Question Has the consumer gaming GPU technology timeline lost it's way? Where are the AFFORDABLE 16GB VRAM cards to choose from?

[VirtualLarry]

Some possible hope for prices to come down.

Right now just seems like a bad time to buy. Lucky me that I've got a couple of 5900X CPU rigs with 64GB of DDR4-3200 in them, that need decent GPUs (at least, one of them does), to become my "main rig" to get back into some slight gaming.

I have a couple of 4K UHD TVs using as monitors over HDMI2.0 @ 60FPS. No real HDR to speak of.

Was thinking about a RX 6800 / 6800 XT, but I still can't stomach paying $600 for one. Maybe $300-350.

WTF prices are still like mining is a thing, is beyond me. RTX 4090 IS profitable to mine with, if you don't mind an ROI of like 10 years.

 

[Rigg]

When it does get here, let's hope AMD has learned the lesson from the 7900XT. Price it correctly out of the gate. Instead of getting destroyed in reviews because of the perceived value. Then dropping the price later, after the damage is done, and potential customers have already dismissed your product.

I agree but I'm not holding my breath. They've consistently launched both CPU's and GPU's the last few generations with questionable MSRP's and quickly followed up with price cuts. Trying to milk early adopters isn't a good strategy for gaining mind share if you ask me. I've said it before and I'll say it again, AMD's sales and marketing doesn't do them any favors. Cutting prices a few months after launch isn't a good look. Consistently naming your products a tier or two above the tier they actually compete against isn't either. I hoped it would get better with Hallock out the door but not much seems to have changed.

 

[Thunder 57]

Recommend never buying any of these grey market cards.

Ex miner cards that were "cleaned" and new cheap HSF placed on it by unscrupulous Chinese sellers in order to sell it as a "new" GPU.

Yea, that was a rookie move. After seemingly doing everything right with RDNA 2, AMD can't help themselves from screwing up RDNA 3. It must be really bad news if we haven't heard about N32 yet.

 

[Aapje]

It can mean all kinds of things. That they do a respin, that they have excess 6000-cards to get rid of first, that the performance is relatively low compared to the cost to make them, etc. We'll only know what's up when they get released.

I'm not getting my hopes up. I would like it to be attractive, but good chance that we need to wait for price drops or next gen.

 

[DeathReborn]

When it does get here, let's hope AMD has learned the lesson from the 7900XT. Price it correctly out of the gate. Instead of getting destroyed in reviews because of the perceived value. Then dropping the price later, after the damage is done, and potential customers have already dismissed your product.

Was it the 7000 series where they launched then increased the price by $50? That's something they should avoid too.

 

[jpiniero]

When it does get here, let's hope AMD has learned the lesson from the 7900XT. Price it correctly out of the gate. Instead of getting destroyed in reviews because of the perceived value. Then dropping the price later, after the damage is done, and potential customers have already dismissed your product.

The lesson AMD's going to learn is that regardless of performance, they just simply can't get gamers to buy very expensive cards. AMD really needs GCD scalability so they can distribute the dies between entry level and "Mid Range" and only do High End if say mining comes back.

 

[Aapje]

The lesson AMD's going to learn is that regardless of performance, they just simply can't get gamers to buy very expensive cards. AMD really needs GCD scalability so they can distribute the dies between entry level and "Mid Range" and only do High End if say mining comes back.

I think that the issue is that they don't understand how significant their deficiencies are in the eyes of consumers and thus always start off by pricing their cards too close to the price/performance of the Nvidia cards.

But I disagree that they couldn't get people to buy a card if it was way faster. If they make an 8950 XT that is sort of like SLI on a card, with multiple compute dies, and it is much faster than a monolithic 5090 Ti, then they could ask a lot of money.

PS. Mining won't come back. GCD scalability could do wonders for AI farms and such, though.

 

[fleshconsumed]

Back in the day of 3870/4870/5870 AMD essentially had 50% marketshare. They did it in the past, but it's going to be a lot harder to do it now. AMD would need to execute flawlessly against nvidia for 3 generations, and they would also have to win back the mindshare away from DLSS/FG/NVENC hype. That's going to be tough.

Back on topic, if one wants 16GB card, best options are used 6800 which are going for around $400 or Arc770 for $350. These prices aren't great, but not too horrible either considering past 2 years.

 

[Heartbreaker]

The lesson AMD's going to learn is that regardless of performance, they just simply can't get gamers to buy very expensive cards. AMD really needs GCD scalability so they can distribute the dies between entry level and "Mid Range" and only do High End if say mining comes back.

7900 XTX apparently sold much better and was more expensive than 7900 XT.

Initial 7900 XT pricing was just stupid. I think the lesson should be less stupid initial pricing.

Even in a vacuum without a NVidia competitor, 7900 XT was stupidly priced just vs the 7900 XTX.

