Jump to content
NotebookTalk

jaybee83

Member
  • Posts

    2,041
  • Joined

  • Last visited

  • Days Won

    40

Everything posted by jaybee83

  1. heyho waddhup guys, back from the (almost) dead here. so seems like after almost 4 years of successfully avoiding it, that damn covid bug finally got me. and OF COURSE i infected both my wife and lil babygirl in one fell swoop, ugh... keep your fingers crossed, i seem to be kinda on the mend, so hope i can take over babygirl care tomorrow seeing as my wife is crashing today... anyways, so that happened while i was in the middle of figuring out the details of my custom IHS install, my desktop is still in pieces on my desk waiting for me to get back my strength haha. ill keep you guys posted on developments. dude thats some insane pricing, id have gotten a kit as well in your place. ahaha THE FIRST INSTANT THOUGHT i had when i saw this video pop up on yt was our forum 😄 Janktop FTW!
  2. enough already, send us some! 😛
  3. interesting, the highest end RTX 5000 is equipped with a 4080 die.
  4. music to my ears, switching from iphone to pixels 😄 what kinda cases did u guys go for?
  5. noice, that blue looks amazing! i got the 7 pro in white, super happy with it 🙂 now if i just found the time to upgrade to Android 14.... 😄
  6. nice, bookmarked the article. should a good read for my lunch break later 🙂 nice, good job! now to tighten up all the subtimings as far as they can go 😄
  7. noice uve definitely found the right forum for Clevo freaks 😄 welcome!
  8. thought it was good! plus its a nicely wrapped story at the end... going out with a BANG (no spoilers 😛 )
  9. nope thats why id be curious to get one 😄 other users have reported a considerable bump in system snappiness tho 🙂
  10. ah no, im not talking about the small capacity optane ssds used as caches to speed up slow HDDs. there are larger capacity optane drives such as the 1.6 TB ones to be used as the main drive in a machine. as such, they sport insane latencies and super high IOPS at low queue depths, plus they have absolutely insane writing durability and do not degrade in performance even when they are filled to the brink. not even the fastest regular NAND ssds can compete with that, thats why optane is priced so ludicrous (that and the comparably low storage density, were basically talking SLC here). in short: an optane SSD is basically the 4090 of SSDs, ultra performance at ultra pricing 😄 here a review if youre interested: https://www.tomshardware.com/reviews/intel-optane-ssd-dc-p5800x-review even compared to pcie 5.0 drives the P5800X is miles ahead:
  11. i admit im not a big GTA fan, i was rather old school and totally hooked into the OG GTAs of old with the birdview perspective: and yes i know ive now outed myself as OLD but thats ok hahaha 😛 nevertheless im looking forward to all the "hidden lore" storylines in GTA VI, i thoroughly enjoyed watching the summary videos on the alien / bigfoot / mount chilliad of GTA V on yt 🙂
  12. some good points right there, thanks bud 🙂 the 3500 / 3800 speeds were with her previous 2x8GB 3800 sticks tho. i have since upgraded her to 2x32GB 3200 sticks, so curious to see how far i can push those. probably not as far, but across the whole lifetime of her machine (im expecting 10yrs) capacity is more important than speed 🙂 plus, of course, its fun to keep upgrading it 😄 im currently preparing aan 8TB QVO 870 ssd for her as a xmas present, shes running out of space on her external 5TB spinner, so thats gonna be a nice bump both in capacity and speed for her. QLC should be fine in her case, were tallking mostly media stuff like movies, music, pics, etc.
  13. i think u missed the point, its 1.6 not 16 hahaha. optane is friggin expensive per GB!!!
  14. i might as well just go all out with a 14900K 😄 but tbh im not planning to purchase one anytime soon, might wait until 15th gen comes out and prices come down. so far shes quite happy with the 12600. mind you, thats a laptop shes rocking, so naturally the mobo traces and so-dimms are not able to compete with desktop hardware haha
