Question are video card prices headed down yet?

[moonbogg]

Seems a little pricey for N21 Refresh... I was thinking something closer to the following... Note: I only put like 5 min of thought into this so cut me some slack lol

NVIDIA	AMD	Price	Raster Performance
	7950XT (Full N31)	$2499	>2x RTX 3090/6900XT
RTX 4090 (Full AD102)	7900XT (Cut down N31)	$1999	~2x RTX 3090/6900XT
RTX 4080 (Cut down AD102)	7800XT (N32)	$1299	~1.6x RTX 3090/6900XT
RTX 4070 (AD103)	7800 (Cut down N32)	$899	~1.2x RTX 3090/6900XT
RTX 4060Ti (Cut down AD103)	7700XT (N33)	$699	~RTX 3090/6900XT
RTX 4060 (AD104)	7600XT (Cut down N33)	$449	~RTX 3070/RX6700XT

Too cheap. RXT 3070 equivalent performance should cost $700 and RTX 3090 equivalent should cost $1500. This would be a significant price reduction for which gamers would be grateful. If they improve price/performance, then everything will be vaporware and end up on ebay for 3X msrp. If you want products to stay in stock for more than 5 seconds, you can't offer people value anymore. The masses have demanded and voted for garbage value, so that's what I hope they get. I want Nvidia to charge $800 for a 4060 and offer nothing below. That's my wish.

 

[Saylick]

Too cheap. RXT 3070 equivalent performance should cost $700 and RTX 3090 equivalent should cost $1500. This would be a significant price reduction for which gamers would be grateful. If they improve price/performance, then everything will be vaporware and end up on ebay for 3X msrp. If you want products to stay in stock for more than 5 seconds, you can't offer people value anymore. The masses have demanded and voted for garbage value, so that's what I hope they get. I want Nvidia to charge $800 for a 4060 and offer nothing below. That's my wish.

... People with deeper pockets than I always ruining it for the average Joes who want something reasonable without paying out the wazoo...

 

[Glo.]

Will they cut the PCIe on the 7500XT to x4 or actually learn from their mistakes?

Why would it be?

Its the same N23 as RX 6600 and 6600 XT.

 

[moonbogg]

... People with deeper pockets than I always ruining it for the average Joes who want something reasonable without paying out the wazoo...

It's people with deeper pockets, but it's also people willing to bury themselves in debt for a gaming toy they can't afford. The hobby has always had people with relatively deep pockets; 4 way SLI users etc. During those times, there were good value products, but not anymore. This whole gaming thing was only ever exciting because just about everyone could participate. You see little forum participation among those with $1500+ GPUs. It becomes a small, exclusive club. It is what it is, but it's sad to see the vast majority get completely priced out of the enthusiast products. Most enthusiasts have been forced down to the midrange products now. People who used to spend 5-$600 on a flagship are starring 60-class cards in the face now wondering if it's even worth it anymore. I say it's not.

 

[jpiniero]

N22 is going to be used for X600 SKUs when RDNA3 releases, N23 will be used for X500 SKUs when RDNA3 releases.

For 7500 XT, with RX 6600 XT performance I'd say we can expect at best 249$ MSRP

Sub-$300 is dying/dead me thinks.

 

[Saylick]

Sub-$300 is dying/dead me thinks.

I think so, too.

*hugs PS5*

 

[lakedude]

I heard some tariffs were about to be discontinued, putting downward pressure on prices...

 

[biostud]

Too cheap. RXT 3070 equivalent performance should cost $700 and RTX 3090 equivalent should cost $1500. This would be a significant price reduction for which gamers would be grateful. If they improve price/performance, then everything will be vaporware and end up on ebay for 3X msrp. If you want products to stay in stock for more than 5 seconds, you can't offer people value anymore. The masses have demanded and voted for garbage value, so that's what I hope they get. I want Nvidia to charge $800 for a 4060 and offer nothing below. That's my wish.

3090 is only in that price bracket because of the 24gb vram, the 3080 is not that much slower but is priced much lower.

 

[biostud]

Maybe there is a large market for super expensive video cards, but wouldn't it make most sense to target the price range where they can sell most volume?

Sure they will make $1500 video cards, but neither nvidia or AMD would survive if they didn't have competitive products in the lower price ranges as well.

 

[Glo.]

Sub-$300 is dying/dead me thinks.

On smaller nodes - yes, there will not be sub 300$ GPUs.

But right now - they will still exist, especially considering that N23 and 22 and AD107 and 106 are still on N6 from TSMC.

N5, N4, N3, etc - thats where it starts to become hard on manufacturing costs.

