MSI GeForce RTX 3090 Suprim X

[R0H1T]

Now this is not a straight comparison. They're comparing Samsung's 7nm to tsmc's 7nm but it should give you a decent idea of just how much better Tsmc's node is.

You know that's not how node comparison's work, or do you? While you have physical characteristics that you can direct measure, then electrical characteristics & finally the most important part of the puzzle ~ the uarch. Just because you say TSMC 7nm is "much more better" than SS' 8nm doesn't make it a fact that it'd be also much better for Ampere, unless you have the same GPU/chip/uarch made on the two nodes. The closest comparison IIRC was the VII & Vega64.

Anyone claiming otherwise is just guesstimating! So waiting for you to provide evidence about how superior (or inferior) one is wrt to the other.

Take zen3 for instance, with a (major) tweak in zen2 you not only get higher IPC but also much higher clocks on the same node. Can you now claim that zen3 would clock just as high on 7nm+ or 7LPP, perhaps 6nm?

 

[apamise]

$1750 card that is only 16% faster than a $700 card getting "highly recommended" badge just doesn't feel right.

 

[M2B]

You know that's not how node comparison's work, or do you? While you have physical characteristics that you can direct measure, then electrical characteristics & finally the most important part of the puzzle ~ the uarch. Just because you say TSMC 7nm is "much more better" than SS' 8nm doesn't make it a fact that it'd be also much better for Ampere, unless you have the same GPU/chip/uarch made on the two nodes. The closest comparison IIRC was the VII & Vega64.

Anyone claiming otherwise is just guesstimating! So waiting for you to provide evidence about how superior (or inferior) one is wrt to the other.

Unlike you, I don't lable my thoughts as facts.
I'm not saying TSMC's N7P or whatever is for definite better than Samsung's 8N, but based on what we've seen before it's a pretty safe bet that the tsmc node is better both in terms of performance and efficiency.
And yes I'm aware that you can't compare different architectures across different nodes so easily, Mr. Electrical Engineer.

I'm basically wasting my time with someone who can easily lie to prove his made up nonsense correct.

 

[R0H1T]

Oh really, so I guess you must've mislabeled these thoughts as well?

while RDNA2 is barely 10-15% more efficient than ampere while being on a much better node.

Or how about this one?

And also the most efficient RDNA2 chip is barely beating the 12nm 1660Ti at 1080p (in terms of efficiency),

It'd be nice if you stopped at the point where you said the 3070 is the most efficient GPU in their lineup - to which I even said that there could well be more efficient Ampere GPUs released after it - but then you just had to bring in some other assumptions to prove a point huh?

 

[M2B]

Or how about this one?

What in the actual fuck is wrong with you?
That was a different discussion with someone else to prove my point that 1080p is not the best indicator of the relative efficiency for these high-end cards.
Let's just end this mess here and move on. What a fucktard.

 

[swirl09]

80'c ? pretty bad. ASUS RTX 3090 STRIX OC = 68'c

The Strix default profile is very aggressive. I had to play around a lot to get it to my liking, which wasnt as easy as it should be due to the fact that currently there is an unknown fan offset on afterburner atm - presumably will be fixed in a future update of either the program and/or card. I have mine targeting 2010 / v0.993 / 70C / 60% fan which is reasonably quiet - just audible. There is some variance from game to game.

The out of the box settings for the Suprim gaming bios is much better balanced.

 

[owen10578]

Aw I was hoping for a 500W power limit after having the highest power limit on the 3080 Suprim X out of all the 3080s.

I really don't get people complaining about the power consumption. If you care so much about it why buy a pre-overclocked card with a huge cooler and a strong VRM? Even if you wanted the cooler but not the power consumption then just lower the limit manually to whatever you want. Higher power limits gives the user freedom to overclock higher and that's never a bad thing.

 

[okbuddy]

poor 6900xt

 

[kruk]

Really? You lot still going there? I mean if I really wanted to heat my room I would have bought a 295X2.

Yeah, but 295X2 destroyed the competition in 4K:

This one only destroys your wallet

 

[nguyen]

@W1zzard Can you utilize Nvidia PCAT (which does measure performance per watt) and note down the perf/watt of each GPU in a few games and average them down ? I think it would be more precise than using power consumption in only 1 game and 1 resolution at that.

Yeah, but 295X2 destroyed the competition in 4K:
This one only destroys your wallet

Didn't you read the 2011 memo, Average FPS for Xfire/SLI setup are useless numbers, all you get are incomplete frames or microstuttering, that's why Nvidia burried it when they are the first to invent it.

 

[mak1skav]

Isn't this card already recalled? Why people keep arguing about imaginary graphic cards?

 

[W1zzard]

Can you utilize Nvidia PCAT (which does measure performance per watt) and note down the perf/watt of each GPU in a few games and average them down ? I think it would be more precise than using power consumption in only 1 game and 1 resolution at that.

I've told NVIDIA I want to check out PCAT, they said they'll look into it but so far they dont have enough units to give me one

Isn't this card already recalled?

