NVIDIA GeForce RTX 5090 and RTX 5080 Specifications Surface

erek

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
12,725
"When it comes to the GeForce RTX 5080, NVIDIA has decided to further separate its xx80 and xx90 SKUs. The RTX 5080 has 10,752 FP32 CUDA cores paired with 16 GB of GDDR7 memory on a 256-bit bus. With GDDR7 running at 28 Gbps, the memory bandwidth is also halved at 784 GB/s. This SKU uses a GB203-400-A1 die, which is designed to run within a 400 W TGP power envelope. For reference, the RTX 4090 has 68% more CUDA cores than the RTX 4080. The rumored RTX 5090 has around 102% more CUDA cores than the rumored RTX 5080, which means that NVIDIA is separating its top SKUs even more. We are curious to see at what price point NVIDIA places its upcoming GPUs so that we can compare generational updates and the difference between xx80 and xx90 models and their widened gaps."

1727390062573.png

Source: https://www.techpowerup.com/327026/...tions-surface-showing-larger-sku-segmentation
 
if the rumoured gap are true, could be a sign of a $1000 or less 5080.

The $1000 4080 super refresh make it look like the 4080 $1200 is not sticking.

256 bits, 5% more cores than the 4080 super could that fit under 325 mm with the latest TSMC 4N they used, only 300 watt of cooling pwoer, that could even leave the $800 price (4070 ti msrp) point open with incredible high margin, a logical where the xx80 in a pre 2020 world, considering where they place the 2080super/3080 back in the days, $700->$800 would feel quite reasonable considering the inflation since.

Probably test $1000 first and then make the $800 24GB super version later on if needed.
 
Last edited:
GDDR7 aside, still having 16GB on a $1000+ GPU is a bit lame. This makes me wonder what the 5070 will look like. You can get a 16GB 4070ti Super right now for under $800. I assume a 5070 will perform better, but likely by not a whole lot. They're more or less be asking $600 (my assumption) for a marginally faster card with less VRAM. Unless GDDR7 is that much better than GDDR6, seems like quite a mediocre upgrade.

I feel like they will plan on a 5080ti/Super/whatever between the 5080 and 5090 with the gap being that big.
 
Flogger23m said:
I feel like they will plan on a 5080ti/Super/whatever between the 5080 and 5090 with the gap being that big.
Click to expand...
There rumours that 3GB GDDR7 chips will not be ready at first but will next year, making a super/ti refresh for that gen quite certain (but then again when was the last gen not having one...) I feel like they will all have those now, the question is more will it be a binned down 5090 or an extra 5080, than will there have one.
 
erek said:
"When it comes to the GeForce RTX 5080, NVIDIA has decided to further separate its xx80 and xx90 SKUs. The RTX 5080 has 10,752 FP32 CUDA cores paired with 16 GB of GDDR7 memory on a 256-bit bus. With GDDR7 running at 28 Gbps, the memory bandwidth is also halved at 784 GB/s. This SKU uses a GB203-400-A1 die, which is designed to run within a 400 W TGP power envelope. For reference, the RTX 4090 has 68% more CUDA cores than the RTX 4080. The rumored RTX 5090 has around 102% more CUDA cores than the rumored RTX 5080, which means that NVIDIA is separating its top SKUs even more. We are curious to see at what price point NVIDIA places its upcoming GPUs so that we can compare generational updates and the difference between xx80 and xx90 models and their widened gaps."

View attachment 681966
Source: https://www.techpowerup.com/327026/...tions-surface-showing-larger-sku-segmentation
Click to expand...
Seriously.... Probably looks like this ...
 

Attachments

  • 745621_1665366087347.png
    745621_1665366087347.png
    1.3 MB · Views: 0
Legendary Gamer said:
Seriously.... Probably looks like this ...
Click to expand...
Needs more power ports.

DukenukemX said:
You wish. The 5090 will probably be over $2k because Nvidia knows you'll buy it. The 5080 being $1,200 is a bit of a hard sell considering the 4080 didn't sell well for a $1k.
View attachment 681999
Click to expand...
I certainly don't think they added more RAM to anything without having a big ol' premium attached.