 

[Aapje]

Back in the day of 3870/4870/5870 AMD essentially had 50% marketshare. They did it in the past, but it's going to be a lot harder to do it now. AMD would need to execute flawlessly against nvidia for 3 generations, and they would also have to win back the mindshare away from DLSS/FG/NVENC hype. That's going to be tough.

I think that AMD's best chance is if Nvidia gets completely obsessed with AI and start neglecting video cards. Although that's not necessarily good news for consumers, certainly as AMD also doesn't seem that interested in stepping up their investment.

 

[igor_kavinski]

I think that AMD's best chance is if Nvidia gets completely obsessed with AI and start neglecting video cards. Although that's not necessarily good news for consumers, certainly as AMD also doesn't seem that interested in stepping up their investment.

We would be spared having to upgrade every few years and game developers will actually get bored enough to start optimizing their code when they won't have new hardware covering up for their lazy code.

 

[Mopetar]

AMD could get back to 50% market share if they dropped a 16 GB 7800 that's a 30% performance uplift over the 6800 XT for $500 and offered similar good value follow ups to the 6700 XT and 6600 XT.

The key to market share is getting a lot of sales in the midrange and below. The only issue is that they probably won't have the greatest margins doing this and they might be more profitable matching the prices set by NVidia and only having half the market share.

Anyone who's been paying attention knows that Lisa Su will go for profit over market share.

 

[Rigg]

AMD could get back to 50% market share if they dropped a 16 GB 7800 that's a 30% performance uplift over the 6800 XT for $500 and offered similar good value follow ups to the 6700 XT and 6600 XT.

That's roughly the % uplift the 7900 XT gets over the 6800 XT. That 7800 scenario is pure fantasy.

 

[Heartbreaker]

That's roughly the % uplift the 7900 XT gets over the 6800 XT. That 7800 scenario is pure fantasy.

Unlaunch the 7900 XT, relaunch as 7800 XT. Problem solved.

 

[coercitiv]

In case anyone needs the picture:



It's a raster only line-up but the exact 30% figure was too sweet to let go

 

[DeathReborn]

HUB just dia a re-testing of the 3080 10GB vs 6800XT 16GB & it's as you were back when they launched. It does kinda tie into the VRAM issue and if 16GB is the thing we really think it is. I know that is going to get under some peoples skin but just telling you what he found out. Does AMD still have that Finewine? Is 10GB enough for typical gaming at 1440p/2160p?

Timestamps for the above video with % win/loss in the spoilers if you just want some numbers (full list of games in video of course)
11:30 for 1440p average

Spoiler

6800XT 6% slower

11:55 for 4k average

Spoiler

6800XT 8% slower

12:25 1440p No RT average

Spoiler

6800XT 2% slower

12:39 4k No RT average

Spoiler

6800XT 8% slower

 

[fleshconsumed]

Yes, 3080 is top dawg right now, the question is for how long? Steve himself said at the end of the video that 10Gb is "just enough" for right now. Give it a year and it'll probably start running out of VRAM just like 3070.

If you really care about RT and acknowledge that 3080 is living on a borrowed time then yes, 3080 is better choice. On the other hand if you don't care about RT 6800XT is the sensible choice as it'll age much better than 3080.

 

[Heartbreaker]

Yes, 3080 is top dawg right now, the question is for how long? Steve himself said at the end of the video that 10Gb is "just enough" for right now. Give it a year and it'll probably start running out of VRAM just like 3070.

If you really care about RT and acknowledge that 3080 is living on a borrowed time then yes, 3080 is better choice. On the other hand if you don't care about RT 6800XT is the sensible choice as it'll age much better than 3080.

The thing is, where 3080 is choking because settings impacting VRAM, 6800XT is also unplayable at those same settings, for other reasons (like inadequate RT HW).

If you want to postulate that eventually 3080 starts getting choked because of VRAM, I see no reason that 6800 XT wouldn't also be unplayable as it is today, maintaining their relative positioning.

But as far as buying either today, it makes little sense since the 4070 replaces the 3080 at a lower price, with more VRAM, greater efficiency, and more modern tech.

 

[DAPUNISHER]

Yes, 3080 is top dawg right now, the question is for how long? Steve himself said at the end of the video that 10Gb is "just enough" for right now. Give it a year and it'll probably start running out of VRAM just like 3070.

If you really care about RT and acknowledge that 3080 is living on a borrowed time then yes, 3080 is better choice. On the other hand if you don't care about RT 6800XT is the sensible choice as it'll age much better than 3080.

It's odd seeing him do his usual testing methodology then concluding the way he did. He admitted during the 16GB vs 8GB analysis that his trusty bar graphs can fail to tell the whole story. I contend that is happening in this revisit. He left out Resident Evil 4 and Last of Us, 2 of the key titles in his VRAM comparison. By reverting to his usual testing suite, that would indicate he did not play the games long enough to see if factors like variable rate shading were coming into play. If assets aren't loading or the quality is being reduced, you'd never know without playing the game long enough.

