Nvidia Ampere Discussion [2020-05-14]

Besides mass murdering everyone who can't help themselves and wait a couple of weeks before purchasing something which will be sold for two years what measures can they take?

 

idk, but from the Nvidia-page (good luck opening it atm):
„To combat this challenge we have made the following changes: we moved our NVIDIA Store to a dedicated environment, with increased capacity and more bot protection. We updated the code to be more efficient on the server load. We integrated CAPTCHA to the checkout flow to help offset the use of bots. We implemented additional security protections to the store APIs. And more efforts are underway.“

 

Their measures were apparently letting people know this time beforehand that there's only limited supply

 

It was a bit like challenging 4chan to a DDoS.

 

Did the Nvidia store limit the number of cards per order?

 

Yeah, 1 per customer at least for 3080's

 

They'd likely have needed to do a 3 month+ aggressive inventory buildup before launching to offset the initial demand. Essentially you'd be looking at something more along the lines of a console product roll out (in terms of everything from announce to availability), however given the market/product differences I don't see something like that being doable due to a variety of factors. Of course ultimately what they'd essentially really be doing is saying everyone has to wait 3 months+ compared to when they could've potentially gotten one.

On a semi-related note interestingly something "negatively" (depending on how you view it) affecting this launch is that there was no staggered AiB custom availability and opted for a "better" reference offering. Having a basic reference only model (or more expensive I guess) from Nvidia only (just rebranded for the AiBs) would've both helped launch inventory (much simpler logistics) as well as blunted demand (as people would've wanted to hold off for customs).

I just managed to run both versions (since I had a copy of the older one) through a PDF compare and it does look like the Titan RTX spec fix and the GA104 clarification/details were the only two changes.

 

It didn't help consoles to stock for months. At least I have been unable to get a preorder in for ps5 :/

 

I really doubt that we're looking at three months of satisfying the initial demand. It'll more than likely be resolved by the end of October completely.

It's also very possible that many of those who couldn't get a 3080 right now will instead wait for 6900 and 3080 20GB cards since we're basically about a month away from the former at least. So the demand will die down a bit over the next couple of weeks IMO.

 

Cool, apparently some other stores were not limiting per customer.

 

You'd need to account for that if GPUs did switch to a console type roll out it'd only been done if it's universally for both vendors and their entire lines, so essentially everything would be pushed back by the artificial ramp up period.

Also that type of launch roll out in itself would increase launch demand (or preorder demand) due to the longer build up (hence why consoles would really need a 6 month+ build up vs. the maybe 3 month+ build up they have now to solve their issue), so effectively demand would be even higher than now but offset more so by the longer supply build.

 

Same for 3090. I could just see that after putting it in the basekt and before their shop-system refused to let me advance through the purchasing process any further.

That's my impression too.

And consoles were at a completely different price point for a whole system, not a component only.

 

https://www.techpowerup.com/review/asus-geforce-rtx-3090-strix-oc/38.html

 

Haven't even managed to get through half of the available reviews yet, but Computerbase comments a bit about the more than 10GB VRAM being beneficial in today's games ,by pointing out the 3090 has a repeatable 22% higher frametimes compared to the 3080 in Ghost Recon Breakpoint.
https://www.computerbase.de/2020-09.../2/#diagramm-ghost-recon-breakpoint-3840-2160

 

With this kind of performance, I'd rather revise my statement from above, which was "3090 or bust": A 20 GByte 3080 with the 3090 cooler would be nice. For a three-digit price.

 

Since most mainstream video cards still have 8GB or less VRAM, it's unlikely that the majority of games be using more than 8GB of assets at the same time. So, in a way, 10GB is probably fine for the foreseeable future.
On the other hand, as more games are now cross platform, consoles now have a considerable impact on PC game development. Since next gen consoles have 16GB RAM, it's possible that at least some games, especially those designed for consoles, could be using more than 10GB of assets at the same time. Of course, textures are quite easy to handle, as it's easy to lower texture resolution to save space. It's just that sometimes you probably can't use the highest texture settings on a flagship card, and that can be disappointing.

As CarstenS said, personally I'd like to see a reasonably priced 20GB 3080 or 3080 SUPER for an upgrade (I'm using a 2080 Ti right now so 3080 is not a great upgrade for me).

 

https://www.igorslab.de/en/what-rea...tabilities-of-the-force-rtx-3080-andrtx-3090/
To sum it up:
Board partners got surprised by record smashing power consumption, circuit designs (within specification) working perfectly well for smaller GPUs don't work for the 3080/3090. Used the wrong capacitor types around the power pins of the GPU. Several custom designs (like ZOTAC Trinity 3080, Gigabyte, MSI) are pretty much broken by design, and won't run stable unless parts are replaced or boost clocks are drastically limited.

Apparently not yet decided who is to take the fall, as the faulty choice of components was apparently permitted by NVIDIAs specifications.

 

That's all good, but these issues affect Founders Editions too, not just AIB cards

 

Yes, and also reported cases for MSI and EVGA models which used the same parts list as the Founders Edition. Albeit not reported at the same frequency as models using POSCAP on the NVVDD rail too. May be same issue, but with a higher error margin, may be unrelated issues. Difficult to tell apart from PSU problems.

One unknown user (assuming it's not fake, Igor didn't link to a source) apparently got a Zotac GPU stable at boost clocks by replacing POPSCAP on NVVDD by MLCC group, which provides a strong point in case. Assuming that user had sufficient knowledge about electrical engineering not to fall for PSU issues.

 

Some comments from buildzoid on the capacitor article by igor's lab