• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

NVIDIA's 3xxx Reference Cooler Costs $150 By Itself, to Feature in Three SKUs

llien

Member
According to the report, the cooling solution, which looks a lot more overengineered than the company's RTX 20-series Founders Edition cooler, costs a hefty USD $150, or roughly the price of a 280 mm AIO CLC. The cooler consists of several interconnected heatsink elements with the PCB in the middle. Igor's Lab reports that the card is estimated to be 21.9 cm in length. Given its cost, NVIDIA is reserving this cooler for only the top three SKUs in the lineup, the TITAN RTX successor, the RTX 2080 Ti successor, and the RTX 2080/SUPER successor.

FOUREJTRyX2TcqWV_thm.jpg


All three will use the same cooling solution, and a common PCB design codenamed PG132. Further, all three cards will be based on a common ASIC, codenamed "GA102," with varying hardware specs. The "SKU10" (TITAN RTX successor) could ditch the TITAN brand to carry the model name "GeForce RTX 3090," max out the 384-bit wide memory bus of the GA102 ASIC, and feature a whopping 24 GB of GDDR6X memory, with 350 W typical board power.

The next SKU, the SK20, which is the RTX 2080 Ti successor, will be cut down from SKU10. It will feature 11 GB of GDDR6X memory across a 352-bit wide memory interface, and have a 320 W typical board power rating. This board will likely feature the RTX 3080 Ti branding. Lastly, there's the SKU30, which is further cut-down, features 10 GB of GDDR6X memory across a 320-bit wide memory interface, and it bears the RTX 3080 model number, succeeding the RTX 2080 / RTX 2080 Super.

When launched, "Ampere" could be the first implementation of the new GDDR6X memory standard, which could come with data-rates above even the 16 Gbps of today's GDDR6, likely in the 18-20 Gbps range, if not more. Lesser SKUs could use current-gen GDDR6 memory at data-rates of up to 16 Gbps.

TPU

44k9rn.jpg
 
$150 doesn't sound believable to me, even for the highest end $1000+ cards. Of course you can ask any price you want, but I bet the BOM is below $50 for something like that.
 

Boxman

Banned
PC master race about to be violently beaten in GPU prices once again. Once the prices for these cards are revealed, I expect many to get a console instead of upgrade their rigs, especially when a 4.8gb/s SSD becomes the baseline.
 
I don't think us people aiming for the Titan, the xx80 Ti and the xx80 setups are really worried about pricing. If you want the best visuals, framerate, and performance, you cannot get these for 5 or 6 hundred bucks. Everyone knows this, and it's argued over and over, whenever console fans argue against PC. You gotta pay to play. Cant wait for these to release so I can surpass my 2080 ti, which has already surpassed next gen already.
 

-YFC-

Member
Damn this is expensive AF. A bit weird if you ask me. It's a chunk of aluminum/copper and 2 fans attached to it. I mean I get that machining time costs money, but that it costs 150 bucks to make in large volumes? Somehow I'm skeptical towards this statement. To me this seems like Nvidia's trying to justify the huge cost this gpu will be. I dunno. But coolers produced in high volume don't cost this much. Weird if ya ask me.
 
Damn this is expensive AF. A bit weird if you ask me. It's a chunk of aluminum/copper and 2 fans attached to it. I mean I get that machining time costs money, but that it costs 150 bucks to make in large volumes? Somehow I'm skeptical towards this statement. To me this seems like Nvidia's trying to justify the huge cost this gpu will be. I dunno. But coolers produced in high volume don't cost this much. Weird if ya ask me.
Definitely agree with this. When you are machining and tooling for a very small handful of products, prices will definitely be high. But for a card like this, it will be produced on a higher scale than a handful, which should decrease prices. I could see half of the stated price, but definitely not $150
 

KungFucius

King Snowflake
I cant wait to strip it off and add a custom waterblock cooler that costs 350.

Not really. I don't know why they would do this unless it is absolutely required to get the performance they are shooting for. It seems like terrible engineering to make the cooling system such a big expense. Either the design is fake, the cost is bullshit, or this enables huge gains in performance or stability that is critical to the product..
 

kiphalfton

Member
Well in any case, at least the aib versions should be the same cost as the FE version. Which I find really weird since Pascal or whenever they started this trend of "premium" reference cards. Why buy the reference design when you can get a better aib version for about the dame price. I understand at release the FE version is the only thing available, usually for a couple months, but then after that it's kind of pointless. That gap was closed a bit with the Turing FE cards, as they actually look pretty good, but even so. If a EVGA FTW was the same price as a FE card, I would take the former all day every day.
 

Sethbacca

Member
The prices on this shit have gotten stupid, and blown straight by into full retard. Clearly Nvidia has not had competition in the high end for a while.
 
