news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

new wisdom report

editor: so sleepy

[introduction to new wisdom]with the release of a large number of revelations, lao huang’s rtx 50 series graphics card seems to be closer to us!

recently, the specifications of nvidia's next-generation flagship graphics card rtx 5090 and rtx 5080 have been leaked again!

compared with the previous generation 4090, the number of cuda cores of the rtx 5090 has reached 21,760, an increase of 33%.

the video memory adopts the new gddr7 architecture, with the capacity upgraded to 32gb and the bit width returned to 512 bit, both increased by 33%.

after a wave of upgrades, the power consumption of the rtx 5090 has also reached an astonishing 600w. compared with the 450w of the 4090, it is still a 33% increase.

however, the upgrade to the rtx 5080 is lackluster.

the number of cuda cores is 10,752, which is 11% higher than 4080 and 5% higher than 4080 super.

the video memory capacity and bit width have not been improved, and are still 16gb and 256 bit.

the power consumption has reached 400w, an increase of 25%.

next, let us take a closer look at what information is currently known.

RTX 5090

first up is the flagship geforce rtx 5090.

according to kopite7kimi, the rtx 5090 is expected to use pg144/145-sku30 pcb design and be equipped with gb202-300-a1 gpu core.

of these, the 5090 will enable 170 of the total 192 sms (streaming multiprocessors), with a cuda core count of 21,760 instead of the full 24,576.

this represents an 11.4% reduction in core count, which is slightly higher than the 11.1% reduction in the rtx 4090 relative to its full ad102 chip.

it is worth noting that this reduction in the number of cores is a common practice in high-end gpus in order to optimize yield and performance.

in terms of video memory, the rtx 5090 shows amazing performance - a new 32 gb gddr7 video memory, a peak width of 512 bits, and a bandwidth of up to 1.792 to 2.00 tb/s.

in terms of power consumption, the tbp (total board power) rating of the rtx 5090 will reach 600w.

interestingly, although the tbp has reached 600w, nvidia is likely to continue to use the dual-slot radiator design on the official founders edition.

this means that compared to the previous generation of products, the new generation of graphics cards will have a significant improvement in heat dissipation efficiency.

it is understood that the n4p process node used by blackwell is only a slight upgrade compared to the n4 used by the previous generation ada. therefore, netizens have speculated that this generation of founding edition graphics cards is likely to be some kind of "liquid-cooled beast."

RTX 5080

next, the xx80 series that netizens complained about again - rtx 5080.

the rtx 5080 graphics card will be based on the pg144/147-sku45 pcb design and use the gb203-400-a1 gpu chip.

it is not difficult to tell from the naming that the 5080 will use the complete gb203 gpu chip, with 84 sm (streaming multi-processor) and 10,752 cuda cores.

although the value is not small, compared with the rtx 5090, it is directly cut in half (less than 50%). the previous difference between rtx 4090 and rtx 4080 was 40%.

in addition, the rtx 5080's video memory configuration is also halved - 16 gb capacity, 256 bit width, and the video memory bandwidth is expected to be between 896 gb/s and 1024 gb/s.

the tbp (total board power) rating of the rtx 5080 will reach 400w, which is a 25% increase from the power consumption limit of the previous generation product.

taken together, the overall performance gap between rtx 5090 and rtx 5080 will be very significant.

netizen: it can be called a replica of 4080 turned into 4070 ti

as the specifications leaked, the most heated topics were the 5090's 600w power consumption and the 5080's disappointing 16 gb video memory specifications.

you know, the 4080, which is not so far behind the rtx 4090, is already one of the worst value performers among the 80 series products. it was not until the "super" version was launched that the situation improved slightly.

but it might not be entirely a bad thing that the rtx 5080 is so emasculated.

some netizens said that if the space left between 80 and 90 is large enough, amd may be able to successfully tempt amd to re-enter the high-end gpu market.

is there also a 24gb version?

not to mention, not long after the above specifications were leaked, there were new revelations about the rtx 5080.

according to polymorph, a well-known member of the well-known hardware forum chiphell, nvidia may launch a 24 gb version of the rtx 5080.

it is worth noting that polymorph has previously leaked the first image of the ga102 "ampere" gpu core, revealed the configuration of the rtx 40 series graphics card, and introduced the specifications of the rx 6000 "rdna 2" gpu in detail, so this revelation should also be of greater significance. high credibility.

there are currently two possible ways to implement a 24 gb nvidia geforce rtx 5080 graphics card.

the first, and least likely, way is for nvidia to use the gb202 gpu core, which would support a 384-bit memory bus and 24 gb video memory configuration.

another more possible path is to use a 3 gb video memory module using gddr7 technology.

previous 2 gb modules were available in 32 gb and 16 gb video memory capacities. but if you use a 3 gb module, the rtx 5080 can reach 24 gb of video memory capacity while maintaining the same pcb design, core specifications, and 256-bit memory bus configuration.

however, since gddr7 adopts a new architecture, if you want to achieve higher memory speeds and larger capacities, you still need to wait for the manufacturing process to reach a sufficiently mature level.

has the new card been sent for testing?

compared with these specifications, the more exciting news is that the first batch of rtx 50 series gpus are said to have been sent to the testing laboratory!

recently, netizen harukaze5719 discovered several rtx 50 gpu configurations in the shipping list, mainly involving the 16 gb rtx 5080 model:

699-1G144-0030-TS1（RTX 5090）

699-1G144-0050-TS1

699-1G144-0045-TS1（RTX 5080）

691-1G145-2030-TS1（RTX 5090）

699-1G147-0070-TS1

699-1G147-0050-TS1

699-1G147-0045-EB1（RTX 5080）

however, we still don't know exactly when these graphics cards will be available.

after all, there is no reason why lao huang has to release the rtx 50 series in 2024 - it has no competitors, and the performance of the rtx 40 series is still strong.