Connect with us
What will be the GeForce RTX 4090 Details about the What will be the GeForce RTX 4090 Details about the

Components

What will be the GeForce RTX 4090. Details about the configuration and parameters of the AD102 GPU have appeared

Published

on

A lot is already known about GeForce RTX 40 video cards, and gradually more and more data appears on the Web. For example, the GeForce RTX 4090 is credited with a TDP of 600 W, but for the RTX 4070 it will already be 300 W. Fresh data shed light on the GPU configuration of new products.

What will be the GeForce RTX 4090. Details about the configuration and parameters of the AD102 GPU have appeared

So, the AD102 GPU, which will serve as the basis for the RTX 4090, will have 12 GPC blocks. This is 70% more than the GA102, which has only 7 GPCs. Each GPC will consist of six TPCs and two SMs, corresponding to the GA102 configuration.

What has changed is the FP32 and INT32 core configuration. Each sub-core will include 128 FP32 blocks, but the total number of FP32+INT32 blocks will increase to 192. This is because FP32 blocks do not share the same sub-core as IN32 blocks. The 128 FP32 cores are separated from the 64 INT32 cores.

Thus, each sub-core will consist of 128 FP32 blocks plus 64 INT32 blocks, for a total of 192 blocks. Each SM will have a total of 512 FP32 modules plus 256 INT32 modules for a total of 768 modules. And since there are 24 SM modules in total (two per GPC), we have 12,288 FP32 modules and 6,144 INT32 modules, for a total of 18,432 CUDA cores in common terminology.

What will be the GeForce RTX 4090. Details about the configuration and parameters of the AD102 GPU have appeared

The new GPUs will have 192KB of L1 cache per SM block, which is 50% more than the Ampere architecture. This is a total of 4.5 MB of L1 cache in the top AD102. For comparison, the GA102 only has about 1.8 MB.

L2 cache will be 96 MB, which is already 16 times more than the GA102! In fact, this is a kind of analogue of Infinity Cache for AMD cards.

Also in the new architecture will increase the number of rasterization units (ROP). There will be 32 units per GPC block, which is twice as many as now. And since the number of GPCs will also increase, as a result, the AD120 will have 384 ROPs versus 112 for the GA102. From the rest, fourth-generation tensor cores and third-generation RT cores are reported.

Click to comment

Leave a Reply

Your email address will not be published.

Components

AMD wants to show how to make ultra-cheap modern graphics cards. Radeon RX 6300 may be getting ready for release

Published

on

AMD wants to show how to make ultra cheap modern graphics

It seems that in the current difficult times, AMD and Nvidia have decided to play a game of who will release a more affordable graphics card.

AMD wants to show how to make ultra-cheap modern graphics cards.  Radeon RX 6300 may be getting ready for release

AMD already has the Radeon RX 6500 XT and RX 6400 with MSRPs of $200 and $170, respectively. Nvidia only has the $250 GeForce RTX 3050. If someone wants to buy an Nvidia card cheaper, they will have to take something from the old generations. And now we do not take into account the real prices in stores.

But the other day it became known that Nvidia is preparing a GeForce GTX 1630 (although this is still an old generation), now there are rumors that AMD wants to release a Radeon RX 6300.

The information was found in the AMD graphics driver code for Linux. A new adapter with the code name Beige Goby and ID 0x7424 was found there. Beige Goby is the codename for all desktop cards on the Navi 24, i.e. RX 6400 and RX 6500 XT. And this means that the novelty will be built on this GPU.

Sources suggest that this will be the Radeon RX 6300 model, if only because the mobile RX 6300M ​​already exists. In addition, the RX 6500 XT already uses Navi 24 to the maximum, and there is no point in making another adapter between the RX 6500 XT and RX 6400. Thus, the assumption regarding the RX 6300 looks the most logical. Such a card could have 512 stream processors. Of course, the performance will be very modest, but for someone it does not matter.

Given that the RX 6400 is already only $170, the RX 6300 will cost $150 or even less.

Continue Reading

Components

Even the mid-range GeForce RTX 4060 will get GDDR6X memory. New rumors about RTX 40 cards have surfaced

Published

on

Even the mid range GeForce RTX 4060 will get GDDR6X memory

Another piece of data about GeForce RTX 40 video cards has appeared on the Web, and this time it also concerns mid-budget models.

Even the mid-range GeForce RTX 4060 will get GDDR6X memory.  New rumors about RTX 40 cards have surfaced

The reliability of the data is questionable, so it’s worth treating them accordingly, but there are even very few rumors about the same RTX 4060, so these, if only for this reason, deserve attention.

So, according to the source, the GeForce RTX 4060 will be based on the AD106 GPU and will have 4608 CUDA cores. It will also have 8 GB of memory and a 192-bit bus. Moreover, the memory will be GDDR6X, that is, with a high frequency.

The RTX 4070 will be powered by an AD104 GPU with 7608 CUDA cores and the same 192-bit memory bus. But the memory itself will be 12 GB. For Nvidia it is uncharacteristic to use not the widest memory bus in this segment, but here it is worth remembering the greatly increased amount of cache memory in the Ada Lovelace architecture, which will be similar to Infinity Cache in AMD cards. It is he who can allow you to compensate for a narrow tire. In addition, fast memory will partially do this.

Even the mid-range GeForce RTX 4060 will get GDDR6X memory.  New rumors about RTX 40 cards have surfaced

The RTX 4080 will get an AD103 GPU, 10,752 CUDA cores, a 256-bit bus, and 16GB of memory, while the top-end RTX 4090 will be based on the AD102 with 16,128 CUDA cores and a 384-bit bus. The memory of the card will be 24 GB.

The TDP of the cards, in the order they are listed, will supposedly be 200, 300, 350 and 450 watts.

In general, the information looks very plausible, and partially coincides with early data from trusted sources, but this does not mean that it is correct.

Continue Reading

Components

Intel is finally ready to release its desktop graphics cards. A link to Arc A3 maps appeared on the company’s website

Published

on

Intel is finally ready to release its desktop graphics cards

Looks like Intel is finally ready to release its Arc discrete desktop graphics cards. Yesterday, a link appeared on the Intel site regarding the Arc A3 models.

Intel is finally ready to release its desktop graphics cards.  A link to Arc A3 maps appeared on the company's website

True, the link itself did not lead anywhere, and it was later removed, but most likely this means that Intel is already preparing to launch.

In addition, this fact confirms the early data that it is the junior adapters that will be the first to enter the market, as is the case with mobile cards. AMD and Nvidia always do the opposite.

As a reminder, at first, Arc maps will appear only in China and only for PC builders. Then they will be available in retail, but again only for the Chinese market. And only sometime later this summer they will enter the global market, already together with older models.

Intel is finally ready to release its desktop graphics cards.  A link to Arc A3 maps appeared on the company's website

Considering that the Radeon RX 7000 and GeForce RTX 40 graphics cards are due out in the second half of the year, and the first ones may appear in the fall, it is completely unclear how and with what Intel adapters will compete. In such a situation, the company will only have to significantly reduce prices.

As for the first models, they should be Arc A310 and A380 with 512 and 1024 cores, respectively.

Continue Reading

Most Popular