Separate names with a comma.
Discussion in 'nVidia Flavor' started by Kyle_Bennett, Nov 9, 2018.
I'm really more curious as to why the artificating on these cards looks like space invaders.
First, is anyone seeing this happening on 2080 (non-Ti) cards in big numbers?
And also, I think Nvidia really did try to build it for overclocking. I feel for them in this respect: The power delivery really is quite the heavy duty overkill. Whether or not they officially let us pass 130% power targets right now or not the PCBs are really, really good.
This is probably going to be either a GDDR6 failure or some type of GPU core early mortaility they just didn't see coming before mass production.
Based on the type of garbage on the screens I'm kind of still leaning toward GDDR6 issues. My Gigabyte card has pretty poor memory overclocking compared to many people. I keep hearing how many people get +1200 max out on the slider but I know exactly where my artifact limit is. It's +845. And I've seen quite a few other people top out in the +800 range. Such a massive difference in RAM capability makes me wonder how much variation or change to the process is happening on the RAM front.
I think... if was purely GPU Nvidia would have an easier time identifying it. And they would have halted sales or initiated a full recall by now to stop the pain from increasing.
Any time you see those dot pattern or color grids on the screen the first thing you should think is something to do with the memory. I'd be expecting to see a lot more just straight up black screens and game crash messages and blue screens and dead no video issues if it was always the GPU going.
Not to mention that it’s very possible that the current RTX line will not be optimal for performance with this technology. This wouldn’t be the first time Nvidia’s touted a cool technical feature that no one uses because of the titanic amount of system resources it requires (see: Gameworks).
We won’t know until actual benchmarks are released in actual gaming situations, but if I were a betting man, I would say that maybe the RTX 2080 Ti is almost adequate, but below that, don’t bother, and you can’t SLI the 2070 in the event you buy one and realize you would need extra horsepower to use the feature.
im glad i was able to pick up an evga 1080ti black for 649 last month at microcenter. i cant find a new 1080ti now for under 800 bucks. if i was kyle i would take that 2nd working 2080ti and repeat what he did to the first one to see if it fails also. maybe it is the brief oc that eventually kills it. would be good info to know.
It would be a dot on the landscape if the OC hastens its demise because many have failed immediately or after a while of use with no OC.
The problem doesnt look to be OC related.
Quite a bit of bullshit and crystal-ball gazing going on in this thread...
It's too early to know anything, other than "some people's cards died.." If we can get actual failure rates from a manufacturer, then we will have something to talk about.
"Kyle got a bad card omfg!" LOL.. as much hardware as Kyle see's, statistically he was probably overdue for getting a defective item.
It sucks of course. I was hoping [H] would have some nvlink performance benchies at some point.
Love that pic. The turbos are bigger than the engine itself. ….
Turing that dies... hrm TurDing? Because they are crapping out.
Kyle_Bennett did you have 2 cards fail? or 2 failures on 1 card?
Glad I passed on the initial batch of cards. Guess I'll hang on to my 1080 Ti a while longer until this mess is sorted out!
Heres nvidia's idea of a vapor chamber system:
Small dinky fin stacks that are tightly clustered together.
Now, here's sapphires off the 290 vapor x.
Larger fins, space between them for airflow, heatpipes as well as a vapor chamber. Nvidia's implementation just looks anemic and half assed in comparison.
Can't really compare the HBM solution to the GDDR solution- they are different heat densities. As to fin spacing, within a reasonable range it's just a tradeoff. The tighter spacing with more fins has more surface area and thus needs less fan speed to remove the same amount of heat, but it will also likely be noisier at the same fan speeds. Tune one way or the other- I'd bet that the tighter spacing with more fins is a more expensive process, but at the same TDP and with the same fans would result in lower noise.
I agree it's easy to echo chamber on this. A half dozen noisy people on a few forums seem like an army of failures but if you've been around the tech forums long enough you get a sense of real vs. imagined issues. I think it's safe to say we at least have failure rates that are above the expected one or two percent for hardware.