MSI Pauses Shipments of RTX 30-series SUPRIM X, No Recall in Progress

MSI had paused shipments of its RTX 3090 SUPRIM X and RTX 3080 SUPRIM X graphics cards on suspicion of a quality issue with their cards. Earlier today, a Reddit post by someone who was in queue with a retailer to purchase an RTX 3080 SUPRIM X graphics card was allegedly informed by the retailer...

www.techpowerup.com

$1750 card that is only 16% faster than a $700 card getting "highly recommended" badge just doesn't feel right.

I have between $1500 and $2000 to spend on a graphics card, what would you recommend?

 

[Sovsefanden]

474w peak in games
....oooookay then
fermi 2.0

No its not, simply look at performance per watt, which looks fine, for reference cards that is, which is the only thing that matters

AMDs 6800 series watt usage also skyrockets when overclocked, 6800 xt goes to 300+ watts easily in gaming when overclocked, 3080 goes to 350 but performs better on avg + has DLSS and better ray tracing

3090 beats them all and hits 400+ at times, im sure 6900 xt is going to be upthere as well post oc

Could not care less about 3090 or 6900 XT since its 1000+ dollars cards (on paper), worst perf value you can get and 6900 xt is going to be missing in action for months

Hell even 6800 series are MiA and most retailers say people will likely wait deep into 2021 to get one, AMD fucked their launch up more than Nvidia

 

[Vayra86]

Right, so let's see your facts that back up the claim that TSMC 7nm is much more superior than whatever Ampere's made on? When you get the numbers, assuming you have the same exact GPU on the two separate nodes, then wake me up! Till then keep your claims to yourself & FYI the most efficient RDNA GPU was in a Mac & way more efficient than the likes of 5700xt so if you think that the 6800 is the efficiency king, just wait till you see these chips go into a SFF or laptops with reasonable clocks

Not a bad idea actually, though you have to agree (or not) that this brings us virtually a full circle in the computing realm ~ first zen3 & now RDNA2.

Lol. It was exactly the reason AMD was going to catch up prior to this release, in the eyes of many people. And here you are saying the opposite AFTER the cards are out? I think its crystal clear now that AMD gains the advantage through a combination of architecture (they finally !!! caught up, look at how the GPUs work with boost, there is feature set parity, etc.), and having a better node. So yes, its pretty logical to assume - even without a shred of direct evidence beyond benchmarks and perf/watt numbers and die sizes/specs, that TSMC has that much of a better node.

On the other side of the fence, we see Nvidia releases GPUs that are not as optimal as they usually do. Power budgets have risen tremendously, after half a decade of strict adherence to 250 ish watt GPUs on the top. Now they exceed that by a royal margin. Not 20W or something as they did with Turing, but in the case of x80, nearly 100W depending on what gen you compare to. That's huge and its def not up to architecture where Nvidia has been leading and kinda still does, as their time to market was and still is better, despite them 'rushing' Ampere.

I have between $1500 and $2000 to spend on a graphics card, what would you recommend?

Save half, upgrade faster, duh. Or buy two x80s so you can SL...oh wait.

 

[W1zzard]

Save half, upgrade faster, duh.

But I'm rich, I want the best, and it should be quiet

 

[XL-R8R]

I have between $1500 and $2000 to spend on a graphics card, what would you recommend?

Spending half the amount for 85%+ of the performance and not buying silly overpriced GPU's to support the market of nonsense pricing scheme thats arrived over the last 6 years? Is what I'd personally recommended.... not this.

Put the rest of the money you just saved, in shares for your favorite tech company.


Edit:

"and it should be quiet" - FE edition card with waterblock - still cheaper and probably as fast (if not faster with tweaks)

 

[spnidel]

No its not, simply look at performance per watt, which looks fine, for reference cards that is, which is the only thing that matters

474w is 474w regardless of what the performance per watt is - it's not a card I would put in my machine

AMDs 6800 series watt usage also skyrockets when overclocked

which is where an undervolt overclock comes into play...

3090 beats them all and hits 400+ at times, im sure 6900 xt is going to be upthere as well post oc

idk given that the 6900 xt has the same tdp as the 6800 xt, and taking into zen 2 and zen 3 tdp numbers into account, I wouldn't be surprised if the power draw difference was minimal between the 6800 xt and 6900 xt

 

[medi01]

@W1zzard

I realize what I'm asking is lots of work, but I think OC performance is relevant to many reviewers out there.
It would be cool if OCed GPU would be tested across the board (all games/charts, not just one)

3090 beats them all and hits 400+ at times

400+ is not how I'd refer to a card ringing at 500W door, and that before being OCed:

 

[spnidel]

@W1zzard

I realize what I'm asking is lots of work, but I think OC performance is relevant to many reviewers out there.
It would be cool if OCed GPU would be tested across the board (all games/charts, not just one)


400+ is not how I'd refer to a card ringing at 500W door, and that before being OCed:

View attachment 177181

yeah there's no way that thing doesn't peak at under 500W when overclocked... power consumption wise these cards are a fucking joke, jesus christ

 

[R0H1T]

Lol. It was exactly the reason AMD was going to catch up prior to this release, in the eyes of many people.

