Buildzoid Hey guys here Buildzoid guys Hey here Buildzoid here Hey guys Buildzoid Hey here guys Buildzoid guys here hay Buildzoid Hey guys here Hey buildzoid here guys Hey here buildzoid guys Hey guys buildzoid here Hey buildzoid guys here Hey here guys buildzoid Hey guys here buildzoid Guys buildzoid Hey here Guys Hey buildzoid here Guys here buildzoid Hey Guys buildzoid here Hey Guys Hey here buildzoid Guys here Hey buildzoid Here guys Hey buildzoid Here guys buildzoid Hey Here buildzoid Hey guys Here buildzoid guys Hey Here Hey buildzoid guys Here Hey guys buildzoid
Would it be worthwhile to revisit the “easy memory timings for Hynix DDR5 with Ryzen 7000” video now with the new bios? Dummies like me hope we can get updated easy settings with even better performance. 😅 pretty please 🙏
What do you get in aida64 with those settings. I can't seem to get past 65ns but I also don't have power down disabled because I don't want to wait for slow ass memory training every boot and you have to enable power down with memory context restore enabled.
Thank you for sharing DDR5 overclocking experience! Following the settings in the video, I successfully increased the memory frequency to 8000MT/s and passed 30 rounds of VT-3 testing. The spec used is as follows: CPU: Ryzen 9 7950x3d MB: GIGABYTE AORUS X670E MASTER(BIOS version:F30,AMD AGESA 1.1.7.0) MEM: Hynix M-die 2x24GB2 kit
Год назад+14
BuildZoid, an ideia that only you could pull up: how much does DDR5 speed and timings matter? As you are usually using your 7900x, you could lock the cpu clock and run some benchmarks of your choice to measure the ram impact (Geekbench, y-cruncher, Py Prime 32m, you name it..). You should run the benchmarks on each RAM that you tune up and build a database. Your videos are amazing and I think it would be really nice to see that.
Hey BZ. Just a data point since you mentioned 1.2vsoc being best for your chips - mine is definitely lower, peaking at around 1100 or 1125. Setting 1100 vs 1200 is night and day on some configs with no throttling on 1100 vs 30% performance loss at 1200. I currently run DDR5-8000 with a 4dimm board (x670e carbon) and 2x16GB of Hynix 16a, which is probably considerably worse than what you have - makes me wonder if 8400 is doable. Sadly there is no infinity fabric ratio for uclk:fclk sync at 8200 mem, but maybe we can get that in a future AGESA. I'm running that right now on 1.125v soc, but it runs fine on 1.10 so this is just solid safety margin (1.075 shows some performance loss). Part of the reason i like the synced config is not having to run fclk up against the wall and deal with all of the funky instability that can happen because your room temperature changed or the planets are alligned wrongly - 2000 fclk just works, no questions asked, 100% of the time at 1.1vsoc or 1.25vsoc or whatever you want. Likewise the IMC is very happy to run at 2000mhz rather than driving it to 3200 plus in gear 1. I also use a VDDP of 1050 but have not properly explored lower values yet, but also suspect too high causing failure. I use VDDG 850 but that's completely untuned - substantially lower is probably optimal. My CPU will do 2167 fclk on 850 vddg, but it's a bit peculiar about it with only 2133 being super safe. This is generally slower than 2000 with uclk:fclk sync. +5% IF isn't enough to overcome desync penalty. Between 5-10% is the crossover point most of the time, with 10% being better on almost everything. Oh, and my stable primaries are 36-46-36 at 1.4/1.4/1.38 on the mem. 10 whole ticks
I think AMD"s memory controller is actually a bit better tuned for 16Gb A-die ATM. The LiveMixer and HDV both kinda run 8000 on A-die but really struggle with this 2x24GB kit.
Another data point I can add is that 8000 1:2 really likes low vsoc (1:2 in general), 6000-6400 1:1 likes it higher. There are people running 8000 at ~0.95vsoc. I think it's mainly because you're running the mclk really low, so there's no need for higher soc voltages.
I have a R9 7950X3D and I run into BSOD when running SoC to 1.200v but I'm running it currently with 1.235v and testing the Corsair Vengance 6400MT/s 2x48GB , Hynix M-Die and with the help of this channel I seem to be able to run them in 30-36-30-30 @ 6200MHz (MCLK 3100, UCLK 3100, FCLK 2100). Not sure if it can do any tighter timings. Still playing around still, but the SoC voltage really make an impact to the CPU idle temps, and I'm all about getting best performance at the best temps :).
I started laughing so hard at the 4023 mhz that I had to stop the video ... 40 x 99.9 = 4023 LOL . JUST THE USUAL AMD STUFF .. great job though and I love your channel BZ!
on the 7000 cpus, the infinity fabric is running dual 32yte/cycle interconnects. That doubles the total system memory bandwidth available to the cores. A single 2000mhz IF interconnect is limited to 64GB/s of total bandwidth. The dual interconnects creates a potential maximum of 128GB/s to the cores
@@BusAlexey No, Since threadripper 3000 introduced the dual interconnects to each CCD, the 5000 and 7000 series have all used dual interconnects to each ccd. That is why the gaming performance improved with the 5000 series chips. Memory read traffic and GPU traffic no longer needed to fight to use the same saturated bandwidth. It is also the reason why the chip here is able to get 96000 MB/s read/write/copy memory bandwidth over infinity fabric links that run at 2000mhz
@@JJFX-I clocked a Dominator DDR5-6200 kit with Hynix 16Gbit A-die to 8600 on Z690 Unify-X. Bench stable only, but the point stands. It's all in the ICs you get, the RAM vendor isn't that important.
We really need to harass AMD into giving us a single CCD SKU with GMI3-W (both infinity fabric package links on the CCD enabled) so we can properly take advantage of the bandwidth improvements in games.
That is if games actually even use more bandwidth in the first place. Games are generally latecny limited not bandwidth. No DRAM can get near SRAM/Cache level of low latency.
@@SolarianStrike Newer games, mostly open-world with "fast" traversal and RT, seem to actually be scaling quite well with bandwidth in many cases. This seem to be a bit more of a trend now ever since games started to be made only for the current gen consoles, maybe because consoles having high bandwidth/high latency memory (compared to PC system memory) and engines being adapted to run well on those?
