electrosoft

Holding out: @Mr. Fox @Papusan Going with the flow: @electrosoft Definitely buying: @johnksss @jaybee83 @tps3443 @Talon Did I miss anyone? 🤣

Something must have been up with 7.02 BIOS for the Asrock Z790i Lightning when no matter what I did, TM5 was failing even on 6400 sticks (those same sticks running rock solid on the Hero right now at 2133/6400). I went back to 6.02 and the problems disappeared. 8.02 just dropped so as a test I went BACK to 7.02 and TM5 was failing again. Went to 8.02 and now TM5 passes no problem again. Yeah, hopefully the reviews will be a bit better but actual real benchmarks out in the wild aren't exactly showing that 30-35% uplift both in OpenCL and Vulkan along with those game tests but the deluge of reviews next week will provide clarity. Translation: "Hope I score a 5090, but if not I'm good with my 4090....but I really would like to score a 5090!" 🙂 4090 is still a monster....

Hmmmm, @jaybee83 this looks interesting. I didn't even notice this post here, but I'll take a looksy, thanks! Do you have some screen shots of the BIOS post unlock and some results @ecxedxrip?

Same reaction to not having an 18" version of this approach but at least they're trying. If they can aquire mobile dies and fab their own GPUs like this, that opens up possible upgrades. They just need to keep this form factor going forward. ----- 5090 benchmarks looking not spectacular..... ~27% uplift in Vulcan and 14% in OpenCL: https://videocardz.com/newz/nvidia-geforce-rtx-5090-appears-in-first-geekbench-opencl-vulkan-leaks ~20% gaming uplift with all the bells and whistles turned off but testing methodology is odd in early review leaks of the 5090D (which isn't game capped): https://videocardz.com/newz/geforce-rtx-5090d-reviewer-says-this-generation-hardware-improvements-arent-massive

He's not wrong on a few points. Same Node. Same tech basically. More cuda cores More voltage More memory It reminds me of Nvidia's 500 series a bit when we considered that hot and power hungry stretching the node. This is also why I said before I feel like a bit of Intel at play here and Raptorlake a touch. ---- I keep going back and forth. Still in that "Go big (5090) or go home (Used 7900xtx/9070XT/5070) mindset this time around. Do I want the absolute best performance? Am I ok with a lesser card? Tune in January 30th for the next exciting episode of, "As the tech turns." I like the uncertainty of what I am going to do in less than two weeks. 🙂 I'm eagerly awaiting the various reviews of the 5070, 5090 and 9070xt.

You mean next generation? 9070XT would be at best a side grade to your 7900XTX but will actually be less powerful if reports are correct. I'm tempted to just snag a used 7900XTX and call it a wrap depending on the 9070XT reviews.

Can you trade it in before launch day so you have a credit waiting? lol, that posting over on OC offering $780 for a Strix 4090.....

Nah, the 5090 is going to come in at 30-35% which is more than enough for me to pull the trigger if I disregard proper fiscal responsibility. If it turns out to be faster, woot! But all reports are leaning towards temper your expectations. That uplift is coming mainly from the cuda core uplift (which is about ~33%) and Nvidia opening up the bus and increasing memory frequency enough to properly feed them / that memory interface. The more I read up on the FE, the more I'm now leaning equally towards all three now with the FE and Suprim tied assuming the Suprim has the superior PCB design. Astral will be good, but I know that price is going to be outlandish. I will pull the trigger on whatever model presents itself and makes it into the basket and checks out. 🤣 4080 Super -> 5080 is an easier read since the bus is the same (256 bit), and you are getting ~5% uplift in CUDA cores but faster memory. It really comes down to the break even point between the cuda cores, memory bus and if the faster GDDR7 is fully being utilized to feed it but we know it is to a degree as the reported uplift is going to be right around 10-11%. All I know is at 4k, the gap between the 5080 and 5090 is going to be monstrous. I am expecting the 4090 to be faster than the 5080 by ~15-20% but the 5080 to still be a slight upgrade over the 4080 Super and using DLSS4 is that's your cup of tea. With AMD being non competitive and Nvidia entertaining a Titan again along with AI (!), I am not expecting a 5090ti variant. Plus the 5090 as-is is already pushing the 600w limit of the connector and sanity even taking into consideration PCIe power delivery. This node is being "Intel"'d a little bit methinks for the 5000 series. I have no idea what Nvidia is going to bring for the 6000 series but this is AMD's chance to do something....

4080 Super to even 4090 is a good boost, so extending that another 33% would make it a monster boost. Which laptop are you using for testing? VDD/VDDQ being limited to 1.1v will definitely keep you capped. Oddly (maybe this is just me), but I tend to focus more on secondary timings for AM5 than primary but both are important. I start with Trefi (due to heat and sensitivity) and work my way through secondary first. why? Dunno. I take things in swaths usually three settings at a time when able unless things work synergistically with each other like a few timings do. You can start with the traditional approach and it still works too for AM5. Before anything, I always keep my timings as loose as possible, pump up SA/SOC and see where the IMCs can go on both Intel and AMD. Find what is stable the most, then shave off voltage where able. Somewhere I end up striking a compromise which is usually a little less on the IMC/memory and tighter timings as a trade off. I am pushing for 2200/6600 but realistically, I know I'm going to end up at 2133/6400 1:1 on the 9800X3D. I was able to get the 14900KS running at 8600, but I settled for 8200 and much better timings with the TG 8200 sticks. As for WoW.... I've started some preliminary testing on the 4070 Super with these Crucial POS sticks as a baseline before I jettison them for the 48GB kit included and eventually the TG kit but Flight runs are basically capped to 99%+ because the 4070 Super just doesn't have the juice to keep up with the 14900KS or 9800X3D at all. I'm going to try some raids next but if I want to run some meaningful tests at 4k Ultra 10 RT on to get some meaningful separation, I'm going to need a 4090 again (no?) or a 5090 (yes?) 🙂 I will hesitantly give the 9800X3D some props because those "hitches" (dips) I would sporadically hit on FP runs on the 7800X3D/7950X3D were gone but I won't really know till I hit it with raids/PvP where they were nails on a chalk board. Make sure to give Orta a whirl out on AK this week to compare 4k Ultra 10 RT on to collect some data.