You can keep LOLing all you want, unless you have a GA104 die on 8nm Samsung & 7nm TSMC you cannot say for sure how the latter is superior (or inferior) to the former ~ that's a fact. Anything else is just conjecture, & that's what I was fighting ~ the assumption that one node is much superior to the other without actual hard numbers to back any claim to the effect. So unless now you're gonna claim that you have that data with us, how about you keep the same assumptions in check?

It wasn't the only reason, if zen2 -> zen3 & RDNA -> RDNA2 has taught us is that uarch is a major part of the equation which simply cannot be ignored. IMO Ampere is a dud relative to Turing, now you can claim that TSMC 7nm would've made a big difference here but until we get real data to cross reference how performance can vary between nodes ~ you're just making castles in thin air.

 

[W1zzard]

I realize what I'm asking is lots of work, but I think OC performance is relevant to many reviewers out there.
It would be cool if OCed GPU would be tested across the board (all games/charts, not just one)

I agree it would be cool, but it's simply not feasible with all the tests and games that I have

 

[Sovsefanden]

yeah there's no way that thing doesn't peak at under 500W when overclocked... power consumption wise these cards are a fucking joke, jesus christ

I doubt people that pay 1750 dollars for a card will care about it using 100 watts more or less. This card is 31dB load, which is very quiet for this amount of performance and flagship gpu's always use alot of power, big dies

Simply don't buy this card and don't overclock if you think watt usage is important, both AMD 6000 and Nvidia 3000 series use way more power when overclocked and OC scaling is low on both (for 24/7 OC, not LN2 testing and other useless methods)

Custom cards with higher boost clocks yields ~2% extra performane. Powerdraw increases by 20% this is true on both AMD 6000 and Nvidia 3000 - OC some more and see another 2-3% performance, with another 20% bump on power. You are now up 40% in powerusage, but only 4-5% up in performance, worth it?

Not to mention that you pay several hundred dollars extra for some of these custom cards, which delivers almost nothing over the reference solution (which are good on both AMD and Nvidia now - which is why custom cards barely performs better)

Hell some 6800XTs end up at 800 and even 900 dollars, thats insane for a 650 dollar card
And this is the MSRP price, not the scalper price

 

[the54thvoid]

I agree it would be cool, but it's simply not feasible with all the tests and games that I have

For arguments sake, is the OC % gain applicable to most other scenarios? If so, there's absolutely no need to test on all games. A generalisation of 5% better means 5% better. Unless other games don't scale well with clock increases? And then again, these days, across the cards of AMD and Nvidia, we see, what, 5% uplifts? A few fps, or an extra 10 at 200fps... It's 'almost' not worth it.

 

[W1zzard]

For arguments sake, is the OC % gain applicable to most other scenarios? If so, there's absolutely no need to test on all games. A generalisation of 5% better means 5% better. Unless other games don't scale well with clock increases? And then again, these days, across the cards of AMD and Nvidia, we see, what, 5% uplifts? A few fps, or an extra 10 at 200fps... It's 'almost' not worth it.

It's close enough

 

[Vayra86]

You can keep LOLing all you want, unless you have a GA104 die on 8nm Samsung & 7nm TSMC you cannot say for sure how the latter is superior (or inferior) to the former ~ that's a fact. Anything else is just conjecture, & that's what I was fighting ~ the assumption that one node is much superior to the other without actual hard numbers to back any claim to the effect. So unless now you're gonna claim that you have that data with us, how about you keep the same assumptions in check?

It wasn't the only reason, if zen2 -> zen3 & RDNA -> RDNA2 has taught us is that uarch is a major part of the equation which simply cannot be ignored. IMO Ampere is a dud relative to Turing, now you can claim that TSMC 7nm would've made a big difference here but until we get real data to cross reference how performance can vary between nodes ~ you're just making castles in thin air.

But there Is real data and it says you are wrong every time. Sources were provided, did you even get into them or? Do you read reviews? We know how efficient the arch and the node is, the products are out... We know Ampere has more efficient RT, but less efficient raster. We know the end FPS, bandwidth/memory/power usage, we know die size and we know the architectural improvements. If you still didn't figure it out by now, you never will. Your opinion is irrelevant, the facts are there. Its like a graph with lots of dots, you just need to connect them with straight lines and you've got your comparison well laid out.

For arguments sake, is the OC % gain applicable to most other scenarios? If so, there's absolutely no need to test on all games. A generalisation of 5% better means 5% better. Unless other games don't scale well with clock increases? And then again, these days, across the cards of AMD and Nvidia, we see, what, 5% uplifts? A few fps, or an extra 10 at 200fps... It's 'almost' not worth it.

Exactly. Its as much of a non-issue as the additional OC headroom this specific card offers. Margin of error territory at best.