@@LordApophis100 not entirely true. They are decoupled. But before this infinity fabric wouldn't go to 3000 and memory speed of 4000ish was abysmally low. So maybe 1:1:2 be a slightly better config. Won't know until actually tested. But I assume 8400 is difficult to stabilize
@LordApophis100 test it. At 6200 I lose latency performance at 2000, 2033, 2100 fclk but at 2067 I gain it back. I do gain read/write speed at 2100 vs 2067 tho. So throughput scales but latency like to be at an even ratio. It's definitely not entirely true.
Video request: Many people actually want to run 4 sticks at least at 6000 MT/s. Considering latest AGESA updates that improve memory clock, it would be really nice if you could do another "easy settings" for 4 sticks with Ryzen 7000 to run 6000 MT/s. (EDIT: preferably Hynix-M die cause that's what I have ;) )
This, I'm putting a PC together and looking at a kit of 2 16GB sticks at 6000. The mobo Memory QVL claims it would support 2 or 4. So I'm thinking 32 now and another 32 later. But AMD sites says 4 stick will get them to 3600. Running at 6000 is already thanks to EXPO/XMP from the official 5200. So wonder if 4 stick will have them struggling to get the 6000. The ones I'm looking are also Hynix-M from ADATA, SS. QVL also says it support 24 and 32 GB sticks but only 2, officially. Or at least that's what I get from the "DIMM Socket Support 2" while it says "2/4" with the other sticks.
having certain results costs effort and time, you probably don't deserve certain results if you don't want to work to achieve them. buy a ready-to-use PC that has already been tested and you will be more satisfied instead of going to someone who has to overclock components that are not designed to go at those speeds
I have the 8200 vcolor kit stable at 8000cl36 on an x670e extreme. I'm pretty much running your exact timings. Trfc is stable to about 660 but you should be able to do much better. I've always had trouble with my fclk 2100+ so I am taking the win and leaving it at 2000 lol. I mean some of my timings mightve been 8 where yours were 9, but your video pretty much described exactly where I landed on my 4dimm board with a 7950x3d
@nathalyberko1139 I have the v color 8200 cl38 kit. I have no idea what the performance jump is from 7000. There are so many variables to this. On a dual ccd chip going from 6400 in 1/1 to 8000 in 2/1 have about the same latency and much higher read/write speeds. On a single ccd chip there is zero reason to go 2/1 so you may as well stick to 6000-6400 in 1/1 and shoot for lowest latency. It is not easy to get 8000 working on amd and the actual fps gains are probably extremely small especially on an x3d chip that doesn't even require the fastest memory. Sorry there is no straightforward answer. Tldr, it depends, and probably less than you think unless you have a 7900 or 7950 AMD is fundamentally different from Intel so I'm not sure if any of this would apply to Intel
Any way we can get a video of 6400MHz 1:1:1 ? I heard some people already got 6400MHz with good tight timing I already have 6200MHz 30cl at 1.38v.. but I couldn't get stable 6400MHz with my Hynix A-die
You helped me in 2017 with a purchase of a MSI Pro Carbon AC. I have done nothing but praise you since as it has never failed me, even after maxing it out and changing the CPU to a 5600 from a 1600 in Feb/March. I just grabbed a 4090 and once I recover from that purchase I need to purchase a new mobo, ram, and CPU for it. I am looking at the MSI Carbon X670e paired with a 7900/7950 (still not sure if x3d or not) but then I see all of this memory stuff, and I am lost. I need 64GB (currently have 48GB) but I don't know what to do. I am even more confused after this vid. Any help I would GREATLY appreciate. Thank you again.
Hello, I was basically in the same situation as you, studying current MBD market for several days. Ended up with X670E Carbon WIFI because I just could not justify spending double money on an ACE just for 10G WLAN, Power / Reset Buttons onboard and ECLK. If the Carbon would have ECLK it would be the #1 Go-To board today. Even without it, it is a great board that (I hope) will carry me for somewhere between 5 to 10 years. Running 7800X3D, currently tweaking memory.
If the uncertainty around X3D chips applies to both 7900 and 7950 I would encourage you to avoid the 7900X3D. It only has 6 cores on the v-cache chiplet so you will lose out on some gaming performance in comparison to a 7800/7950X3D and also not perform as well in other workloads as a 7900X.
@@wctgttm Well, I am doing this in stages and the next stage is to replace my 6 year old monitor then in 2024 is time to replace the mobo, cpu, and ram. I am hoping AM6 arrives as they said it would in 2024 and maybe a new gen of Ryzen as the 7k lineup is pretty bad. I watched about this and it seems some is non x3d and there is no way to control which one it disables so a game will load up and be on the non X3D chiplet or next time be on the X3D chiplet. That is rubbish so not for me like that.
@buildzoid - with your current experience with the new AMD AGESA and new high speed kits, does this change your 'low effort timing settings' as published half a year ago? For 32GB kits, 64GB kits?
Hoping one of you can help me out. Picked up a 7800x3D and an Asus X670E-F MB and can't for the life of me figure out which RAM I should go with. I'm eyballing 2x24 or 2x32GB kits but I'm seeing so much conflicting information on whether Ryzen 7000 benefits from speeds over 6000, or if that's a bad idea. I know I should be able to use XMP ram fairly easily, and my Mobo officially supports up to 6600 (well 6600+ according to ASUS). I see some comments saying over 6000 doesn't benefit Ryzen 7, but usually without evidence or citation. Anyone know if I should go with higher speeds or not?
2x24 will be highest perf for a while, the latest BIOS enabled 2167 fabric clock and 6400 is trivial on all 24gb density Hynix M dies. There's Micron A die floating around though, if the CAS, RCD, and RP are all the same timing it's Micron; stay away.
I was able to hit 8000mhz on my 7950x3d but not sure if it is stable. I have been running memtest pro for a few hours and got a few (1 or 2 errors) I assume that to be not stable right?
*Buildzoid* : Don't go above 6000mhz DDR5 on Ryzen chips, it's the sweet spot. *Also, Buildzoid* : 8000mhz on Ryzen chips, that no Intel 13th gen can run properly. LOL!
VT3 is so mean that i've proven it crashes my cores at +10 curve optimiser when i have this fast of a memory overclock on. I've ended up cranking it to +30 for mem testing purposes. I don't think that it's any issue with the IMC, Interconnect or RAM as i've documented stable CO and core frequency getting worse with faster memory on all sorts of systems - the core gets fed more data, does more work, pulls more current, gets hotter and more. It's interesting though to see VT3 borderline failing at spec (the worst core i saw failed at -1 CO with 5200 JEDEC) and then losing 11 or more ticks of CO stability with the memory OC.