  15. nope, as opposed to Intel Alder/Raptor/RPL-R Ryzen 7000 only supports DDR5, no hybrid IMC... My lady's machine sports 12th gen paired with DDR4, curious to see how much RAM OC will improve once I bump her up to 14th gen 🙂 with the 12600 currently installed, she can do up to 3500 in Gear 1 and 3800 in Gear 2.
  16. thanks, appreciate it 😀 very good question, actually forgot to add an info relating to that: The Misc voltage is a general power delivery path from the mobo to the CPU, the VDDG voltage is actually derived from that and controls the IF / FCLK voltage. funny enough, i had to increase the Misc voltage for RAM OC but VDDG didnt have any impact whatsoever, either for IF or RAM OC. I think that the stock VDDG voltage is already enough for any kind of IF or RAM OC, at least in my case. Could be that at even higher RAM or IF speeds one would need to also adjust VDDG in the future. Ill add it to the above post 🙂
  17. Aight gaiz, i know its been LONG overdue now, but finally getting around to writing up my AM5 / Ryzen 7000 DDR5 OC adventures 😄 hope this helps my fellow AM5 owners! Stress testing: Ive been using Tm5 with Anta777 Absolut profile, which gave me very nice results in terms of stability. anything that passes that will pass everything else in my everyday usage. For my current 2x24GB kit, it needs roughly 3 hours to complete at SPD settings (5200 Mhz) and 2 hours at tuned settings (6200 or 7600 with tightened timings). I have also seen lots of recommendations for Karhu, but havent tested that yet. Infinity Fabric (IF / FCLK) - Stock is at 1800 - 2000 Mhz, depending on motherboard and bios revision - Depending on CPU quality, typically maxes out around 2133 to 2200 Mhz (in my case 2167 Mhz) - As opposed to Ryzen 5000, the IF is now decoupled from the Memory Controller (MCLK) and Memory itself (UCLK), so you can run it asynchronously - Depending on the setup and settings, I found that sometimes synchronous, other times maxxed out IF clocks are better for the resulting memory latency (i.e. DDR5-6200 with IF at 2067 = 3x ratio synchronous vs. DDR5-6200 with IF 2167 = 2.86 asynchronous), so best to check both! "Gears" = UCLK/MCLK ratios - "Gear 1" = UCLK/MCLK = 1:1 ratio, i.e. Memory Controller at 3000 Mhz and DDR5-6000 at 3000 Mhz actual clocks - "Gear 2" = UCLK/MCLK = 1:2 ratio, i.e. i.e. Memory Controller at 2000 Mhz and DDR5-8000 at 4000 Mhz actual clocks - As with Intel, Gear 2 has a latency penalty but allows for much higher RAM frequencies and thus bandwidth. - In my case, I maxed out at the following: DDR5-6400 bootable and DDR5-6200 stable in Gear 1 DDR5-8000 bootable and DDR5-7600 stable in Gear 2 - From what ive seen, typical max ranges for Gear 1 are 6200-6600, for Gear 2 7600-8200. How stable those were though, I cannot judge 🙂 your results WILL vary depending on motherboard, CPU IMC quality and RAM sticks, as per usual... Voltages - VDDG does not have any impact on RAM OC (at least in my case, see next point directly below) - The Misc voltage is a general power delivery path from the mobo to the CPU, the VDDG voltage is actually derived from that and controls the IF / FCLK voltage. funny enough, i had to increase the Misc voltage for RAM OC but VDDG didnt have any impact whatsoever, either for IF or RAM OC. I think that the stock VDDG voltage is already enough for any kind of IF or RAM OC, at least in my case. Could be that at even higher RAM or IF speeds one would need to also adjust VDDG in the future. - SOC, VDDIO, Misc, VDDP, as well as RAM VDD and RAM VDDQ are important for RAM OC on AM5 - With a bit of trial and error, as well as lots of researching and comparisons I have found the following voltage ranges safe for testing (all of the listed max values I have been running for months during tuning of timings, so at least from my side I consider those safe for 24/7, your mileage may vary!): SOC = Stock 1.00V, max. 1.30V (max possible with newer Bios revisions) VDDIO = Stock 1.10V, max. 1.58V Misc = Stock 1.10V, max. 1.55 V VDDP = Stock 0.80V, max. 1.18V RAM VDD = Stock 1.10V, max. 1.63V RAM VDDQ = Stock 1.10V, max. 1.63V Naturally, after finding ur best settings for timings, I would suggest to go back and lower the above voltages as much as possible to save on voltage wear and heat as much as possible. Here were the lowest settings I found