it should be 28-35-35-74. Check the tRP and tRCD are 35 as they should be.
If you're on Asrock, your Vsoc might be a bit low. Asus and Gigabyte default to 1.25V. You could also try increasing the cas latency to 30-or increase VDD which improve the cas latency achievable.
@@marcgii so, i changed the trp and trcd to 35 like @IIHydraII said, and shit im running now 6200 cl28 at 1.22vsoc 2166 fclk, i tried 6400 at 1.25v it boots but wont even get to the first iteration of vt3 XD, but thanks for the comment
@@mnaufalalauddin even with that kind of B-die memory tune isn't nearly to what i just felt when i switched to AM5 R5 9600X 6000MT/s. It is a night and day difference.
My kits also worked out of the box at 8000/2000/2000 w/o changing, a thing. They also work at 8400/2100/2100 and also 8400/2200/2100 For PHYS synced on A and B channel : Try ARdPtrInitVal P0 = 0, 1, 2 (3 is default). But prolly need more voltage Why is GDM on ? SMU 98.78 is bugged increases latency by 7-9ns at least on b650e-e strix. Msi has "latency killer" but you have it disabled for some reason. Maybe this settings are "stable" coz of the increased latency on 1202A 98.78 SMU compared to 1202 (non-a and 98.77 smu) ? Wonder what latency on your settings vs my 8000/2200 gdm off (~57ns)
the "latency killer" description makes it sound like it just messes with CPU power management to stop the CPU from idling when running AIDA's latency test. I kinda doubt it helps in any real workloads especially when MSI warns about the "latency killer" hurting CPU performance(probably because it messes up the power management and boosting). GDM is on because turning it off sometimes doesn't work and doesn't really affect performance that much. So basically a convinience/laziness thing. Thanks for the tip about the tPHYRDLs
@@ActuallyHardcoreOverclocking RL perf agesa A vs non-A cyberpunk 720p 287fps vs 305 Dawntrail 720p laptop 47.5k vs 50k Wukong 720p 279fps vs 285fps So yeah i tested it multiple times. 1202A is broken. Idk if the latency killer is a gimick but im 100% sure 1202A is broken.
Lol this is a brand new one that copies ASUS/GIGABYTE. They used to have a convoluted pile of crap that was very love or hate. It always gave me a headache.
32gb sticks behave a bit differently in terms of overclockability as it puts more strain on your IMC. What are your voltages? (Vdd/q, soc, vddp, vddio) . Also, as a side note, generally speaking, and especially on a 3d vcache chips, the difference between a tightly tuned 6200 and 6400/8000 is marginal at best, so even if you end up with 6200 and you tune it properly, you end up with ~98% of peak performance available. Besides synthetic benchmark results dick swinging, it makes close to 0 difference in real world performance...
Yes you can tell which manufacturer GSkill is using by the third digit from the end. In this case it ends in 20A, 2 means Hynix. 1 is Samsung, 3 is Micron, 4 is PSC (powerchip), 5 Nanya and 9 is for JHICC
JHICC? So, dip the sticks in moonshine and it gains better speeds and timings? :V
Thanks for the video! I was asking you for more information on this board on reddit, and the video absolutely delivered.
Very tempted to go pick one of these up and give 8000 MT/s a go.
I appreciate you uploading this video BZ. I have a 7800 CL36 XMP gskill kit on most am5 qvls that didn't work on my 7950x3d but work on the 9800x3d I got today. Was able to bump it up to 8000/2000 and use your settings and it seems totally stable so far after an hour of prime 95, vt3, and karhu each which is good enough for now as I'm still playing around with things.
About the sticker that ends in 820A, I'm pretty sure the 2 means Hynix. And a 1 there instead would mean Samsung, like 810B would be B-die. The third to last digit. Some Hynix sticks end in 821x, but the 2 is the important one.
Top notch video. I have the gskill royal neo 8000 kit (16x2) and it worked out of the box on my Asus crosshair hero x870e and 9800x3d with vsoc at 1.05. Used your secondary timings and it worked great.
Dude that perfect bec that exactly the incoming build I have coming I got thr royals and the hero and a 9800x3d did you have to change anything at all any recommendations for me I’d love it if you could discord me the setting in bios
I cant add manuel voltage in ram section ..for example if i type 1.35 v it ends up in 0.0001 ..same for soc and Ram fields..tried go enable high memory voltages and got A6 error on x870e Carbon ..