Very useful review, and well-timed for me as I have just picked up a 2x 24 GB M-die kit, 7950X3D and an Aorus 670E Master (latest Agesa, bios F13C). Your timings and DDR5 8000 from this vid work just fine 'as is' on that. You mentioned that you were thinking about a 4-Dimm board to test with the new 24 GB M-die - perhaps try that board.
I have a question for ddr4. Currently running a 5900x with trident z ram 3600cl15 1.35V (tuned to 3800cl16-15-8-12 - 1.55V) 4 dimms in total, Gigabyte aorus Master x570. 2 of em is bad and has a bad time running at those speeds, but if you put the good dimms in the right slots which is A2 and B2 then its running ok. the score wont be affected from going from 3533 mhz and up. Could it be because of the 2 bad memory dimms??
@@ActuallyHardcoreOverclockingvery weird. I run 7600 on mine, basically just fine? Probably an error or two happening under the hood, since I haven’t don’t real stability testing. But first time I’m hearing someone having such poor luck using that board
@@ActuallyHardcoreOverclocking Fair enough. I thought I was already pushing it with 7600, so not surprised 8000 was a step too far. Am surprised with seeing your recent vids on AM5 though. Part of reason I went Apex was purely looks and another was for higher speed DDR5; looks like AMD just need more time. Hopefully, Intel 15th gen and AM zen 6 lets us see both platforms really take advantage of these kinds of kits.
Год назад+1
So, I have a 7700x. What RAM clock (bandwidth) is the highest my CPU can benefit from?
if i had to guess as to why refresh is so common, it's probably rowhammer prevention... so lowering refresh could mean fancy rowhammer-style memory tests fail.
Regarding VDD/VDDQ discussion at around 12:00 to 15:00, It is not dangerous (or so It was said in early intel DDR5 platforms) to have a big difference between the two? I mean, would you stay with 1.4VDD/1.35VDDQ at most or is it safe to 24/7 lets say 1.43VDD/1.3VDDQ? Thanks
I would like to ask if there is someone who runs ryzen 7000 cpu with 4 sticks of ram at expo 6000 clock? I spend almost month trying but best what I can do is 4800 mt/s with 4 sticks and expo enabled (with ryzen 7800x3d).
Same for me too. Got 7950X3D, two mobos (TUF X670E and X670E Carbon) and 2 kits (A-Die) 2x16GB Kingston Fury Beast RGB 6000CL36. with 2 dimms, it can easily run 6200 CL30 on both TUF and Carbon. with 4 dimms, it can only run 4800, more than that will cause BSOD immediately or Corrupt windows (need to Restore windows). Trust me, I've tried everything (VDDIO/VDD/VDDQ 1.4v, Soc 1.3V, even messing with BUS configuration)
Can I do this with Kingston renegade fury 6000 CL30 kit on my 7800x3d or does this kit not overclock that high? I know this is a high end kit so I'm curious. I currently have it at 6400 CL30 at 2200 FLCK.
When running karhu don’t you run like 97% of your memory so like 1gb or less is free? I’ve got 32gb kit so wouldn’t I run like 31gb or so ? Sorry for the Noob question
Hi Buildzoid. I am a longtime fan and youtube subscriber. I am currently shopping for a new gaming PC. Not planning to overclock as I don't have as much time due to work. However, I want the fastest performance out of the box. Getting ROG STRIX B650E-E and rtx 4090. I am currently puzzled in regard to the RAM kit to get. 32 gb will not be enough for me, I need 48 or 64. I am in Europe. I found a kit that looks the fastest out of the box and with the new ryzen bios update I hope it should work: Teamgroup T-force Delta RGB 48 GB, DDR5-7200, CL34-42-42-84. However, as I understand it requires manual tuning since it doesn’t have AMD EXPO. In that case perhaps ADATA XPG LANCER 64GB, DDR5-6400, CL32-39-39 or Corsair Vengeance 64GB, DDR5-6000 CL30-36-36-76? Could you please share your opinion on whether these are good choices and advise on some alternatives? Thank you very much in advance, I am really hoping for your reply.
8000 series 12c/24t single ccd x3d with 1.5:1 memclk to uclk would b amazing. 8000 memory with 2666 uclk could put us into heavily tuned intel territory for latency and throughput.
Hi buildzoid I had 4200 MHz cl 16 16 28 on my z390 9900ks. After setting CPU voltage to adaptive from manual of to test if I could save some power . I can't return to 4200 anymore. It even freezes in bios. I noticed adaptive put my voltage past 1.5v. is my CPU messed up? It can do 3900 stable though
A dye test it with my good imc 13900KS (using my regular setup, P62,E48,R54 Ht off, already proven to take care of stable 8800) VDDQ TX: 1.32V, SA : 1.18V, IMC VDD: 1.43V DRAM VDD/VDDQ: 1.45V.
can you make a video about 4 dimm ddr5 overclock possibilities? i only get 5200 mhz on hynix ram stable to run. as soon as i set 5400 mhz, the system becomes more and more unstable over time. an expert video about possibilities and problems would be nice
Hello I am curious why SOC voltage needs to be so high. I understand when you are running in 1/1 mode and the MCLK is at 3000+ but in /2 mode it's running far slower at ~ 2000MHz.
From what I have heard/seen SOC needs to be high on ryzen 7000 systems for high mem clocks, can’t say with full confidence why though so that’s something to fully research
It doesn't have to be, i think this is a lack of testing since the config is new. 1.1 soc works for me at 8k. It's like you said, 2000 uclk is way easier than 3200.
Seen demo of 192GB (4x48GB) stable at 6400g2 with 2133 fclk. G2 is a notable performance hit but having a bazillion ranks per channel helps performance so it's not exactly bad.
Do you have a video on your process to get to these numbers? I bought the teamgroup 7200 48gb kit. I tried running your settings but it would have errors on memtest86. However, running your XMP profile worked for me allowing me to hit 8000mhz, so that gives me a better starting point to work with. I am just wondering how you get to these numbers, I am guessing you dont decrease everything 1by1 and have a general starting point you go towards as testing everything 1by1 would take forever.