to be stable at my tightened timings: DDR5-6200 Gear 1 (G.Skill 2x16GB Kit DDR5-6600) SOC = 1.20V VDDIO = 1.31V Misc = 1.10V (Stock) VDDP = 1.03V RAM VDD = 1.52V RAM VDDQ = 1.52V DDR5-6200 Gear 1 (Teamgroup 2x24GB Kit DDR5-8200) SOC = 1.19V VDDIO = 1.27V Misc = 1.39V VDDP = 0.81V RAM VDD = 1.58V RAM VDDQ = 1.59V DDR5-7600 Gear 2 (Teamgroup 2x24GB Kit DDR5-8200) SOC = 1.26V VDDIO = 1.52V Misc = 1.26V VDDP = 1.10V RAM VDD = 1.59V RAM VDDQ = 1.59V Notes: - VDD can sometimes be lowered further than VDDQ, so check and see whats possible here. In my case only 0.01V was possible. - As you can see, minimum required voltages are also different for each ram kit, so your results may vary, even with the same setup otherwise. Additional Settings Nitro Mode The most recent Bios updates for AM5 boards, a new "Nitro Mode" has been implemented, which allows a more finegrained tuning of RAM training. Loosening up these settings allows for more stability at the cost of latency and potentially longer boot times. It includes the following settings: Nitro Mode = Enabled / Disabled Robust Training = Enabled / Disabled (Enable for better stability) Nitro RX Data / Nitro Tx Data / Nitro Control Line / Nitro Rx Burst Length / Nitro Tx Burst Length = basically, the higher the factor, the higher the stability / latency / boot time. What I did was to max out everything, same with the voltages, then tighten up all the timings, afterwards go back and find the lowest values that are still stable to improve on latency and boot times. Other There are two additional settings important for RAM tuning in this case: "Memory Context Restore" and "Power Down Enable" For best latency and performance (i.e. reaching tighter timings / higher clocks) I recommend disabling them both. This will, however, significantly increase our boot times due to RAM retraining going on for every boot. I leave them both disabled cuz i dont find the longer boot times that bad, were talking maybe 30-40 seconds vs. 10 seconds, definitely worth the tradeoff for better (and stable!) performance 🙂 Cooling This is more of a general thing and not specific to AM5: DDR5 RAM is finicky when it comes to temperatures! In my case I leave the RGB switched off, which bring about 5-10C lower temps (crazy, right?). In addition, I have a 140mm fan pointed directly at the RAM sticks to further lower the temps. I am usually around 50C under full load with TM5 stress testing and in the high 40s when gaming and the whole case is heated up by the GPU and CPU. Ideally, you would replace the stock heatsinks with aftermarket ones or add liquid cooling to it. Whatever you do, just keep an eye on temps! The lower, the more tight the timings and higher the clocks can be with good stability 🙂 Timings I am including my tuned timing results for both my previous G.Skill 2x16GB DDR5-6600, as well as my current Teamgroup 2x24GB DDR-8200 kits. Notes: - The G.Skill kit SPD was at DDR5-4800, whereas the Teamgroup kit sports DDR5-5200 at stock. - The G.Skill kit (on an older bios) profited from synchronous IF, whereas the Teamgroup kit preferred the max. IF clock at 2167 Mhz in both Gear 1 and Gear 2. - For tightening the timings, i maxed out the voltages as well as Nitro Mode settings and then went back after tuning to lower everything again as much as possible. Results I checked performance with AIDA64, 10 consecutive runs for each setup and calculated the average for each parameter to account for any variability. I am using the DDR5-4800 G.Skill stock as the 100% baseline here. G.Skill Stock DDR5-4800 Read = 54835 Write = 52866 Copy = 52664 Latency = 96.1 G.Skill Tuned DDR5-6200 Gear 1 Read = 90968 (+66%) Write = 94858 (+79%) Copy = 81608 (+55%) Latency = 55.4 (+74%) Teamgroup Stock DDR5-5200 Read = 64214 (+17%) Write = 64992 (+23%) Copy = 58463 (+11%) Latency = 94.7 (+1%) Teamgroup Tuned DDR5-6200 Gear 1 Read = 91237 (+66%) Write = 95211 (+80%) Copy = 85265 (+62%) Latency = 54.6 (+76%) Teamgroup Tuned DDR5-7600 Gear 2 Read = 99155 (+81%) Write = 101802 (+93%) Copy = 91402 (+74%) Latency = 57.9 (+66%) Conclusions - Based on the obtained results and input that i saw from other AM5 users online, I am likely held back by the CPU IMC. The RAM sticks did