The RTX 4090 Looks SICK

[AlexTheGreatish]

Nvidia just announced the RTX 4090 and 4080s. Overall Ada looks great... but Nvidia left out some crucial details that have me concerned.

 

 

 

[Luscious]

Thanks but I'll pass. Rather have four cards 1 slot thick than one card 4 slots thick 

 

 

 

[Shimejii]

it pretty much only looks sick because the rest of the lineup was purposely gimped for a year. 1.76x performance is good, its just the rest of the lineup that is currently lacking for an increased price. 

 

As it currently stands, 4090 probably costs around 850$-1000$ per card to make, so i dont blame them for that price. Its the behavior of calling a card a name it really shouldnt be, and they have been doing it for years now. that 4080 12gb should not have the 80 name.

[Pitboy64]

Just imagine the prices for these in Canada.
Not just the exchange, but the added Canadian tech hardware tax, in evidence anytime we know the American dollar price.

[Ravendarat]

35 minutes ago, Shimejii said:

As it currently stands, 4090 probably costs around 850$-1000$ per card to make

What information do you use to reach that conclusion? I assume you are talking about Nvidia themselves and not a board partner?

[silentdragon95]

That thing is going to be €1949 over here (with taxes obviously). That's just ridiculous. And don't even get me started on the "RTX 4080 12 GB" (read: RTX 4070) that is €1099. Sure, double the performance for the next generation is cool and all, but that's also twice the price of an RTX 3070. At this rate, no thanks.

 

I sincerely hope that AMD isn't going to match these prices and that Intel doesn't cancel ARC and gets their software issues in order. We need a third competitor in the market now more than ever.

[ThousandBlade]

Remember when top tier GPU was $500 dollars??

[starsmine]

46 minutes ago, Ravendarat said:

What information do you use to reach that conclusion? I assume you are talking about Nvidia themselves and not a board partner?

17k n4 wafer (was a tad over 16k in 2020 for n5, but TSMC has raised prices twice since then somewhere in 10-20%)
608mm^2 die
13 dollars per gig memory. 
 

= 330 dollars per die at .08 defect rate, so nvidia wanting there cut will be selling those to AIBs for 500 dollars per die
+ 312 dollars for the ram.

Cooler costs, shipping costs, and VRM costs are all higher then the 3090. 

AIBS literally are selling 3090s at 100s of dollars of loss at 1000 dollars. 

I honestly think its priced lower then expected due to the increased manufacturing cost vs last gen.
And I think its because RDNA3, which should be very close in performance, will be so much cheaper to manufacture in comparison. Much higher yields due to chiplet, much less die space on n5, with most of it on n6, and with significantly less power requirements leading to cheaper coolers, cheaper shipping, and cheaper vrms. 

[Ravendarat]

5 minutes ago, starsmine said:

17k n4 wafer (was a tad over 16k in 2020 for n5, but TSMC has raised prices twice since then somewhere in 10-20%)
608mm^2 die
13 dollars per gig memory. 
 

= 330 dollars per die at .08 defect rate, so nvidia wanting there cut will be selling those to AIBs for 500 dollars per die
+ 312 dollars for the ram.

Cooler costs, shipping costs, and VRM costs are all higher then the 3090. 

AIBS literally are selling 3090s at 100s of dollars of loss at 1000 dollars. 

Id be kind of surprised if Nvidia isnt going for a higher margin on that die, but I admittedly never cared to look at their financials to see what kinda margin the run as it it would take some serious breakdown to get to a reasonable guess.

[starsmine]

3 minutes ago, Ravendarat said:

Id be kind of surprised if Nvidia isnt going for a higher margin on that die, but I admittedly never cared to look at their financials to see what kinda margin the run as it it would take some serious breakdown to get to a reasonable guess.

  

14 hours ago, starsmine said:

https://www.jonpeddie.com/news/evga-wont-offer-nvidia-next-gen-series

OOF 

 

yes you are correct, nvidia wants and gets more margin, I just would hope they are trying to get some good will back from AIBs after screwing them over on ampere.

Or not, its nvidia, they are known to be dicks. 

[Luscious]

1 hour ago, ThousandBlade said:

Remember when top tier GPU was $500 dollars??

I can remember when a top tier WATER COOLED GPU was $800 - EVGA hydro copper 980Ti at launch in 2015.

[DoctorNick]

2 hours ago, ThousandBlade said:

Remember when top tier GPU was $500 dollars??

Yes 8800 ULTRA baby!

[Solarisfire]

Is anyone else worried that DLSS 3 is just the same thing as "Motion Smoothing" that we've seen in TVs for a while now? Seems to just take rendered frames and make up (albeit with fancy AI algorithms) totally new frames that didn't exist before... And then they're counting those frames towards the FPS that the card is spitting out on all of the performance graphs in the announcement... Just doesn't seem right to me, but I guess we need to see this thing work before drawing a conclusion.

