A new generation of graphics cards has apparently been long awaited in the autumn of 2020, when Nvidia’s launch of the Geforce RTX 3080 and RTX 3090 was plagued by demand that far exceeds supply. Nvidia recently announced that the launch of the Geforce RTX 3070 for desktops is delayed, from the previous date 15 October to the new premiere on 29 October, where the background to the delay is precisely to achieve better access from the start.
However, problems with saturating demand do not prevent Nvidia from continuing to invest in graphics cards with the Ampere architecture under the hood. The company is also preparing a version of Geforce RTX 3070 for laptops, according to a leaked product image that shows the graphics circuit in a system intended for quality testing (via the Twitter profile 9550pro). In addition to the circuit itself, some additional details can be discerned in the image. The circuit name is specified as GN20-E5-A1, which is a reference to the graphics circuit GA104 used in the Geforce RTX 3070 desktop.
Among the details are markings on the memory circuits that indicate H56C8H24AIR, which are GDDR6 modules from SK Hynix. This is consistent with the fact that the Geforce RTX 3070, unlike the RTX 3080 and RTX 3090 siblings, uses GDDR6 memory instead of GDDR6X. Details that do not appear are how many CUDA cores the mobile circuit is equipped with, and what clock frequencies the graphics processor runs at.
Specification | Geforce RTX 3070 | Geforce RTX 3070 Max-Q* | Geforce RTX 2070 Max-Q | Geforce RTX 2070 |
---|---|---|---|---|
Architecture | Ampere | Ampere | Turing | Turing |
Graphics circuit | GA104 | GA104* | TU106 | TU106 |
CUDA cores | 5 888 | 5 888* | 2 304 | 2 304 |
Clock frequencies | Bass frequency: | Unknown | Bass frequency: | Bass frequency: |
Graphics memory | 8 GB GDDR6 | 8 GB GDDR6* | 8 GB GDDR6 | 8 GB GDDR6 |
Memory interface | 256-bit | 256-bit * | 256-bit | 256-bit |
Clock frequencies graphics memory | 16 000 MHz | 12 000 MHz* | 12 000 MHz | 14 000 Mhz |
Memory bandwidth | 512 GB/s | 384 GB/s* | 384 GB/s | 448 GB/s |
Manufacturing technology | 8 nm | 8 nm | 12 nm | 12 nm |
TDP | 220 W | Unknown | 80–90 W | 115 W |
* Specifications are based on leaked information and should be considered speculative until confirmed.
SK Hynix H56C8H24AIR operates at a speed of 12,000 MHz elsewhere where it is used, which if mobile Geforce RTX 3070 also uses 256-bit memory interface gives a bandwidth of 384 GB / s. This can be compared to the stationary equivalent whose graphics memory operates at 16,000 MHz over 256-bit memory interfaces, for a total bandwidth of 512 GB / s.
Nvidia’s graphics circuits, which are developed specifically for thinner laptops, are called Max-Q, where clock frequencies are typically reduced to their desktop counterparts. The GA104 circuit is fully equipped with 6,144 CUDA cores, and the stationary variant of the Geforce RTX 3070 has 5,888 of these activated. The base frequency tuffs at 1,500 MHz while the boost frequencies reach 1,730 MHz.
The Geforce RTX 3070 Max-Q will probably come with lower clock frequencies than that, but exactly what these land in remains to be seen. As competitor AMD unveils the desktop graphics cards in the Radeon RX 6000 series only towards the end of October without any indication of new mobile graphics chips, Nvidia’s Geforce RTX 3070 Max-Q and any siblings will probably rule alone on the performance throne when it takes place in upcoming laptops.