Would be interested to see how far you can push Samsung memory dies with the latest update, knowing that Hynix is far superior. Unfortunately for me I didn't win the memory lottery. Samsung Memory stable at - Memory clock: 6400mhz Fabric clock: 2,100 Timings: 36-36-36-96 Voltage: auto/default Current build CPU: 7800x3d Mobo: B650M AORUS ELITE AX Memory: G.SKILL Trident Z5 NEO RGB 32GB DDR5 2x 16GB - 6000MT/s - CL36 - 1.35V - 36-36-36-96 - AMD expo Optimized
Used exact settings. B650e aours elite ax board. Boots fine. Games are stable while playing but is not stable in stress tests. Definitely performs better in my games compared to 6400 1:1
The ONLY game I’ve seen that may be “memory bandwidth” heavy is that unoptimized mess called Starfield. Game seems to be deliberately unoptimized in an attempt to sell more AMD GPU’s. Never a good idea to use benchmark results of unoptimized ports as a bases to predict the future of gaming in general. So I wouldn’t base this anomaly as a bases to predict anything.
Saw the new agesa pop, loaded it, and nothing but issues going to 6400 lol. Not sure what the deal is. Just set it back to 6000. Was kinda bummed. ASUS x670e-e/ddr5 cl30
Hey Buildzoid. I have been following your channel for the past month and your content is amazing! Could you possibly do a Hynx A 6600 UCLK=MCLK with the highest FCLK you can achieve? Specially if you ran it on that Asus motherboard you did the "timings for am5" video 👁️👁️
And just as a heads up, I'm using this video to try to get similar results with my 7600 + hynx a hahaha I know you said it's useless for sure CCD but it's just for shows
I had to increase VDDG CCD + IOD Voltage to get DDR4 3866 to work or rather , to get the IF/Cache in 1:1 stable without Cache WHEA Error in HW info . I use VDDG CCD at 1,09v and VDDG IOD at 1,125v on a Asus B550 Gaming E with an 5800X3D , it will start with DDR4 4000 / IF 2000 as well , but i cant fix the HWEA error without raising Voltages to probably insane Levels there
Hey Buildzoid, guys here
Guys here, Buildzoid hey
Buildzoid guys, hey here
T Shirt material.
Buildzoid Hey guys here
Buildzoid guys Hey here
Buildzoid here Hey guys
Buildzoid Hey here guys
Buildzoid guys here hay
Buildzoid Hey guys here
Hey buildzoid here guys
Hey here buildzoid guys
Hey guys buildzoid here
Hey buildzoid guys here
Hey here guys buildzoid
Hey guys here buildzoid
Guys buildzoid Hey here
Guys Hey buildzoid here
Guys here buildzoid Hey
Guys buildzoid here Hey
Guys Hey here buildzoid
Guys here Hey buildzoid
Here guys Hey buildzoid
Here guys buildzoid Hey
Here buildzoid Hey guys
Here buildzoid guys Hey
Here Hey buildzoid guys
Here Hey guys buildzoid
Would it be worthwhile to revisit the “easy memory timings for Hynix DDR5 with Ryzen 7000” video now with the new bios? Dummies like me hope we can get updated easy settings with even better performance. 😅 pretty please 🙏
Just what I was thinking...😂
What do you get in aida64 with those settings. I can't seem to get past 65ns but I also don't have power down disabled because I don't want to wait for slow ass memory training every boot and you have to enable power down with memory context restore enabled.
Bump
Thank you for sharing DDR5 overclocking experience! Following the settings in the video, I successfully increased the memory frequency to 8000MT/s and passed 30 rounds of VT-3 testing.
The spec used is as follows:
CPU: Ryzen 9 7950x3d
MB: GIGABYTE AORUS X670E MASTER(BIOS version:F30,AMD AGESA 1.1.7.0)
MEM: Hynix M-die 2x24GB2 kit
BuildZoid, an ideia that only you could pull up: how much does DDR5 speed and timings matter? As you are usually using your 7900x, you could lock the cpu clock and run some benchmarks of your choice to measure the ram impact (Geekbench, y-cruncher, Py Prime 32m, you name it..). You should run the benchmarks on each RAM that you tune up and build a database. Your videos are amazing and I think it would be really nice to see that.
Nice! I really enjoy your Ram stuff. Always learning new stuff. Thank you!
Hey BZ. Just a data point since you mentioned 1.2vsoc being best for your chips - mine is definitely lower, peaking at around 1100 or 1125. Setting 1100 vs 1200 is night and day on some configs with no throttling on 1100 vs 30% performance loss at 1200.
I currently run DDR5-8000 with a 4dimm board (x670e carbon) and 2x16GB of Hynix 16a, which is probably considerably worse than what you have - makes me wonder if 8400 is doable. Sadly there is no infinity fabric ratio for uclk:fclk sync at 8200 mem, but maybe we can get that in a future AGESA. I'm running that right now on 1.125v soc, but it runs fine on 1.10 so this is just solid safety margin (1.075 shows some performance loss).
Part of the reason i like the synced config is not having to run fclk up against the wall and deal with all of the funky instability that can happen because your room temperature changed or the planets are alligned wrongly - 2000 fclk just works, no questions asked, 100% of the time at 1.1vsoc or 1.25vsoc or whatever you want. Likewise the IMC is very happy to run at 2000mhz rather than driving it to 3200 plus in gear 1.
I also use a VDDP of 1050 but have not properly explored lower values yet, but also suspect too high causing failure.
I use VDDG 850 but that's completely untuned - substantially lower is probably optimal. My CPU will do 2167 fclk on 850 vddg, but it's a bit peculiar about it with only 2133 being super safe. This is generally slower than 2000 with uclk:fclk sync.
+5% IF isn't enough to overcome desync penalty. Between 5-10% is the crossover point most of the time, with 10% being better on almost everything.
Oh, and my stable primaries are 36-46-36 at 1.4/1.4/1.38 on the mem. 10 whole ticks
I think AMD"s memory controller is actually a bit better tuned for 16Gb A-die ATM. The LiveMixer and HDV both kinda run 8000 on A-die but really struggle with this 2x24GB kit.
@@ActuallyHardcoreOverclocking Don't have any 24M to test, but a 2 dimmer would be nice :D
Another data point I can add is that 8000 1:2 really likes low vsoc (1:2 in general), 6000-6400 1:1 likes it higher. There are people running 8000 at ~0.95vsoc. I think it's mainly because you're running the mclk really low, so there's no need for higher soc voltages.
