It’s a situation that made reviews difficult at Custom PC, as
we had a print deadline at the same time. We had time to test
the AMD GPUs, but didn’t get any RTX Super samples in time
for full review last month. We ended up not being able to give
scores to the AMD GPUs until we’d tested the Nvidia
competition, and since then AMD has also announced price
cuts, while the price of pre-Super RTX GPUs has dropped a
bit. Are you following? Good.
Thankfully, the dust is now starting to settle, and we’re
now at a point where we can properly put the RTX 2060 and
2070 Super through their paces, and also make some
recommendations. Over the next few pages, we’ll take you
through the new Super GPUs, and recommend which GPU
you should buy for your budget.
What’s so super?
There’s essentially nothing new in the ‘Super’ RTX GPUs.
They’re based on the same 445mm² TU106 and 545mm²
TU104 chips we saw in last year’s RTX graphics cards,
but the model numbers have shuffled along a bit, with
different numbers of streaming multiprocessors enabled
on the various chips.
Let’s start with the RTX
2060 Super; like the original
RTX 2060, it’s based on the
same TU106 GPU as the RTX- However, the RTX
2060 Super is much closer to
the original RTX 2070 in
terms of spec. It has 34 of its
SMs enabled, giving it 2,
stream processors, 34 RT
cores for hardware ray
tracing and 272 Tensor cores
for Nvidia’s deep learning
features, such as its DLSS
anti-aliasing mode.
Comparatively, the original
RTX 2060 only has 1,
stream processors, 30 RT
cores and 240 Tensor cores,
while the RTX 2070 is only a
small step up, with 2,
stream processors, 36 RT
cores and 288 Tensor cores.
Just as importantly, the
RTX 2060 Super now comes
with 8GB of 14GHz (effective)
GDDR6 memory, unlike the
6GB of its predecessor.
What’s more, it’s attached to
a 256-bit interface, rather
than its predecessors 192-bit
interface, and it has the full
count of 64 ROPs, rather than - This move brings the RTX
2060 Super into line with its three bigger siblings, with a total
memory bandwidth of 448GB/sec, compared to just
336GB/sec for the RTX 2060.
Another result of this move is that board manufacturers
don’t necessarily need to make a new PCB for the RTX 2060
Super – the ones for their RTX 2070 cards should be fine. It’s
a situation that effectively means the RTX 2060 Super is
much more practical to water-cool than its predecessor, as
Antony Leather discusses on p106. Clock speeds have also
changed slightly, with a 1470MHz base clock and 1650MHz
boost clock, compared to 1365MHz and 1680MHz
respectively for the original RTX 2060.
Next up is the RTX 2070 Super, which moves from the
TU106 GPU of its predecessor to the TU104 GPU used by the
RTX 2080. You can think of it as a bit like the GTX 1070 Ti in
that respect. The newly announced RTX 2080 Super (see
p12) has all 48 of the TU104 chip’s SMs enabled, giving it
3,072 stream processors, a small step up from the original
RTX 2080’s 46 SMs containing 2,944 stream processors.
The RTX 2070 Super has 40 of the TU104 GPU’s SMs
enabled, giving it 2,560 stream processors, 40 RT cores
and 320 Tensor cores.
SUPER MAN
+ 2060 Super
handles 2,560 x
1,440 ray tracing
+ 2070 Super is
extremely fast
+ Futureproofing
SUPER HANS- RX 5700 XT
quicker than 2060
Super - Expensive
- Palit Game Rock
card is massive
Nvidia’s TU104 chip has all its parts enabled in the RTX 2080 Super.
Remove eight SMs and you get the RTX 2070 Super