not show much difference between the two kits in Gear 1. Also, I have seen differing results on the X670E Extreme and Hero boards, so that leads me to think that the CPU is my current bottleneck. Curious to see what Zen 5 will bring! - As mentioned before, Gear 2 lets the latency suffer a bit, but provides a nice boost to bandwidth. I will stick to my Gear 2 settings for now, as the latency penalty is a mere 3 ns in my case, so nothing to worry about. Overall quite happy with the results at roughly 60-90% performance uplift vs. stock speeds of my previous G.Skill RAM kit 🙂I didnt even really bother testing with the XMP / EXPO profiles tbh, just used them as a quick baseline for additional tuning 😄 Aight, with GPU and RAM tuning out of the way, Ill now shift my focus on tweaking my 7950X 🙂
  18. An Intel Optane drive, specifically the P5800X with 1.6 TB capacity: would be super curious to see firsthand how it fares in real world scenarios. but yeah, at currently 8000€ probably not gonna get that.... 😄 used to be a "super value" 3000€ tho! maybe i can get a good price on ebay hahaha edit: yup, listings on ebay start at 2400€, total steal 😛
  19. couldnt agree more with this. the main factor i focus on is "usage time". so get the king card right when its released and enjoy it for the longest time. IF a 4090 Ti ever comes out (which i doubt, no monetary incentive for Nvidia), ill be totally chill without any need to upgrade. at best, with full fat 102 die, itll be what? maybe 10-15% more performance? plus with my card oced to the max that gap gets quite small, so couldnt be bothered to upgrade. as you said, ill just wait and see what the 5090 brings to the table, until then i can save up enough money to do that upgrade without having to sell off my 4090 beforehand. no downtime, max usage time and some money back from selling off my 4090 in the end 🙂 and if the 5090 is NOT available ANYwhere due to "insert random tech hype for scalper prices" then i still have my 4090, all good! oh yes, very happy with my AC LF II 420, especially with the 6x Noctua IndustriallPPC 3000 fans in push-pull, quite the cooling beast 😄
  20. dude wtf, almost 50 million views inside 9 hours 😄 i think ppl are excited? hahaha
  21. oh yes definitely way more finicky. ive popped several intel cpus and compared to this total piece of cake 😄 but when it does work, all the more satisfying! likely wont be coming around to installing the custom TG ihs until next weekend, i need the machine working in the meantime 😛
  22. all cleaned up, protected the SMDs with some fingernail polish from my wife (applied after i took this pic) and off back under the stock IHS with LM and kryonaut on top for some initial testing. luckily the cpu still works 😁 got some baseline results before delidding with CB2024 60 min loop so had smth to compare this to. results aint too bad, -2.5C to -4.7C depending on CCD. so with the custom TG IHS should be even better, especially with double LM under and on top of it 😎
  23. very nice deal indeed, although shipping and customs would bring that up considerably for international customers 😅
  24. sad to hear man, but thanks for the valuable first hand info on current AMD gpus. as for X3D parts dying...welp that in combo with the limited tuning capabilities kinda keeps me from switching over. quite happy with my 7950X, curious to see what its actually capable of, so far ive been focusing on the 4090 and ram, now its cpu tuning time! speaking of which: was a bit more finicky than with intel cpus, where its just one "plopp" and done. due to the restricted movement area @SMDs its a lot of back and worth to make the indium solder layer fail via material fatigue. still, works nicely with the delid-die-mate! cant test the cpu yet since were on the road (wifey driving, baby sleeping, daddy casually delidding on his lap in the backseat 😂), but result looks very promising, no obvious chip damage. ill clean up the pcb, dies and stock ihs and update again once done 😁
  25. yep same here, enjoying my suprim x wwaaay too much, especially with the 1000W XOC vbios 🙂
×
×
  • Create New...

Important Information

We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue. Terms of Use