Bro try 2100:2100:4200(8400) give it a go. Someone will probably make it happen.
I have a R9 7950X3D and I run into BSOD when running SoC to 1.200v but I'm running it currently with 1.235v and testing the Corsair Vengance 6400MT/s 2x48GB , Hynix M-Die and with the help of this channel I seem to be able to run them in 30-36-30-30 @ 6200MHz (MCLK 3100, UCLK 3100, FCLK 2100).
Not sure if it can do any tighter timings.
Still playing around still, but the SoC voltage really make an impact to the CPU idle temps, and I'm all about getting best performance at the best temps :).
I started laughing so hard at the 4023 mhz that I had to stop the video ... 40 x 99.9 = 4023 LOL . JUST THE USUAL AMD STUFF .. great job though and I love your channel BZ!
on the 7000 cpus, the infinity fabric is running dual 32yte/cycle interconnects. That doubles the total system memory bandwidth available to the cores. A single 2000mhz IF interconnect is limited to 64GB/s of total bandwidth. The dual interconnects creates a potential maximum of 128GB/s to the cores
Only on ryzen 9
@@BusAlexey No, Since threadripper 3000 introduced the dual interconnects to each CCD, the 5000 and 7000 series have all used dual interconnects to each ccd.
That is why the gaming performance improved with the 5000 series chips. Memory read traffic and GPU traffic no longer needed to fight to use the same saturated bandwidth.
It is also the reason why the chip here is able to get 96000 MB/s read/write/copy memory bandwidth over infinity fabric links that run at 2000mhz
@@bradmorri what do you mean, "chip here"? Chip here is ryzen 9, exactly what I was saying. so you're not proving anything for ryzen 7/5.
@@BusAlexey Im not proving anything. I am telling you what the facts are. I dont care if you believe me or not.
Thanks for the info! ❤
Incredible how AMD managed to unlock so much higher speeds "in one-go".
@fuckingmetaldudefabric clock speed is too low so AMD 8000mhz ddr5 won't improve any gaming performance
I hope my CMH64GX5M2B6000Z40 works with x670e Carbon at 6000mhz :" i'm gonna pray to god.
@@lemonhaze715 if you are going for 6000, why choose that expensive motherboard? I am using b650e strix e and running my 6400 cl40 kit at 8000 cl34.
@@lemonhaze715It's really time to stop buying Corsair memory.
@@JJFX-I clocked a Dominator DDR5-6200 kit with Hynix 16Gbit A-die to 8600 on Z690 Unify-X. Bench stable only, but the point stands. It's all in the ICs you get, the RAM vendor isn't that important.
G'day Buildzoid,
WOW! this running stable is crazy after so many years of AMD Memory being nowhere close to intel
We really need to harass AMD into giving us a single CCD SKU with GMI3-W (both infinity fabric package links on the CCD enabled) so we can properly take advantage of the bandwidth improvements in games.
The only way is for Intel to beat AMD so bad that AMD kind of wakes up and start seriously competing
Most likely not happening
That is if games actually even use more bandwidth in the first place. Games are generally latecny limited not bandwidth.
No DRAM can get near SRAM/Cache level of low latency.
@@SolarianStrike Newer games, mostly open-world with "fast" traversal and RT, seem to actually be scaling quite well with bandwidth in many cases. This seem to be a bit more of a trend now ever since games started to be made only for the current gen consoles, maybe because consoles having high bandwidth/high latency memory (compared to PC system memory) and engines being adapted to run well on those?
@@slimjimjimslim5923>implying they re not competing seriously
nice a video on 3 things im interested in. 24Gb dimms, 8000MHZ RAM, and AMD on 8000MHZ RAM !!
thanks for sharing :D
11:15 about the soc voltage being related to high infinity fabric speed or imc speeds. Is this the same on zen 2 cpus too?
That's cool at 2:1 mode 8000mbps still having good latency like 1:1 mode 6000mbps
I wonder if it can go to 2100 infinity fabric, 2100 uclk, 4200 memclk. Having the advantages of high speed and synchronization both
afaik Zen 4 has it decoupled, so having 1:1 does not give any latency benefits anymore.
@@LordApophis100 not entirely true. They are decoupled. But before this infinity fabric wouldn't go to 3000 and memory speed of 4000ish was abysmally low. So maybe 1:1:2 be a slightly better config. Won't know until actually tested. But I assume 8400 is difficult to stabilize
@LordApophis100 test it. At 6200 I lose latency performance at 2000, 2033, 2100 fclk but at 2067 I gain it back. I do gain read/write speed at 2100 vs 2067 tho. So throughput scales but latency like to be at an even ratio. It's definitely not entirely true.
lol that ram fan!! Best use of an amd box cooler ever!
thats a big gg
Video request:
Many people actually want to run 4 sticks at least at 6000 MT/s. Considering latest AGESA updates that improve memory clock, it would be really nice if you could do another "easy settings" for 4 sticks with Ryzen 7000 to run 6000 MT/s. (EDIT: preferably Hynix-M die cause that's what I have ;) )
This, I'm putting a PC together and looking at a kit of 2 16GB sticks at 6000. The mobo Memory QVL claims it would support 2 or 4. So I'm thinking 32 now and another 32 later. But AMD sites says 4 stick will get them to 3600. Running at 6000 is already thanks to EXPO/XMP from the official 5200. So wonder if 4 stick will have them struggling to get the 6000. The ones I'm looking are also Hynix-M from ADATA, SS. QVL also says it support 24 and 32 GB sticks but only 2, officially. Or at least that's what I get from the "DIMM Socket Support 2" while it says "2/4" with the other sticks.
I wish I could pay this guy to go through my Bios settings on my PC. Half the settings I have no idea what they do
having certain results costs effort and time, you probably don't deserve certain results if you don't want to work to achieve them. buy a ready-to-use PC that has already been tested and you will be more satisfied instead of going to someone who has to overclock components that are not designed to go at those speeds
🔥OMG! 8000MHx on AMD that is big !!🔥
I have the 8200 vcolor kit stable at 8000cl36 on an x670e extreme. I'm pretty much running your exact timings. Trfc is stable to about 660 but you should be able to do much better. I've always had trouble with my fclk 2100+ so I am taking the win and leaving it at 2000 lol.