 

EDIT: I really should have watched the video first...

[pureexe]

4 minutes ago, Solarisfire said:

Is anyone else worried that DLSS 3 is just the same thing as "Motion Smoothing" that we've seen in TVs for a while now? Seems to just take rendered frames and make up (albeit with fancy AI algorithms) totally new frames that didn't exist before... And then they're counting those frames towards the FPS that the card is spitting out on all of the performance graphs in the announcement... Just doesn't seem right to me, but I guess we need to see this thing work before drawing a conclusion.

Yes, i'm worry about it. 
If image going to make an image smooth like a butter, i going to reduce graphic setting instead of turn on DLSS.

[Solarisfire]

Rendering a frame completely decoupled from the CPU has to introduce input lag of at least 1 frame no matter what... Maybe that matters less the more frames you have as each frame persists for a shorter amount of time, but a lot of gamers still use 60hz monitors so it will be input lag for no benefit. But no matter what game I'm playing, competitive or otherwise, I don't want more input lag.

[iLikeBananas]

The 3070TI is 669€ in Germany right now, hope the release of 40-Series cards will further drive the price down. But with how badly priced the 4080 is, I can totally see lot's of people buying 30-Series cards instead, driving those prices up.

[LapsedMemory]

Disclaimer - I have no actual insider knowledge, so the following is pure speculation: 

 

Simple supply and demand, just not in the way it's usually attributed.  

 

There is not currently a huge demand for these cards, so the price doesn't seem justified at first glance.  

There is not currently a huge supply of these cards, so the price doesn't seem justified.

 

What there is, however, is a metric shiateton of 30 series inventory sitting on shelves that has to be moved with some form of profit margin for the board partners, or NVIDIA won't have any board partners left.  NVIDIA was between a rock and a hard place here.  They had to release the 40 series because they've already ordered way more chips than they will need now that the crypto market tanked, but if they price it too low, the 30 series market will tank, which is already struggling due to an influx of used cards on the market.  What's the only viable solution?  Jack the prices on the 40 tier, make a higher margin on people who just have to have 'latest and greatest', and ride out the inventory issue.  Prices should be reduced once the 30 series inventory is dried up.  

 

The other factor in play here is that NVIDIA possibly pricing this first wave for professional environments.  There are very few consumer level applications that actually NEED the performance boost of the 40 over the 30.  Hell, the average GPU in steam's hardware survey is still the GTX 1080, and most people are still running 1080p@60hz.  Corporations that need the extra graphics power will snatch these up as "cost of doing business" as the extra processing power will just factor into a longer ROI.  

 

I'm not saying I like what NVIDIA is doing, but from a business perspective, it makes sense.

[PikachuEXE]

5 hours ago, Luscious said:

I can remember when a top tier WATER COOLED GPU was $800 - EVGA hydro copper 980Ti at launch in 2015.

I am still using 980Ti (for 1080p)

 

Gonna switch to AMD for my next card (just not sure if 7xxx or 6xxx)

[Vishera]

7 hours ago, ThousandBlade said:

Remember when top tier GPU was $500 dollars??

Bought my HD 6970 for $400 in 2011.

[Dedayog]

9 hours ago, Luscious said:

Thanks but I'll pass. Rather have four cards 1 slot thick than one card 4 slots thick 

 

 

 

Nice brag.  Don't forget the jerking off emoji next time.

[Solarisfire]

If you guys have a 4090 in hand... Please include an analysis of things like head-bob and camera shake when reviewing DLSS3 in the initial review. From the footage I've seen so far it looks like the technology reduces those atmospheric effects that developers build into a game.

[DietzDiet]

the beginning

[Dedayog]

10 hours ago, AlexTheGreatish said:

Nvidia just announced the RTX 4090 and 4080s. Overall Ada looks great... but Nvidia left out some crucial details that have me concerned.

 

 

 

Oh and for aesthetics, it's a bland card.   Not exciting in any way.

 

Who made the curvy one?  Zotac?

[Solarisfire]

6 minutes ago, Dedayog said:

Oh and for aesthetics, it's a bland card.   Not exciting in any way.

 

Who made the curvy one?  Zotac?

MSI announced theirs and they look gorgeous...

MSI reveals its custom RTX 4090 and RTX 4080 series GPUs | OC3D News (overclock3d.net)

[Ubersonic]

If Nvidia are quoting 2-4x performance improvement then you can bet a kidney that it will be an absolute max of 2x, as usualy they will prob be quoting DLSS/RTX numbers most people don't care about to make their numbers look better.