Nvidia wants to be Apple so badly that they treat memory just like Apple does.
 
Ur_Mom said:
One model does.

Leaked information from some shady website -

NVIDIA 5090 32GB P DIDDY EDITION W/ 55G HIGH QUALITY LUBE - $2499USD
Click to expand...
I think that one comes with some muscle relaxant as well, it's a very large GPU.
 
Twisted Kidney said:
I think that one comes with some muscle relaxant as well, it's a very large GPU.
Click to expand...
"Oh, behave!" - Austin Powers :p

In the words of Spreadsheet Steve from HUB (regarding the 'new and improved' vanilla 4070) "You're not getting royally screwed..." - No not royally, just nvidia screwed!
 
LukeTbk said:
There rumours that 3GB GDDR7 chips will not be ready at first but will next year, making a super/ti refresh for that gen quite certain (but then again when was the last gen not having one...) I feel like they will all have those now, the question is more will it be a binned down 5090 or an extra 5080, than will there have one.
Click to expand...

With the assumed performance gap I think a 5080ti will come sooner rather than later. I'm sure they'll do "Supers" in 2026.
 
Has the xx80 stopped to exist or the xx90 since the 4090 a new product class that did not exist before is a bit of a philosophical question, if the 10% faster than a 4090 rumours are true

Going by TPU gpus specs score

1080->2080->3080->4080->5080

1080->2080: 39%
2080->3080: 63%
3080->4080: 49%
4080->5080: 39%
 
pututu said:
This pretty much sums up the RTX5080 spec... 50% less cores than the flagship.


View: https://www.youtube.com/watch?v=cvL-Mplhog8

View attachment 683123
Click to expand...


I was going to mention something similar and saw the HW Unboxed video, more or less sums up my thoughts. The supposed 5080 is more like a 5070. More or less the numbers are moving down. The **80s used to be near top of the line, but it seems like now they are trending more towards upper mid range. Seems like they are planning an 80, 80ti, 90, possibly 90ti, etc. The **70s are more or less becoming the **60 of a decade ago.
 
Lakados said:
Maybe the “90” isn’t the 90 and Nvidia brings back the Titan branding.
Click to expand...
Exactly. However, as a person with many years of no experience with branding at all, I think that would be a mistake. Sure, you can fool a couple of people to buy the 5080 thinking it is an *80 GPU. But when they find out, your brand has lost the trust of the consumers. And next time, they wont buy anything until the entire generation has been released.

In other words, they spend years building expectations into these numbers. Why throw it away?
 
And resolve all the issues in the Standard Model? Including dark matter? :rolleyes:
 
Skyblue said:
Exactly. However, as a person with many years of no experience with branding at all, I think that would be a mistake. Sure, you can fool a couple of people to buy the 5080 thinking it is an *80 GPU. But when they find out, your brand has lost the trust of the consumers. And next time, they wont buy anything until the entire generation has been released.

In other words, they spend years building expectations into these numbers. Why throw it away?
Click to expand...

Nah.

We aren't getting yearly uArch updates anymore. We get process and yield improvements that trickle down from the HALO A.I. parts throughout 2-4 years. With die sizes and transistor densities what they are now(even on the mainstream parts), they can't launch a $900 xx80 part every year. Node development and uArch development is much slower than it once was. We now have to rely on the process/yield improvements and economy of scale to bring faster and faster consumer parts. These humongous monolithic dies have certainly created an issue. If we can get an interconnect to have low enough latencies, we'll see cheap chiplet designs similar to what AMD is doing with Ryzen. Until then, crumbs are expensive.
 
On top of that, the 5070 is STILL only 12GB of VRAM (according to WCCFTech...so hopefully not?).

But the specs seem to indicate that they are building into the original launch room for "super" versions in between the tiers.
 
kirbyrj said:
On top of that, the 5070 is STILL only 12GB of VRAM (according to WCCFTech...so hopefully not?).

But the specs seem to indicate that they are building into the original launch room for "super" versions in between the tiers.
Click to expand...
Super versions might have the 3gb memory chips
 
You must log in or register to reply here.
Back
Top