I mean some of my timings mightve been 8 where yours were 9, but your video pretty much described exactly where I landed on my 4dimm board with a 7950x3d
So what is the performance gain in frames per second in a game if you go from 7000 to 8000mhz? Which ram kit do you have by the way?
Here for the response.
@nathalyberko1139 I have the v color 8200 cl38 kit.
I have no idea what the performance jump is from 7000. There are so many variables to this. On a dual ccd chip going from 6400 in 1/1 to 8000 in 2/1 have about the same latency and much higher read/write speeds. On a single ccd chip there is zero reason to go 2/1 so you may as well stick to 6000-6400 in 1/1 and shoot for lowest latency.
It is not easy to get 8000 working on amd and the actual fps gains are probably extremely small especially on an x3d chip that doesn't even require the fastest memory.
Sorry there is no straightforward answer.
Tldr, it depends, and probably less than you think unless you have a 7900 or 7950
AMD is fundamentally different from Intel so I'm not sure if any of this would apply to Intel
@@Cblan1224 Thanks for explaining, really appreciate it.
Do you have the timings I have the same board but with the 8000 Corsair dominator ram can't get it stable thanks
45:21 based bz at it again with the facts
Any way we can get a video of 6400MHz 1:1:1 ?
I heard some people already got 6400MHz with good tight timing
I already have 6200MHz 30cl at 1.38v.. but I couldn't get stable 6400MHz with my Hynix A-die
nice to see you use Karhu RT
You helped me in 2017 with a purchase of a MSI Pro Carbon AC. I have done nothing but praise you since as it has never failed me, even after maxing it out and changing the CPU to a 5600 from a 1600 in Feb/March. I just grabbed a 4090 and once I recover from that purchase I need to purchase a new mobo, ram, and CPU for it. I am looking at the MSI Carbon X670e paired with a 7900/7950 (still not sure if x3d or not) but then I see all of this memory stuff, and I am lost. I need 64GB (currently have 48GB) but I don't know what to do. I am even more confused after this vid. Any help I would GREATLY appreciate.
Thank you again.
Hello,
I was basically in the same situation as you, studying current MBD market for several days. Ended up with X670E Carbon WIFI because I just could not justify spending double money on an ACE just for 10G WLAN, Power / Reset Buttons onboard and ECLK. If the Carbon would have ECLK it would be the #1 Go-To board today. Even without it, it is a great board that (I hope) will carry me for somewhere between 5 to 10 years. Running 7800X3D, currently tweaking memory.
ECLK confuses me as that is a new one on me.@@easa1912svk
If the uncertainty around X3D chips applies to both 7900 and 7950 I would encourage you to avoid the 7900X3D. It only has 6 cores on the v-cache chiplet so you will lose out on some gaming performance in comparison to a 7800/7950X3D and also not perform as well in other workloads as a 7900X.
@@wctgttm Well, I am doing this in stages and the next stage is to replace my 6 year old monitor then in 2024 is time to replace the mobo, cpu, and ram. I am hoping AM6 arrives as they said it would in 2024 and maybe a new gen of Ryzen as the 7k lineup is pretty bad. I watched about this and it seems some is non x3d and there is no way to control which one it disables so a game will load up and be on the non X3D chiplet or next time be on the X3D chiplet. That is rubbish so not for me like that.
Thank you for this, just copied into my board and looks promising, test it now
So this is practically Gear 4? Is DDR5 with Gear4 on AMD comparable with Intel @ Gear2?
you still messing with them gigglebyte boys 😆
Tachyon gang g
The question is does it scale in game tho? 😊. Is it worth the effort?
YAY!
Does a 3000 MHz IMC mean theoretically we should see performance improvements up to 12000 MHz RAM speed at 2:1?
Finally you run Karhu
@buildzoid - with your current experience with the new AMD AGESA and new high speed kits, does this change your 'low effort timing settings' as published half a year ago? For 32GB kits, 64GB kits?
Can’t wait to see the performance of this vs 6200mhz on standard uclk=memclk
sounds like something 'Frame Chasers' would test if someone awares him
@ronteller2217 he has looked into it, its slower
6400c26 1:1> 8000 2:1
@dKenGuru yeah, what's funny is it actually benchmarks lower than my x99 ddr4 system (in memory tests) thanks to quad channel and low latency
@@ronteller2217he hates AMD so bad. Lol
I think a little too much
That said, he's 1 of my fave toobers
Hoping one of you can help me out. Picked up a 7800x3D and an Asus X670E-F MB and can't for the life of me figure out which RAM I should go with. I'm eyballing 2x24 or 2x32GB kits but I'm seeing so much conflicting information on whether Ryzen 7000 benefits from speeds over 6000, or if that's a bad idea. I know I should be able to use XMP ram fairly easily, and my Mobo officially supports up to 6600 (well 6600+ according to ASUS). I see some comments saying over 6000 doesn't benefit Ryzen 7, but usually without evidence or citation. Anyone know if I should go with higher speeds or not?
I'm in same situation
2x24 will be highest perf for a while, the latest BIOS enabled 2167 fabric clock and 6400 is trivial on all 24gb density Hynix M dies. There's Micron A die floating around though, if the CAS, RCD, and RP are all the same timing it's Micron; stay away.
What is the max ram speed for a
7950X3D? 8000 or more? Or is that useless?
I was able to hit 8000mhz on my 7950x3d but not sure if it is stable. I have been running memtest pro for a few hours and got a few (1 or 2 errors) I assume that to be not stable right?
Correct
What kit and motherboard?
if you get any errors it's not stable.
*Buildzoid* : Don't go above 6000mhz DDR5 on Ryzen chips, it's the sweet spot.
*Also, Buildzoid* : 8000mhz on Ryzen chips, that no Intel 13th gen can run properly.
LOL!
VT3 is so mean that i've proven it crashes my cores at +10 curve optimiser when i have this fast of a memory overclock on. I've ended up cranking it to +30 for mem testing purposes.
I don't think that it's any issue with the IMC, Interconnect or RAM as i've documented stable CO and core frequency getting worse with faster memory on all sorts of systems - the core gets fed more data, does more work, pulls more current, gets hotter and more. It's interesting though to see VT3 borderline failing at spec (the worst core i saw failed at -1 CO with 5200 JEDEC) and then losing 11 or more ticks of CO stability with the memory OC.
hi..have you tried to oc or tighter timings on2x48gb 6400 cl32 dualrank sticks ?..i cant find any info or testresults of those sticks..