Cheers boss, and thanks for going over some rationale behind each sub
First time I heard about Karhu Memory Test. Purchased a license and going to try it out later today. Thanks for all your awesome videos
y-cruncher vt3 is really all you need.
@@juice7661 L take
@@juice7661 In my experience, this is more for FCLK instability testing.
In my experience Karhu is something you want to run for a long time, like 18+ hours or so at least. Most of the time testmem5 + anta777 config catches out instability faster than Karhu and y-cruncher vt3 is also great.
@@IIHydraII L take because my recommendation does not require you to buy a license and its just as good if not better? I feel bad for you.
Awesome video as always. Just wanted to point out. I have a 12900KS with a Z790 MSI carbon Wi-Fi motherboard and I bought A-die 7200mhz 32GB and I always heard those memory controllers were horrendous but mine have two profiles 6000mhz and 7200mhz and it booted just fine at 7200 XMP. Runs in windows just fine. I never did any real memory stress test with it but it’s never blue screened. No games have ever crashed so it’s either the A-Die has become really really good or it’s a combination of the MSI motherboard being stable.
But I’m new to playing around with timings and all that and I should’ve had my windows corrupted and it hasn’t plus it was a dirty install so plus one for what you’re saying about the 8000 kit being stable
Intel yes... Likes higher frequencies, AMD not so much
For anyone wondering. This settings are "ok" but they hammer my ram after 3min of tm5 to 52c+ at 1.5V w/o a fan
64k read 85k write 61.5k copy ~60ns latency
Also they are way worse in bandwidth and latency then 8000/2200 +gdm disabled + synced phys even with worse/lose sub timings.
70.2k / 95k / 68k 59-60ns latency and way lower temps in 47C range after hours of tm5_1usmus_v3
What's the gaming fps gain from this compared to 6000 CL30/28?
Probably like 2%
This guy found a 12% improvement in MSFS 2020 at 1080P (which is why I am planning to try it) but it depends on the game (and probably other stuff; I'm not expecting such a difference on the 9800X3D). th-cam.com/video/bz_yA1YLCFY/w-d-xo.html
@@WillFuI This is generally not the case but in this particular scenario it is very much so. I would even go as far as to say this performs exactly like a tweaked ddr5 6000 setup.
On a single ccd chip like the 9700x you will be maxing out the bandwidth of 2000FCLK at ddr5 6000 easy.
Probably 0
Would love to see you get hold of and tune with 9800X3D
1. Silicon Power XPOWER Storm RGB 64 GB (2 x 32 GB) DDR5-8000 CL38
2. Patriot Viper Xtreme 5 48 GB (2 x 24 GB) DDR5-8200 CL38
In 2:1 mode a single CCD has no chance of fully using this bandwidth. Infinity fabric is 32 bytes read + 16 bytes write per clock per link. So for a single CCD @ 2000MHz FCLK, the bandwidth limits are 64 GB/s read + 32 GB/s write. It is capable of read and write at the same time unlike RAM. Assuming your workload has exactly a 2:1 read:write ratio, the theoretical maximum memory bandwidth would be 96 GB/s. This just so happens to be the theoretical max bandwidth you can get from DDR5 6000. (128 bit interface * 6000 MT/s)
That is GMI (infinity fabric) limitation is definitelly from Ryzen 5000. AMD is probably using GMI-wide on Ryzen 7000 / 9000. If AIDA64 numbers can be trusted, Ryzen 7000 / 9000 variant of Infinity Fabric can do 32 bytes read + 48 bytes write per cycle.
Finally! 💯
My x870 Tomahawk was freezing on code 15 on boot after I updated the BIOS but didn't load defaults afterwards. Resetting the BIOS made the first boot work and then the next one would freeze on 15 again. Did this on two memory kids and carried on for about a week until I decided to load the damn defaults. Worked without a hitch since then.
In the timings what should I change for 16GB M-die to be sure that it works?
Damn. Only 31 minutes? /s
I appreciate the content on this motherboard. Makes me rethink my m-die purchase.
What do you think is soo good about this board? The B650mhdv m.2 is better at a fraction of cost.
@@juice7661 I use the 40 Gbps USB-C for my audio interface. On top of that I generally invest a bit more in my pc.
I wouldn't say the HDV m.2 is better but I agree it's a solid choice for it's price for most people. If someone asks me what to buy and they have no particular preferences. I generally point them toward the B650M HDV m.2 as well.