It's entirely possible the problem is a RAM voltage issue that could be fixed by firmware/drivers. I'd recommend against RMA'ing cards at this point unless they're totally unusable... though it's conceivable just due to varying tolerances in the DDR parts you could end up with a card immune to the problem.
On the nvidia forums it sounds like the FE cards aren't being direct replaced by RMA anymore and are being 'repaired' instead which tells me they are receiving way more cards then 'normal'. As to what the 'repair' is we have no idea at this time. I believe nvidia 100% knows what the issue is at this point but are keeping things hush hoping the problem eventually fades away given enough RMA's.
To clarify, the issue was on my DDR4 memory, not VRAM. System seems stable now.
In that case it'd be useful if people took good pictures of their card before sending them in to see if any mods are visible or parts are replaced.
Yeah, this is what I was thinking too...
The fact that all 3 of my RMAs were within hours of contacting NVidia, and now there are threads on their forums of people waiting over a week to get one makes me really believe that something is seriously wrong.
I had an old 400 series card die via its memory failing. Granted the card ran stupid hot for what it was, but it was artifacting and blue screening like all the descriptions of the 2080ti series posts. It never got "space" invaders types of artifacts, but did get weird snow and items displayed on the desktop.
I feel like dissecting the remains of that card to see if it had any scorch marks which could be clues to what to look for on the 2080 cards.
What clocks are you running on the memory and core?
I re-watched the Nvidia video and the guy specifically mentioned "core" clocks when talking about overclocking and didnt seem to reference memory speeds at all. I have not had any issues with my 2080ti, and while I have overclocked it right out the gate, I started with core clocks and only did minor adjustments to memory speeds. Maybe this is the reason the auto-tuners dont touch memory speeds?
According to him, only 1 card fail so far.
More likely that the demand is so high, that they are keeping little stock on hand for RMA's, in an attempt to fill orders as fast as they can.
It seems anytime I think to check RTX card stock, all I see is sold out or [Auto-Notify]
Kyle you have my condolences for your failed card ..i was planning to buy an RTX 2080 \ or ti card by X mas ..but now i think i'll wait ..and glad i voted do not sign the NDA
[H] reviews are what i use to make my ...... check out this hardware = to buy decisions as i have always found this...... [H]ardOCP = the best tech site on the internet .. keep up the great work
and i hope Nvidia sends you a working card from your RMA that has zero issues ..i'll be watching for what happens next ..before i buy .. who knows i may go AMD if the rtx series bombs
Good on the community for keeping Nvidia honest. Hopefully it helps out a few poor sods in their RMA's.
these cards were already a hard sell for most people (pricing and too early ray-tracing) but now you add these performance/hardware issues and Nvidia might be in for some hurt...worse then the GTX 970 VRAM marketing issue
How about "Torturing!"
I was thinking of ordering the FE or Zotac when they were in stock before the Evga FTW3, my girlfriend was like "don't do it", my boss at work was like "stay the course", glad I listened.
On a side note my girlfriend was spamming F5 and stalking Jacob Freeman's twitter then messaged me during the 4 minutes that it was in stock so I could grab it.
I went to the Nvidia page, and there is no mention of a 2080Ti for sale. If you type that in, it will bring you to a 1080Ti page with a 'notify me' button.
You're checking the wrong area. https://www.nvidia.com/en-us/geforce/graphics-cards/rtx-2080-ti/
Thanks. I can't find an RTX 2080Ti to buy to save my life.
That was the last card I upgraded with a bios upgrade.
Careful - there was nothing about that in Huang's presentation, and his claims were hardly modest. Or are you just rationalizing?
Just submitted RMA for my Asus 2080ti, crash to desktops and now blue screens also happening multiple times a day.
This is widely reported issue now. Not just 10 users. Its a 1200 dollar card. Nvidia certainly could have done better at it. May be they were so overconfident that they wanted to milk the desperate fanboys. What you need to realize is almost all cards are dying the same way. That right there screams one thing, its the GPU not anything else.