 

[igor_kavinski]

I don't know why they don't just give up on the huge monolithic dies and just start doing SLI between two or three smaller dies on the same card. Would really help them in lowering the defect rate. Sure, may not be very efficient from a performance standpoint and will introduce SLI-related issues. But still, why do the hard thing when you can do easy? Continued refinement may even end up making SLI perfect due to them having to iron out the kinks from generation to generation.

 

[DrMrLordX]

Maybe there is a large market for super expensive video cards, but wouldn't it make most sense to target the price range where they can sell most volume?

Sure they will make $1500 video cards, but neither nvidia or AMD would survive if they didn't have competitive products in the lower price ranges as well.

Good luck with that.

 

[biostud]

Good luck with that.

With what?

 

[DrMrLordX]

With what?

With AMD or NV lowering their prices so they can push more volume. They're already learned to control volume and raise prices.

 

[jpiniero]

On smaller nodes - yes, there will not be sub 300$ GPUs.

But right now - they will still exist, especially considering that N23 and 22 and AD107 and 106 are still on N6 from TSMC.

N5, N4, N3, etc - thats where it starts to become hard on manufacturing costs.

N6 is not a cheap node. I mean it's not like AMD is selling a card under $300 right now other than the dastardly 6500 XT, and I imagine the 6500 XT is not long for the market. Maybe the 6400 will get a retail release if they want to get rid of Navi 24 quicker but once it's gone it's gone.

The only thing I will say is that maybe I am being too pessimistic on AD106/7's performance despite the small SM increase. But if it forces AMD to hit higher, they will adjust production accordingly and won't have cheaper products. Also I bet you will only see AD102 and AD103 this year.

 

[maddie]

Maybe there is a large market for super expensive video cards, but wouldn't it make most sense to target the price range where they can sell most volume?

Sure they will make $1500 video cards, but neither nvidia or AMD would survive if they didn't have competitive products in the lower price ranges as well.

I agree, but this is an unpopular view.

You need to amortize R&D over as many units of production as possible. If lower cost, AKA biggest volume cards disappear, then the costs of the remaining production rises, maybe by a lot depending on volumes. Intel and AMD will have an advantage due to IGPs using the same tech in all their CPUs. Boutique manufacturers in many consumer fields have mainly ended up merging with mass producers. R&D becomes too expensive in the long term for smaller production runs per generation.

 

[biostud]

With AMD or NV lowering their prices so they can push more volume. They're already learned to control volume and raise prices.

But ultimately, is that what gives them the highest profit?

And if there seems to be any kind of agreement between AMD and Nvidia, wouldn't anti competitive government institutions get involved?

If mining demands for video cards keep the prices high, wouldn't you want to sell as much as possible at a high price?

And then there is the joker with Intel joining in.

 

[DrMrLordX]

But ultimately, is that what gives them the highest profit?

There's more in play than what gives the highest profit. Both AMD and NV have to balance potential market demand against available silicon. AMD in particular shares wafers across a wide range of products. They currently share N6 with Rembrandt and upcoming dGPUs, for example. In the previous generation, they relied on N7 wafers for enterprise, mobile, desktop, mGPUs, and dGPUs. Maybe they will have a little more flexibility with legacy products on N7, GPUs and mobile on N6, and server/workstation/desktop/almost-next-gen mobile on N5. But I wouldn't count on it getting too much simpler. Rembrandt may move some serious volume. In a competition between premium mobile products much-demanded by OEMs, enterprise dGPUs, and gamer dGPUs, it'll be interesting to see how AMD splits up their available wafer supply. Especially when we'll have no reliable way to track sales of enterprise products.

There are other components that also make it difficult to project exactly how many cards they can send to market. Power delivery components and RAM would be the big two that come to mind, though there may be something else obscure or trivial making it difficult to manufacture more cards to drive volume sales.

In the end, it's easier and more-predictable to leave the market supply-constrained. Especially when your competition is doing the same thing. It's almost like they're cooperating . . .

And if there seems to be any kind of agreement between AMD and Nvidia, wouldn't anti competitive government institutions get involved?

They ought to, if that were shown to be the case.

If mining demands for video cards keep the prices high, wouldn't you want to sell as much as possible at a high price?

You get flooding on the secondhand market which can (sometimes) make it hard to sell new product. Especially midrange product if you've been cranking out high-end cards from the previous generation. That being said, it looks like mining may enter a period of decline, so hopefully that will not be a factor moving forward.

And then there is the joker with Intel joining in.

Battlemage may be a force. Alchemist isn't looking like much. The projected shipping volume is too low to have a serious effect on the dGPU market.

 

[Glo.]