@@tkgg Outside from very obscure stuff like audio and people who need atleast 20usb ports because their must juggle their data between all their usbs to keep them active, there is really no reason not to buy a B650M HDV m.2. Supports DDR5 11000+ and has been tested up to 400W so it can handle any Ryzen CPU out there and not throtle.
@@juice7661 If that's what you enjoy, that's what you get. I wanted the Tomahawk so I got the Tomahawk. It was within my budget and had the features I wanted.
@@tkgg Bros trying to say I got the asrock board because I'm broke🤣.
The reason I got it its because like I've said, killer performance and the next tier up after it is the ASUS ROG CROSSHAIR X670E GENE which costs over 5 times as much and is basically discontinued everywhere. People on ebay want $683 for the board used, no thank you. I'm already at IMC limit using $100 board anyways.
Great vid! But why are you in the year 2045?
I got the same motherboard running with Patriot Viper Xtreme 5 32GB (2X16GB) 8200MT/s CL38 kits. Can you try your 8000MT/s Patriot kits on this motherboard please? I am curious because mines are not stable with EXPO (profile 1) and I don't know what timings to tweak to make it stable.
nice.
Hello Buildzoid.. On x870e hero motherboard tphyrdl sync when you set ARdPtrInitVal P0 to 1. This can vary between different brand of motherboards. Imo tphyrdl sync when value set to 1. On some brand boards they sync when set to 0 or 2.
I have also enabled robust training mode along with nitro mode 1 3 1 8x 8x
Can you share insights on this..
I have a kit like this but 24GB (I haven't installed it yet). So hoping it "just works" and also hoping one of your other kits is like mine and you do a video on it :)
I cant enter vull numerical values for RAM ..for example cant type in 1.4V it always eds up as 0.0001 v .. tried enabling high voltage ..got A6 error ..on msi x870e Carbon with 9700x
Are there chances that Asrock x670e pg lightning will support a 8000 mhz overclock with ddr5? I know there's no 8000 mhz kits in qvls, but it's pretty much the same chipset as an x870. I'm asking this to learn if I should bother getting a 8000cl38 kit instead of the cheaper one and just go with 6200-6600 1:1. Of course assuming that I would get very lucky with a cpu controller. 9800x3d cpu. I'd appreciate if you'd help me with that.
On the server side of zen 5 they fixed the lower write bandwidth not sure if they fixed it for desktop
Hello Buildzoid.. On x870e hero motherboard tphyrdl sync when you set ARdPtrInitVal P0 to 1..
Can you share insights on this..
Tell me how dangerous is vddio 1.4 and higher for long-term use?
Would love to see you get hold of and tune with 9700X and 9800X3D
1. Silicon Power XPOWER Storm RGB 64 GB (2 x 32 GB) DDR5-8000 CL38
2. Patriot Viper Xtreme 5 48 GB (2 x 24 GB) DDR5-8200 CL38
any word if any memory manufacturer is going to come out with single rank 32GB sicks of ram in the future? 32 gigs total aint enough for some games these days.
You can get single rank 24GB sticks that uses Hynix M-die which is very similar to A-die, that'll give you 48GB total Ram which is much better than 32GB.
I also want 32GB single rank sticks (my dual rank sticks doesn't get me far enough). They're supposedly coming but I don't know when.
@@spamlucal yeah i was looking at 48GB kits buut i think i will just wait for single rank 32GB sticks
Thoughts on MSi's BIOS? I'm debating to settle on MSI or ASrock for my brand choice on X870E. Asus no longer consideration given your previous info of expensive but poor quality components and Gigabyte's BIOS lags
Could you try Memtest with FCLK 2200 if it speeds up 10% (keep memory at 8000MT/s)
Lower latency = Good for games
More bandwidth = Good for creator programs (encoding, compression, video editing)
Is that the jist?
Would a 2x24gb DDR5 8000 kit work with these new AMD boards with the 9800X3D? This kit is really for Intel and does not have Expo.
Need to know this as well
Can i ask you something, on a x670e steel legend, dominator platinun and 7800x3d i can do 8000mhz with cl 36 46, but my question is about the tertiary timmings, i cannot run TRDRDSC at default 1, isnt just working, i trwoes error at min 1 3, if i raise it to 8 it works for hours and i can guarantee its stable, what can be the cause of this?.
yo buildzoid today there is ram stream ? , i got hynix a die last week i want you to rate them -_-
Is there still value in running DDR5 8000 with synced fclk for gaming on 9800X3D, or is the latency advantage diminished by the v-cache?