Looking like as predicted about a ~33% uplift from 4090 to 5090 and ~15% from 4080 to 5080 which would be about ~11-12% for the 4080 Super to the 5080. Rest of the comparisons are normal variants (non supers) vs 5000 series and if that is the case, the uplift, especially the 4070 Super to 5070 is going to be abysmal. If I had not sold my 4090 at this point, I might have held onto it. This will definitely NOT be a generational shock and awe like the 4090 over the 3090 but still 33% is nothing to scoff at..... https://videocardz.com/newz/nvidia-official-geforce-rtx-50-vs-rtx-40-benchmarks-15-to-33-performance-uplift-without-dlss-multi-frame-generation

Unless SP ranges have changed (which they may have), that's a really decent SP score for a 9950x. That's exactly what enthusiasts are doing ala Intel. They are binning for IMCs first and SP ratings second. That was my objective before even cracking the boxes on both 9800X3D and had the Amazon 9800X3D on order as a backup to test but I see 9800X3D supply is now catching up with demand. As for better fclk on single ccd, whew, tell that too my 7800X3D and 9800X3D! 🙂 My 7950X3D was the best of the three for fclk...but on average you may be right. My 7800X3D before could only do 2067 and could only do 6200 1:1 and my 7950X3D topped out at 2100 and topped out at 6200 1:1 also. My 7600x could do 2200and hit 6400. 6600 wouldn't post. First 9800X3D topped out at 2000 and couldn't do greater than 5200 stable. It was clearly a POS tested on 2x motherboards. It was an SP113 and it was pure garbage. This SP112 9800X3D so far has done 2133/6400 stable on pure auto settings and 2200 stable. Fatal Error on 2233 but 2200 is golden status and 2233 is unicorn status and would require an even more unicorn 6800 1:1 so I have 2133/6400 and I'm gunning for 2200/6600. This crucial memory is garbage for any type of decent OC. There is no headroom anywhere without it borking on TM5. In Crucial's defense, it performs EXACTLY as advertised stock no problem but it is a sub $100 USD kit sent for free eval so it becomes my new baseline testing kit that sits in the drawer. For it's price and placement I wasn't expecting anything meaningful for OC and I was not disappointed. I'm going to give the G.Skill 2x24GB 6400 kit that was part of this combo deal a whirl next but eventually my TG 8200 xtreem sticks will make their way over for some extended testing but I need my main rig now while I dial in the 9800X3D.

woot! Return window till Feb 1st. I can ride this 4070 Super to the wire! (wire = 28th). I'm still banking on my original list but I am always open to being persuaded. 🤣 "congrats...you saw yourself get owned and teabagged at 600fps....aren't you special!" --- Hopefully your mom is on the path to a speedy recover @johnksss

I would LOVE a 12 P-core only CPU for my Z790i lightning.... I watched this a few days ago and while it might be a problem it isn't the reason for the dip/chunk unfortunately. I was hoping he had some insight, but it is just afterburner power monitoring that is causing a problem whereas the problem I experienced on the 7800X3D/7950X3D was all on its own without any software running. 😞 ----- As for testing these Crucials, I know a proper trefi is a no go on them with these insanely cheap heatsinks. as soon as they hit ~60c+, they error out even at 32767. Pity as 65535 knocked off a nice ~5ns-6ns. 😞 I wish these companies made heatsinks like Teamgroup. Those things are monsters!

Doing some fclk testing and usually I drop down to ~5200, but decided to stick to 6000. 2200 pass TM5, but 2233 = Debug Fatal Error atm. This is still on iGPU. I won't get a chance to slap in the 4070 Super for a few days. Return ETA for the 4070 Super end of January. Think I'm going to start dialing in those Crucial sticks at 2133/6400 sweet spot which is where I will most likely end up for the final gaming mode for now so I can get some WoW testing in before I have to return the 4080 Super and wait to snag a 5090 or 5070/9070XT. I also think it is safe to cancel that other 9800X3D order on Amazon that will ship next later next week. The odds of scoring a better 9800X3D are slim to none methinks.

Same on settings at 4k 5080 will definitely be better than the 4080 Super if even only by 10-15% and it will have access to DLSS4 and other things Nvidia reserves for their current line up. Problem was potentially the 7900GRE. When testing a 7900XTX multiple times, running 4k Ultra 10 w/ RT on basically turned the 7900XTX into a 3080 performance wise. I had to keep RT off to get any decent performance uplift and that was with a 7900XTX. I can imagine the 7900GRE was probably performing much worse. Here was my original assessment of the 7900xtx for WoW when I reviewed the XFX 7900XTX a few years ago: ------- I had bandied about the idea of maybe getting a 7950HX3D laptop, but the X3D chunk on the desktop drove me away. I'll be curious to see if the 9950HX3D is any better but I am sure I'll get my own data sets in a few days when I start testing the 4070 Super on the 9800X3D here. Last trip I took, I used the NH55 w/ settings dialed down to 7 and it was silky smooth so that was good enough for me in the moment. I still wouldn't say no to a proper 18" behemoth though.