NVIDIA GeForce RTX 5090 Founders Edition

[Onasi]

Funny how the most interesting part of the review to me was the DLSS 4 analysis and W1zz saying that the new transformer model is just THAT much better than what we had previously. Like, that’s cool. The card itself? Ehhhh, about what we all expected. Amusing how the fastest consumer GPU on the planet is just kind of… unexciting? Pedestrian?

 

[mb194dc]

I keep harping on about the wall that silicon lithography has hit and people keep ignoring me, and then do surprised Pikachu faces when there is no efficiency gain generation-on-generation. Because efficiency, by and large, comes from the node size and that isn't getting appreciably smaller. If y'all are crying this hard about lack of generational performance uplift now, you're gonna be drowning in your tears for a long time, because there ain't any good solutions in sight in the next half-decade at best. Physics is a harsh mistress.

How about if we put 5090 at higher than 4k resolution? I agree with you re the wall. 600w power consumption for 27% increase at 4k shows it.

My guess is at 8k 5090 with massive memory bandwidth will be much more efficient, even if unplayable.

 

[oxrufiioxo]

I think it will blast the side panel/glass, impacting every hardware piece, unless you have a case with a open-mesh design for a side panel for the air to escape that way, then I don't see any issues.

My understanding is it would blow air straight towards the motherboard if vertical mounted.

If you watch the GN video they have some really cool video of the airflow pattern

 

[R0H1T]

I keep harping on about the wall that silicon lithography has hit and people keep ignoring me, and then do surprised Pikachu faces when there is no efficiency gain generation-on-generation. Because efficiency, by and large, comes from the node size and that isn't getting appreciably smaller. If y'all are crying this hard about lack of generational performance uplift now, you're gonna be drowning in your tears for a long time, because there ain't any good solutions in sight in the next half-decade at best. Physics is a harsh mistress.

You know that's not even remotely true, CPU's have been getting more efficient on the same node for eons! Heck Nvidia themselves did it 3(2.5?) times on 28nm

Although getting them more efficient is probably exponentially harder now.

 

[Legacy-ZA]

My understanding is it would blow air straight towards the motherboard if vertical mounted.

If you watch the GN video they have some really cool video of the airflow pattern

Oh yeah lol, of course it will, duh, yeah that won't be good for the mobo or anything in that PC. I would stay clear of using it in a vertical mount, I don't even have to do a test to know it will be bad, I am definitely more interested in the horizontal position.

I create a slipstream in my PC chassis, front intake that blow on my GPU/Memory/CPU, my CPU is a push-pull configuration, removing all air at the back and top. Works really well.

 

[oxrufiioxo]

Oh yeah lol, of course it will, duh, yeah that won't be good for the mobo or anything in that PC. I would stay clear of using it in a vertical mount, I don't even have to do a test to know it will be bad, I am definitely more interested in the horizontal position.

I create a slipstream in my PC chassis, front intake that blow on my GPU/Memory/CPU, my CPU is a push-pull configuration, removing all air at the back and top. Works really well.

The FE is going to be unoptanium for most my guess is 1/100 cards sold will be FE models if not lower than that.. They mentioned themselves that it's a very limited model.

 

[chrcoluk]

I hope this criticism is accepted, the biggest thing I was waiting for in this review is the cooler's performance, and then we have a table which only has a 4080 Super for comparison a much lower TDP card, would it have been too difficult to test the 4090 FE as well as a comparison? as ultimately I dont think I am the only one here who wants to compare the old FE and new FE cooler design, adding AIB results later still wont give us this data. Guess need to wait for the 5080 FE review.

Also curious of the ambient temp as your 4080 Super FE runs about 20C higher than mine under load.

 

[R0H1T]

My guess is at 8k 5090 with massive memory bandwidth will be much more efficient, even if unplayable.

This thing's basically like an OCed 14900ks

 

[Scircura]

I'm eager to try the transformer-based DLSS on my 3070. I was surprised to read W1zzard's dislike for DLSS2-SR, followed immediately by big praise for DLSS4-SR.

My second surprise came from Nvidia patting themselves on the back for pairing a 40dB howler of a HSF with a F$2000 (the F$ is for fantasy $) card, right after finally developing quiet-ish coolers for the 40 series. It's not something to be proud of IMO: the 2-slot HSF is a mismatch for the 5090.

(the 4070 with its 12GB runs out of VRAM when enabling even the Medium Path Tracing (Full Ray Tracing) setting in Indiana Jones And The Great Circle).

I would not call 12GB insufficient. 10 seconds later from your YT timestamp, the video host lowers the texture pool size from max and it goes back to running fine with PT.

 

[Eidolon]

@W1zzard Shouldn't the 5090's memory clock be 2333 MHz instead of 1750 MHz? With G7 the effective data rate is 12x the clock rather than 16x with G6X due to the change from PAM4 to PAM3

Edit: 1750 MHz is correct, my bad

 

[skizzo]

removing the hot spot sensor seems stupid. the only drama was clowns who didn't know what it was. along with the rest of the circus that think the hotspot should never go over some arbitrary value like 50C lol

 

[Legacy-ZA]

DLSS 4 is looking mighty fine, I like what I see, very much:

 

[lexluthermiester]

I keep harping on about the wall that silicon lithography has hit and people keep ignoring me

Not at all. Most of who are in the know completely understand what you're saying and agree to a point. Some of us still think advances are possible. But you're not wrong, we are approaching the atomic wall.

and then do surprised Pikachu faces when there is no efficiency gain generation-on-generation. Because efficiency, by and large, comes from the node size and that isn't getting appreciably smaller.