Well yes but not like this. Synced is when your ram speed is 3 times your FCLK. For example at 2000fclk your latency will be lowest at ddr5 6000. Only way you manage to beat a 2000fclk/6000 ram setup is if you run the 6000 with looser timings than the other profile, but the latency running 2000fclk 6000 ram at like cl24 maxed trefi and all the other good stuff is unmatchable outside that ratio even with 8000 cl32.
@@juice7661 It depends on what is synced and what isn't. There is FCLK, UCLK(memory controller) and MCLK (DDR Speed)
A typical DDR5-8000 setup has FCLK and UCLK synced at 2000 and the MCLK is asynchronous by a rather lovely 1:2.
A typical DDR5-6000 setup has UCLK and MCLK synced at 3000 and the FCLK is asynchronous. With FCLK of 2000 you can get a rather nice 2:3 asynchronity that has the a rather small latency benefit compared to other factors.
But the latency in total get's even less with higher FCLKs. So a FCLK of 2200 with DDR5-6000 should have a lower latency despite not being 2:3 asynchronous. But you rather go for 6200 or 6400. Especially with 6400 you can go for a good 2:3 FCLK of 2133.
People always just talk about 1:1 and 1:2 but that's mostly only UCLK:MCLK. I rather go for FCLK:UCLK:MCLK. By that a DDR5-6400 should be 2:3:3 and a DDR5-8000 should be 1:1:2.
@@juice7661This is just not true. 2000uclk (8000mt/s with 2:1) with 2000 fclk is synced. 3000 uclk does not sync with 2000 fclk (how would it). Afaik you have some benefits in how the data retrieval can be optimized at a 2:3 clock ratio, but that's about it.
Sure, but 2:3 does offer some latency benefits. Which are more work it I'd say than 2:1 especially with ddr5. The bandwidth you lose at 8000 by running your fclk at 2000 as opposed to 2200 is alot.
Year 2045 in the bios - are you a time traveler?
Sorry for the probably stupid question but I have 0 experience with AMD. 2200FCLK at ddr5 6600 could be a good solution? and if so should it go in 3 to 1 mode?
how long do you actually need to overclock a new kit to the max and daily stable?
Depends on size of the kit as in the more ram u got in total the longer it takes, which is why 2x16 is the lowest which is around 18 hours to be 100% sure
Wouldn’t peak performance be at like 8800 so your fclk and mclk is 2200? But that sounds near impressive
Is MSI the only one allowing trcdwr changes?
when i try to memtest my 2x16 8000 a die kit at 8000 it doesnt even error, instead it crashes the whole pc to blackscreen and powers off the mainboard lol
my 9800x3d is so dogshit that it can't even do more than 6200 in 1to1, for some reason it can go down to c26 36 36 tho which is absolutely useless tho
I have same kit, but X870E godlike. With memory try it its running fine at 8000CL36 with 63.4 latency in AIDA64. Is it possible to tweak yours kit to 6400MHz CL28? That would he ideal for 9000 series IMO
can you set VPP to 1.85V safely on any motherboard/memory combination?
yes but it usually doesn't do anything.
I'd love to see Corsair Dominator Titanium 96GB 6600MT/s CMP96GX5M2B6600C32 tested on this motherboard. The RAM kit has two profiles one for 6600MT/s and one for 6800MT/s. According to MSI both profiles are supported on 9000 series CPUs.
That would be really slow on ryzen
@@WillFuI Unfortunately not too many faster options for 96GB kits
@ I would just bring it down to like 6400 if your cpu can do 1:1
test it on the HDV please
Are 48GB A-Die sticks a thing yet? Asking for me. Cuz I'm crazy. lol
Question: On a scale of 1-10, how good do you think it is to use OCCT Memory Test as a stress test to look in to Memory Instability? I ask because I've used it a few times recently, and it seems like if the settings I'm testing are unstable it errors within the first 20-30 minutes, otherwise it just chugs along... but then again I've yet to run it for 24hrs lol
KLEVV CRAS V RGB DDR5 48GB (2x24GB) 6000MHz CL30 A-DIE 1.35V Gaming Desktop Ram Memory SK Hynix Chip XMP 3.0 / AMD Expo Ready - Black (KD5KGUD80-60A300G)
Of course they are. But there's still only 16/24gbit dies, so they're dual rank. Good luck getting anything above 7000mhz. Fine if you're going for 1:1 mode, though.
guys, i bought a 9800x3d and kingbank 6000 CL28 A die, and i cant get pass VT3 8 mins without errors on expo all default settings, what do i do, do i return the ram kit??