Tom/MLID just had another good podcast I was listening to with an effects artist for Call of Duty games. This topic is discussed. The guest was talking about how VRS will reduce the quality; that it is basically an optimization hack. He noted the effect will be that as you play the game, you will be wondering why a game level looks all muddy. His opinion is 12GB should be the minimum for VRAM.

 

[Heartbreaker]

It's odd seeing him do his usual testing methodology then concluding the way he did. He admitted during the 16GB vs 8GB analysis that his trusty bar graphs can fail to tell the whole story. I contend that is happening in this revisit. He left out Resident Evil 4 and Last of Us, 2 of the key titles in his VRAM comparison. By reverting to his usual testing suite, that would indicate he did not play the games long enough to see if factors like variable rate shading were coming into play. If assets aren't loading or the quality is being reduced, you'd never know without playing the game long enough.

TLOU is so broken, it's really not fit to make conclusion about anything using it.

He still had many of the most problematic games from the 8GB comparison, and he also noted when they ran out of VRAM here and became unplayable, so he didn't miss that.

But since that was at 4K with RT, the 6800 XT was also unplayable in those instances.

Edit:
Case in point:

Plague Tale Requiem, RT at 4K:

4K RT blows through the 3080 Buffer and turns game into a stuttering mess. But 6800 XT is only averaging 21 FPS, at those same settings, so it's completely unplayable there as well.

It's a bad faith argument to suggest that future games will impair the 3080 more than the 6800XT, when the main impairment situations involve RT, which impairs 6800 XT just as much, if not more...

 

[fleshconsumed]

4K RT blows through the 3080 Buffer and turns game into a stuttering mess. But 6800 XT is only averaging 21 FPS, at those same settings, so it's completely unplayable there as well.

It's a bad faith argument to suggest that future games will impair the 3080 more than the 6800XT, when the main impairment situations involve RT, which impairs 6800 XT just as much, if not more...

You probably misunderstood. The argument is that games will soon start blowing through 10GB buffer at higher resolutions even when not using RT. In which case 3080 will tank while 6800XT will still be playable.

 

[Heartbreaker]

You probably misunderstood. The argument is that games will soon start blowing through 10GB buffer at higher resolutions even when not using RT. In which case 3080 will tank while 6800XT will still be playable.

That argument, needs evidence, not conjecture.

Even on the HWUB 8GB comparison, done with cherry picked games to prove 8GB limitations, showed it was still mostly RT in games that caused buffer problems.

 

[SMOGZINN]

That argument, needs evidence, not conjecture.

Even on the HWUB 8GB comparison, done with cherry picked games to prove 8GB limitations, showed it was still mostly RT in games that caused buffer problems.

Right to make that claim he would need to do the testing with RT off, but then you don't get the dramatic results and it would be hard to make any strong claim. This looks to be a case of choosing the test to fit the results.

 

[fleshconsumed]

That argument, needs evidence, not conjecture.

Even on the HWUB 8GB comparison, done with cherry picked games to prove 8GB limitations, showed it was still mostly RT in games that caused buffer problems.

Same has been said about 3070 8GB two years ago and look where we are. Can't do raytracing at 1080p.

VRAM usage always goes up as resolutions increase. There is only 2GB difference between 10GB and 8GB. It's very reasonable assumption that 10GB will start having issues soon at QHD/4K resolutions even without RT.

 

[DAPUNISHER]

Right to make that claim he would need to do the testing with RT off, but then you don't get the dramatic results and it would be hard to make any strong claim. This looks to be a case of choosing the test to fit the results.

Ray tracing has been the killer app marketed by Nvidia since the 20 series launch. They emphasised this by changing the name from GTX to RTX. That makes the condition of testing without RT nothing short of comical.

But nevermind all that stuff about ray tracing everyone. In order to not create conditions where the card runs out of VRAM, we should test without the very feature that has been shoved down our throats for the last 5yrs. The foundation upon which Nvidia has based its present gaming technology and marketing strategy. The main talking point used in internet discussions to promote the brand. Seems legit.

 

[Heartbreaker]

Same has been said about 3070 8GB two years ago and look where we are. Can't do raytracing at 1080p.

VRAM usage always goes up as resolutions increase. There is only 2GB difference between 10GB and 8GB. It's very reasonable assumption that 10GB will start having issues soon at QHD/4K resolutions even without RT.

Game developers don't increase game VRAM requirements linearly over time.

They increase it in jumps when consoles jump up in VRAM capacity, because AAA games these days are mostly console ports.

Now we are in an era when Consoles have >8GB of VRAM, which can be problematic for cards with only 8GB (mostly with RT on).

It's really only going to get problematic for 10GB when we get next gen consoles with >10GB of VRAM.