Traumatized by ddr5 8000 on intel systems. 😂 Love your videos. Super informative.
Very useful review, and well-timed for me as I have just picked up a 2x 24 GB M-die kit, 7950X3D and an Aorus 670E Master (latest Agesa, bios F13C). Your timings and DDR5 8000 from this vid work just fine 'as is' on that. You mentioned that you were thinking about a 4-Dimm board to test with the new 24 GB M-die - perhaps try that board.
I have a question for ddr4. Currently running a 5900x with trident z ram 3600cl15 1.35V (tuned to 3800cl16-15-8-12 - 1.55V) 4 dimms in total, Gigabyte aorus Master x570. 2 of em is bad and has a bad time running at those speeds, but if you put the good dimms in the right slots which is A2 and B2 then its running ok. the score wont be affected from going from 3533 mhz and up. Could it be because of the 2 bad memory dimms??
You will ALWAYS have to use lower speeds when using 4 dimms. It has always been like that. Sry
@@bismuth7730 ok, tryin to get timings tight then :)
@@bismuth7730 also found out, that my infinity fabric controller do not like a fclk higher than 1767. Like it wont go higher, its just locked there XD
Can you show where is B650 Tachyon can be purchased?
it's completely sold out. GIgabyte only made a a few hundred of them AFAIK.
Did you just completely lose the silicon lottery? Jeez. The fact your apex wouldn’t even boot is crazy to me.
nah the Apex boots it just errors in everything.
@@ActuallyHardcoreOverclockingvery weird. I run 7600 on mine, basically just fine? Probably an error or two happening under the hood, since I haven’t don’t real stability testing. But first time I’m hearing someone having such poor luck using that board
@@aj0413_ 7600 on intel isn't too much of a problem. 8000 is.
@@ActuallyHardcoreOverclocking Fair enough. I thought I was already pushing it with 7600, so not surprised 8000 was a step too far. Am surprised with seeing your recent vids on AM5 though. Part of reason I went Apex was purely looks and another was for higher speed DDR5; looks like AMD just need more time.
Hopefully, Intel 15th gen and AM zen 6 lets us see both platforms really take advantage of these kinds of kits.
So, I have a 7700x. What RAM clock (bandwidth) is the highest my CPU can benefit from?
I hope to see 2x 48gb and 4x 48gb tests
I read everywhere that 8000 MHz is not supported by the Ryzen 9 7900x???best is 5200??
if i had to guess as to why refresh is so common, it's probably rowhammer prevention... so lowering refresh could mean fancy rowhammer-style memory tests fail.
Regarding VDD/VDDQ discussion at around 12:00 to 15:00, It is not dangerous (or so It was said in early intel DDR5 platforms) to have a big difference between the two? I mean, would you stay with 1.4VDD/1.35VDDQ at most or is it safe to 24/7 lets say 1.43VDD/1.3VDDQ? Thanks
I would like to ask if there is someone who runs ryzen 7000 cpu with 4 sticks of ram at expo 6000 clock? I spend almost month trying but best what I can do is 4800 mt/s with 4 sticks and expo enabled (with ryzen 7800x3d).
Same for me too.
Got 7950X3D, two mobos (TUF X670E and X670E Carbon) and 2 kits (A-Die) 2x16GB Kingston Fury Beast RGB 6000CL36.
with 2 dimms, it can easily run 6200 CL30 on both TUF and Carbon.
with 4 dimms, it can only run 4800, more than that will cause BSOD immediately or Corrupt windows (need to Restore windows). Trust me, I've tried everything (VDDIO/VDD/VDDQ 1.4v, Soc 1.3V, even messing with BUS configuration)
Fuck yeah
Can I do this with Kingston renegade fury 6000 CL30 kit on my 7800x3d or does this kit not overclock that high? I know this is a high end kit so I'm curious. I currently have it at 6400 CL30 at 2200 FLCK.
That kit can get 7600mhz from what I have seen so probably can get 8k or so, I don’t think it will be much faster or won’t even be faster than 6400mhz
Depends If it's M die or A die.
@@SAVA47-00 A die is better?
When running karhu don’t you run like 97% of your memory so like 1gb or less is free? I’ve got 32gb kit so wouldn’t I run like 31gb or so ? Sorry for the Noob question
And here I am unable to run a Gskill flare x 32GB kit at 6000 with expo on an ASUS TUF X670E WIFI
Hey BZ, how about that V-Color Manta Xprism RGB? Is it totally wild or what?🎉😊
Was the tachyon ever released in the US?
Wait there are 24GB memory sticks? I thought there where only sticks with powers of 2 of ram
Hi Buildzoid. I am a longtime fan and youtube subscriber. I am currently shopping for a new gaming PC. Not planning to overclock as I don't have as much time due to work. However, I want the fastest performance out of the box. Getting ROG STRIX B650E-E and rtx 4090.
I am currently puzzled in regard to the RAM kit to get. 32 gb will not be enough for me, I need 48 or 64. I am in Europe. I found a kit that looks the fastest out of the box and with the new ryzen bios update I hope it should work: Teamgroup T-force Delta RGB 48 GB, DDR5-7200, CL34-42-42-84. However, as I understand it requires manual tuning since it doesn’t have AMD EXPO. In that case perhaps ADATA XPG LANCER 64GB, DDR5-6400, CL32-39-39 or Corsair Vengeance 64GB, DDR5-6000 CL30-36-36-76? Could you please share your opinion on whether these are good choices and advise on some alternatives?
Thank you very much in advance, I am really hoping for your reply.
Will those timing works on Hynix A die too?
8000 series 12c/24t single ccd x3d with 1.5:1 memclk to uclk would b amazing. 8000 memory with 2666 uclk could put us into heavily tuned intel territory for latency and throughput.
No lol
Nice. Will the tuning be similar with 2x16GB A-die?
Nope they are 2 different dies
kinda funny cuz ddr5 8200/8400 xmp kits ONLY work on intel :D
But how does it play minecraft
Hi buildzoid I had 4200 MHz cl 16 16 28 on my z390 9900ks. After setting CPU voltage to adaptive from manual of to test if I could save some power . I can't return to 4200 anymore. It even freezes in bios. I noticed adaptive put my voltage past 1.5v. is my CPU messed up? It can do 3900 stable though
Oc was 100 stable
would love to see something like this on an Ryzen 8700G
A dye test it with my good imc 13900KS (using my regular setup, P62,E48,R54 Ht off, already proven to take care of stable 8800) VDDQ TX: 1.32V, SA : 1.18V, IMC VDD: 1.43V
DRAM VDD/VDDQ: 1.45V.