I had the same issue, for whatever reason the EXPO profile is 28-*34-34*-74, when it should be 28-35-35-74. Check the tRP and tRCD are 35 as they should be.
@@IIHydraII oooh ok, thanks!
If you're on Asrock, your Vsoc might be a bit low. Asus and Gigabyte default to 1.25V. You could also try increasing the cas latency to 30-or increase VDD which improve the cas latency achievable.
@@marcgii so, i changed the trp and trcd to 35 like @IIHydraII said, and shit im running now 6200 cl28 at 1.22vsoc 2166 fclk, i tried 6400 at 1.25v it boots but wont even get to the first iteration of vt3 XD, but thanks for the comment
my Samsung B-Die TridentZ RGB 3200CL 14 feels nothing to DDR5 Clock Speed nowadays..
you are on XMP man 🤣 I've ran 3666MT/s CL14-8-13-11-240 TRFC tuned sub and tertiaries timings with an R5 2600x on an asus x570-p.
@Taikon_n damn
@@mnaufalalauddin even with that kind of B-die memory tune isn't nearly to what i just felt when i switched to AM5 R5 9600X 6000MT/s.
It is a night and day difference.
@@Taikon_n thats what i'm literally saying, DDR5 transfer speed are crazy nowadays
@@Taikon_n though I didnt tune my B-Die to the peak of it's capabilities 😅
would have loved to see 6400 range and low cl
already did that
Just worked.
i want to see 4dims overclock setup
I wish motherboards didn't come with 4 memory slots.
My kits also worked out of the box at 8000/2000/2000 w/o changing, a thing. They also work at 8400/2100/2100 and also 8400/2200/2100
For PHYS synced on A and B channel : Try ARdPtrInitVal P0 = 0, 1, 2 (3 is default). But prolly need more voltage
Why is GDM on ?
SMU 98.78 is bugged increases latency by 7-9ns at least on b650e-e strix. Msi has "latency killer" but you have it disabled for some reason.
Maybe this settings are "stable" coz of the increased latency on 1202A 98.78 SMU compared to 1202 (non-a and 98.77 smu) ?
Wonder what latency on your settings vs my 8000/2200 gdm off (~57ns)
the "latency killer" description makes it sound like it just messes with CPU power management to stop the CPU from idling when running AIDA's latency test. I kinda doubt it helps in any real workloads especially when MSI warns about the "latency killer" hurting CPU performance(probably because it messes up the power management and boosting).
GDM is on because turning it off sometimes doesn't work and doesn't really affect performance that much. So basically a convinience/laziness thing.
Thanks for the tip about the tPHYRDLs
@@ActuallyHardcoreOverclocking
RL perf agesa A vs non-A
cyberpunk 720p 287fps vs 305
Dawntrail 720p laptop 47.5k vs 50k
Wukong 720p 279fps vs 285fps
So yeah i tested it multiple times.
1202A is broken. Idk if the latency killer is a gimick but im 100% sure 1202A is broken.
Ha, G.Skill thinks I will buy this kit because of this video, but I already have it! That will show em! Iam one step ahead again, muhahaha!
/me who's still running ddr4 and patiently waiting for ddr6 with 12800 MT/s: hmm... interesting 🤔
So 31 min video in 3.1 seconds.. it works on xmp…
MSI has a very nice looking BIOS, I must admit.
Lol this is a brand new one that copies ASUS/GIGABYTE. They used to have a convoluted pile of crap that was very love or hate. It always gave me a headache.
This video is great but makes me sad.
My 9800x3d with a STRIX-E can't boot at 8000 using 2x32gb A-die.
6400 is also unstable AF.
Stuck on 6200cl28. 😢
32gb sticks behave a bit differently in terms of overclockability as it puts more strain on your IMC. What are your voltages? (Vdd/q, soc, vddp, vddio) .
Also, as a side note, generally speaking, and especially on a 3d vcache chips, the difference between a tightly tuned 6200 and 6400/8000 is marginal at best, so even if you end up with 6200 and you tune it properly, you end up with ~98% of peak performance available. Besides synthetic benchmark results dick swinging, it makes close to 0 difference in real world performance...
Where z890 content?
Not happening.
@@ActuallyHardcoreOverclocking Omegalol
nice.