Well in any case, at least the aib versions should be the same cost as the FE version. Which I find really weird since Pascal or whenever they started this trend of "premium" reference cards. Why buy the reference design when you can get a better aib version for about the dame price. I understand at release the FE version is the only thing available, usually for a couple months, but then after that it's kind of pointless. That gap was closed a bit with the Turing FE cards, as they actually look pretty good, but even so. If a EVGA FTW was the same price as a FE card, I would take the former all day every day.
All day, everyday! I'm not sure why the reference cards cost that much, especially when AIB cards crush the performance and overclocked (even auto clocks)
 

kiphalfton

Member
Also nvidia stuff has never really been an issue keeping cool (well besides what, the 400 series stuff)... so what's the need for this super elaborate cooler design?
 

llien

Member
I think you are reading it wrong.
Of course it isn't as expensive on BOM.

It's just an excuse to push pricing even further.

All 3 being based on xx102 silicon is also quite telling (and so is 300+ watt rating)

The first Pascal silicon was 330mm2. Here we are likely in 550mm2+ area.
 
Last edited:
The prices on this shit have gotten stupid, and blown straight by into full retard. Clearly Nvidia has not had competition in the high end for a while.
You'd be surprised. Apparently Nvidia is doomed and so far behind compared to AMD. AMD is absolutely killing it right now, with the best raytracing possible, it's image sharpening is light years above and beyond Nvidia's DLSS2, and way better performance. Nvidia can't hold a candle to AMD RDNA2. At least that's what the console warriors are claiming, which anyone with a few, barely working brain calls could conclude as absolute bullshit. AMD processor team =/= AMD GPU team
 
Last edited:

THE:MILKMAN

Member
$150 doesn't sound believable to me, even for the highest end $1000+ cards. Of course you can ask any price you want, but I bet the BOM is below $50 for something like that.

Nvidia should have a word with the Bloomberg reporter. He knows a supplier that can make a quality cooler for a few dollars.
 

Orta

Banned
The ti cards have always been out of my financial reach so like most I'll be hoping the 3070 will be reasonably priced.

Not to mention will leave the consoles gasping.
 
Last edited:

The Skull

Member
Beefy cooler makes sense if the TBP is over 300 watts. Have they done an Intel and just pushed the clocks like crazy?
 
I understand at release the FE version is the only thing available, usually for a couple months, but then after that it's kind of pointless.
Not really. Reference PCBs are by far the most used template when making waterblocks. Buying an AIB card for custom watercooling is like rolling a pair of dice and expecting to land double 6.

Edit: To say nothing of the fact that FE cards have been getting binned dies so they're basically always great overclockers since Pascal.
 
Last edited:

godhandiscen

There are millions of whiny 5-year olds on Earth, and I AM THEIR KING.
I wonder how loud it is. I remember the GTX480 cooler was also very novel, but super noisy and the card would still reach insane temperatures.
 

Celcius

°Temp. member
They look ugly to me and the design also seems noisy, but I plan to get that Titan this time, along with upgrading the rest of my computer.
 

Agent_4Seven

Tears of Nintendo
If the bottom of the barrel reference cooler alone cost $150, AIB ones for Strix and Aorus (for example) variants can potentially cost even more than that. Fuck all that, all things point towards PS5 purchase this year / early next year.
 
Last edited:

kiphalfton

Member
If the bottom of the barrel reference cooler alone cost $150, AIB ones for Strix and Aorus (for example) variants can potentially cost even more than that. Fuck all that, all things point towards PS5 purchase this year / early next year.

Maybe, but I have faith the aib versions will just take the 1/2/3 fan route as usual instead of a front/rear fan approach.
 

ZywyPL

Banned
The first Pascal silicon was 330mm2. Here we are likely in 550mm2+ area.

RT and Tensor cores do need their die space. I think times of 250-350mm GPUs are gone and 450-800 will be the new standard, at least for NV, while AMD might want to try to incorporate all the functions into more complex CUs, like they did with Rapid Packed Math to incorporate FP16/32/64 onto same cores.

RPM already allows for AI computations as seen on upcoming XBX, so the question is if RT calculations will also be able to run on CUs or will need separate, specialized hardware. But seeing the Big Navi rumors being 505mm GPU with 80CU, as well as XBX 52CU GPU being so small, I think they actually made it. So technically, they should have an upper hand when it comes to pricing.
 
Last edited:

replicant-

Member
I built a nice Rig last year based on Ryzen 3800 last September. For the GPU I went with a 2070 Super as the plan was/is to go Ampere this year.

The 3080 should be a good solution to what i want. The 3080ti cost probably won't be prohibative - I just don't really want to spend over £1K on a GPU. Especially with the PS5 and XSX being on the buy day one list.

Also interested in seeing the 2070 Super replacement.
 

llien

Member
RT and Tensor cores do need their die space.
RT was estimated to be about 8% or less in Turing.

For Tensor Cores => they do not have their own ALUs, according to anand, which is why you cannot use tensor cores and CUs simultaneously.
This is why CUDA/ALU per mm2 is basically the same between Pascal (no tensors) and Volta.

With Lisa sharpening her axe and opening in September with 505mm2 chip (vs Polaris 240mm2) NV could not afford going with 330.
 
Top Bottom