Anyone tried these on the new Z790i Asrock Lightning Wifi? The XMP 8000 CL36 works and stable! Just the latency is horrible topping 68ns.
Any improvement of fps for gaming 8000 mhz to high for this cpu
Reboot stable? mine is stable and when i reboot it is not
can you make a video about 4 dimm ddr5 overclock possibilities? i only get 5200 mhz on hynix ram stable to run. as soon as i set 5400 mhz, the system becomes more and more unstable over time. an expert video about possibilities and problems would be nice
I would love a Video bringing a Hynix a kit 6000 to 8000 on b650e ryzen X3D
I’m curious to see how llama cpp would perform on this
Hello I am curious why SOC voltage needs to be so high. I understand when you are running in 1/1 mode and the MCLK is at 3000+ but in /2 mode it's running far slower at ~ 2000MHz.
From what I have heard/seen SOC needs to be high on ryzen 7000 systems for high mem clocks, can’t say with full confidence why though so that’s something to fully research
It doesn't have to be, i think this is a lack of testing since the config is new. 1.1 soc works for me at 8k. It's like you said, 2000 uclk is way easier than 3200.
You're correct, 1:2 requires low vsoc, you can even get it lower than 1v
IGPU benchmarks?
Can anyone confirm if the 3D CPUs support 4 DIMMs with the new BIOS? I would really like to get 128gb of memory going at 6000mt if possible.
Seen demo of 192GB (4x48GB) stable at 6400g2 with 2133 fclk.
G2 is a notable performance hit but having a bazillion ranks per channel helps performance so it's not exactly bad.
X3D uses the same exact memory controller and substrate as all the other chips...
What is he max speed and timings that work for 7800x3d
What about 4 dimm board 8000 xmp?
command rate 2T?
8200gbits cpu 4164 t4 39hz. Over 6 core 24gb 8gb thst blaist the system 4gb 8gn to 12gb spred xmp
Do you have a video on your process to get to these numbers? I bought the teamgroup 7200 48gb kit. I tried running your settings but it would have errors on memtest86. However, running your XMP profile worked for me allowing me to hit 8000mhz, so that gives me a better starting point to work with.
I am just wondering how you get to these numbers, I am guessing you dont decrease everything 1by1 and have a general starting point you go towards as testing everything 1by1 would take forever.
Waiting on 192 GB tests... 😅
probably not happening this year.
Would be interested to see how far you can push Samsung memory dies with the latest update, knowing that Hynix is far superior. Unfortunately for me I didn't win the memory lottery.
Samsung Memory stable at -
Memory clock: 6400mhz
Fabric clock: 2,100
Timings: 36-36-36-96
Voltage: auto/default
Current build
CPU: 7800x3d
Mobo: B650M AORUS ELITE AX
Memory: G.SKILL Trident Z5 NEO RGB 32GB DDR5 2x 16GB - 6000MT/s - CL36 - 1.35V - 36-36-36-96 - AMD expo Optimized
Nice.
im running 8000mhz cl 38-48-48 on my 7900x also have a 6200 cl 28 profile
How do they compare for latency?
Used exact settings. B650e aours elite ax board. Boots fine. Games are stable while playing but is not stable in stress tests. Definitely performs better in my games compared to 6400 1:1
now AMD just needs to increase fabric speed, 3ghz sounds nice
I love your channel, but I would love it more if your videos would be 4k.
If you patreon him enough to buy a 4k capture card im sure he would do it.
Yeah 4k anything isn't that cheap. Unless you want to buy one for BZ. I'm sure he would take one off your hands.
4k for 800x600 bios and 1080p desktop?
To see the Memory RGB better?
actually I usually have the desktop at 900p or even lower so that the text is more legible on phones. @@Mugris
What does this do for gaming? How much perfomance gains would that give in a cpu bottleneck?
And now my wheels are turning…in a game like Warzone is the 3D cache ccd now worse than the higher clocked ccd at 8000mts memory?
works pretty good with 7600
The ONLY game I’ve seen that may be “memory bandwidth” heavy is that unoptimized mess called Starfield.
Game seems to be deliberately unoptimized in an attempt to sell more AMD GPU’s.
Never a good idea to use benchmark results of unoptimized ports as a bases to predict the future of gaming in general.
So I wouldn’t base this anomaly as a bases to predict anything.
9600 over 300mhz 3x 3200mhz thats bye me max over 6x ram 3200mhz 5.1ghz= 5.5ghz 5.9ghz neu cpu lvl now=
Saw the new agesa pop, loaded it, and nothing but issues going to 6400 lol. Not sure what the deal is. Just set it back to 6000. Was kinda bummed.
ASUS x670e-e/ddr5 cl30
Hey Buildzoid. I have been following your channel for the past month and your content is amazing! Could you possibly do a Hynx A 6600 UCLK=MCLK with the highest FCLK you can achieve?
Specially if you ran it on that Asus motherboard you did the "timings for am5" video 👁️👁️
And just as a heads up, I'm using this video to try to get similar results with my 7600 + hynx a hahaha I know you said it's useless for sure CCD but it's just for shows
I had to increase VDDG CCD + IOD Voltage to get DDR4 3866 to work or rather , to get the IF/Cache in 1:1 stable without Cache WHEA Error in HW info . I use VDDG CCD at 1,09v and VDDG IOD at 1,125v on a Asus B550 Gaming E with an 5800X3D , it will start with DDR4 4000 / IF 2000 as well , but i cant fix the HWEA error without raising Voltages to probably insane Levels there
4166mhz give 30hz 40hz but not more speed
1945 ist max 3900mhz over .176volt 980
Why do i have same aida numbers at 6400c34 on intel system? Think 8000 would be much higher. On intel systems 8000 is like 120-130gbs and low 50ns
Can't decide between 13700K or 7900X .
I would disable Hyperthreading and E-cores on Intel and do the same on AMD + direct die cooling .
7900x eco mode
Why would u disable HT and E-Cores lol 😂
@@od1sseas663 latency
@@starkistuna zero real world improvement , like literally
@@od1sseas663 turning off Smt save a little more power that can be used to overclock a cpu further: ruclips.net/video/-uvhk823e8Q/видео.html