N6 is not a cheap node. I mean it's not like AMD is selling a card under $300 right now other than the dastardly 6500 XT, and I imagine the 6500 XT is not long for the market. Maybe the 6400 will get a retail release if they want to get rid of Navi 24 quicker but once it's gone it's gone.

The only thing I will say is that maybe I am being too pessimistic on AD106/7's performance despite the small SM increase. But if it forces AMD to hit higher, they will adjust production accordingly and won't have cheaper products. Also I bet you will only see AD102 and AD103 this year.

The node itself is relatively cheap, around 9-9.5k per wafer, which equals to 25$ per die of 6500 XT.

What is high in prices is shipments, PCBs and - especially - VRAM.

Also AD106 and 107 are just Ampere GPUs with different SM count, higher amount of L2 cache and higher clocks.

But essentially the architecture is exactly the same as Ampere.

 

[jpiniero]

Also AD106 and 107 are just Ampere GPUs with different SM count, higher amount of L2 cache and higher clocks.

But essentially the architecture is exactly the same as Ampere.

That's basically what Ada is in general. SS8 is such a bad node quality wise that you have to figure that even N6 should allow for huge clock speed gains and maybe the cache will help sustain performance. But we will see. Now I am assuming that AMD is only doing super high end products that isn't just a minor refresh because they fear a mining collapse which is understandable.

 

[Frenetic Pony]

Sub-$300 is dying/dead me thinks.

Nah, it'll be back. The assumption here that the chip shortage and crpytbros are eternal is weird. Every cutting edge silicon company has said the chip shortage should ease by the middle of this year, and the cryptocalypse may already be underway.

Temporary shifts in the market shouldn't make any long term changes. By the end of year everything could be super cheap relative to the past two years.

 

[jpiniero]

Nah, it'll be back. The assumption here that the chip shortage and crpytbros are eternal is weird. Every cutting edge silicon company has said the chip shortage should ease by the middle of this year, and the cryptocalypse may already be underway.

Temporary shifts in the market shouldn't make any long term changes. By the end of year everything could be super cheap relative to the past two years.

Both companies have gotten addicted to high ASPs on top, plus AMD can always shift wafers to other products (read: Epyc) if the alternative is cutting prices.

You might be able to buy a used card for cheap.

 

[Aapje]

Nvidia and AMD run a severe risk of cannibalizing this market if this lasts too long and lots of gamers shift to console gaming or Pokemon. The people who are buying cards for $800+ are usually not the young kids who are getting into gaming. It makes way more sense for those kids to buy a PS5 for just $500 (or even a Quest for $300).

AMD does dominate the current gen of consoles, so they could profit from people moving more to consoles, but it is very risky to put all your eggs in that basket, because missing out on a future contract with Sony/MS would mean a loss of over 10 million units a year. If they lose both the contracts for the PS6 and the next gen XBox, they are looking at losing 20 million units of sales a year (is Nvidia looking at developing their own CPU's because they want to be able to offer a full console solution, to be able to compete with AMD better?).

The PC market is relatively stable and it's good business to nurture a more reliable market, so bad fortune doesn't bankrupt your company as easily. Mining is not a stable market, which is why we've seen the LHR cards in the first place: Nvidia wants to ensure that actual gamers end up with cards.

Also, these huge margins are not going to be sustainable for the long term. If production capacity goes up relative to demand, AMD and Nvidia will compete for market share, which will drive down the prices. Also, companies like Intel are now greatly incentivized to invest in GPU's.

 

[Aapje]

I don't know why they don't just give up on the huge monolithic dies and just start doing SLI between two or three smaller dies on the same card. Would really help them in lowering the defect rate. Sure, may not be very efficient from a performance standpoint and will introduce SLI-related issues. But still, why do the hard thing when you can do easy? Continued refinement may even end up making SLI perfect due to them having to iron out the kinks from generation to generation.

They don't because it's simply not worth it.

They now have a better solution with multi-die processing units. The problem in the past was that the connection between the dies introduced delays and had other problems. They've only recently come up with a better solution, which is why multi-die is only now taking off. Still, it only makes sense for the bigger chips for now. Don't expect it on the low end for now.

 

[jpiniero]

AMD does dominate the current gen of consoles, so they could profit from people moving more to consoles, but it is very risky to put all your eggs in that basket, because missing out on a future contract with Sony/MS would mean a loss of over 10 million units a year. If they lose both the contracts for the PS6 and the next gen XBox, they are looking at losing 20 million units of sales a year (is Nvidia looking at developing their own CPU's because they want to be able to offer a full console solution, to be able to compete with AMD better?).

This console gen is going to last a long time, longer than the typical cycle which is normally 7 years. There's likely to be "Pro" models at some point too.