True to a point. Chemistry is also a factor. If we can find a chemistry that conducts better without creating waste heat we can truly make advancements in efficiency. That's something the IC makers need to do. NVidia is not in the fab market.

 

[Scircura]

You know that's not even remotely true, CPU's have been getting more efficient on the same node for eons! Heck Nvidia themselves did it 3(2.5?) times on 28nm

Wow, I didn't realize GTX 680, 780, and 980 were all TSMC 28nm.

Efficiency nearly doubled in that time!

 

[mtosev]

When will the review of custom cards be published?

 

[3valatzy]

Wow, I didn't realize GTX 680, 780, and 980 were all TSMC 28nm.

What a genius invention !

Because 780 is a rebranded 685 or 690, while 980 is a rebranded 780

 

[lexluthermiester]

Because 780 is a rebranded 685 or 690, while 980 is a rebranded 780

What now? You're joking?

 

[harm9963]

Well as a 4090 owner , $1699.99 retail for a ZOTAC AMP EXTREME AIRO at Micro Center in person , when I went from a 1080Ti to 40 series , got max bangs for my bucks ! those days are gone , comes down to bragging rights , I can warm up to DLSSQ more and more , and FG , which brings up , can't wait to test newer versions , Lossless Scaling 3.0 can do 4x or more , what's funny is Nvidia open a can of worms ,comparing 5070 to 4090 , using x4 Lossless Scaling , in 2077 , I can use DLAA for the first time and looks great standing still

2077 has the new DLSS4 ,Transformer works with my 4090 , now with DLAA and TFG !

 

[lepudruk]

When will the review of custom cards be published?

@24th

 

[3valatzy]

What now? You're joking?

Look at the chips. GK104 294 mm^2, that's in fact a 670-class GPU, then you got a GK110 - proper high-end 561 mm^2, and then you got a backport GM204 which is also not an enthusiast-level GPU with tiny chip only 398 mm^2. In comparison, GB202 is 750 mm^2 which is the largest Nvidia GPU ever made.

 

[mkppo]

What a genius invention !

Because 780 is a rebranded 685 or 690, while 980 is a rebranded 780

IIRC, 770 was a straight up rebrand of 680 but 9xx was a different arch altogether and much more efficient at the same node.

Both the 680 and 770 were worse than the Radeon 7970 though, which was released 6 months before the 680.

So yes, efficiency improvements are entirely possible in the same node, just not that commonplace especially nowadays. 770 to 970 was massive, then we had another massive one with Radeon VII > 5700XT. In both instances it was a major architectural overhaul though, not a minor one like in 5090's case.

 

[TheinsanegamerN]

I mean yeah it’s an impressive gpu but the price unimpressive …. I will tank it with my 3090 for another 2 years for 60 series or whatever amd offering they have then

but folks on 10 series should upgrade and 20 series can consider … 30 series folks I think can skip this 50 series

The best part about stagnation is the sheer lifespan you can get out of older hardware.

That's really the problem here, I don't think there has ever been a new generation with virtually zero improvements in perf/watt or perf/price, even the rebrands usually at least had lower prices. It's quite an achievement.

GTX 700 series. Unless you count the discounted launch price going from 500 to 400, which was barely lower then 680s on sale. Or the 390/x cards.

Yeah exactly - 4090 was really well priced @$1600, so if you bought that in 2022 or 2023, you now have a 5080+ performance... vs if you bought a $1100 4080 and now a $1100-1200 5080 - you would have had less performance for $2400.

So.....you're saying when jensen said "the more you buy the more you save", he was telling the truth?

 

[Upgrayedd]

Cooler seems kinda crappy? It's hot AND loud.

 

[3valatzy]

IIRC, 770 was a straight up rebrand of 680 but 9xx was a different arch altogether and much more efficient at the same node.

Both the 680 and 770 were worse than the Radeon 7970 though, which was released 6 months before the 680.

Yeah, the point is that GK104 (GTX 680) and GK110 (GTX 780) belong to one generation, artificially split by the marketing to "two" virtual generations, hence I call it a rebrand.

This is the same as if AMD decided to call Navi 31 a Radeon RX 8900 XTX, while Navi 32 remained as a Radeon RX 7900 XTX.

 

[lexluthermiester]

So.....you're saying when jensen said "the more you buy the more you save", he was telling the truth?

So with a $2000 price tag, look at all the money you're saving!!!....

To be fair. While the $2k price is steep, you do get the performance you pay for and it's hella cheaper than the professional cards that START at $4000.

Cooler seems kinda crappy? It's hot AND loud.

Um, no You need to read again;

NVIDIA GeForce RTX 5090 Founders Edition Review - The New Flagship

NVIDIA's GeForce RTX 5090 is the fastest graphics card ever released. It comes with a whopping 32 GB VRAM, and support for multi-frame generation, which achieves hundreds of FPS easily. We also managed to disassemble the card, and our review includes performance testing without upscaling or...

www.techpowerup.com

Also; see Jay's review where he demonstrates with a microphone;