With Nvidia’s GeForce GTX 560 Ti, a good thing
gets better. Although nearly architecturally
identical to the GF104 GPU that powered
the GTX 460, the GF114 GPU at the heart
of the GTX 560 Ti has all eight of its SMs
(streaming multiprocessors) enabled. Higher clocks,
a refined cooler, and overclocking headroom combine
to give enthusiasts even more for their $250.
gets better. Although nearly architecturally
identical to the GF104 GPU that powered
the GTX 460, the GF114 GPU at the heart
of the GTX 560 Ti has all eight of its SMs
(streaming multiprocessors) enabled. Higher clocks,
a refined cooler, and overclocking headroom combine
to give enthusiasts even more for their $250.
For those who’ve followed the video
card market for several years, Nvidia
and the “Ti” designation will spark
some memories. In the early 2000s,
Nvidia designated a number of cards in its
GeForce4 series with a “Ti” designation.
Remember the GeForce4 Ti 4800?
Nvidia hasn’t used Ti—short for
“titanium”—for several years, which is
why the recent release of Nvidia’s GTX
560 Ti video board may have video card
enthusiasts thinking back to the past.
Nvidia hasn’t officially explained
why it has decided to resurrect the Ti
suffix with the GTX 560 Ti, but it
does appear that the new video card
fits one of the old definitions of Ti
cards, providing a good mix of value
and performance. The GTX 560 Ti
will be a sub-$250 board, placing it in
what Justin Walker, product manager
at Nvidia, says is the “sweet spot” of the
video card market.
Changes In The GPU
The magic behind the GTX 560 Ti
is the GF114 GPU. Nvidia released
the GF114 as an upgrade to last July’s
GF104, following a similar pattern to
its release of the GF110 in late 2010,
which was an upgrade to the GF100.
All four of these chips are part of the
Fermi family of Nvidia GPUs, but the
GF104 and GF114 have what Walker
calls a streamlined design compared to
the GF100 and GF110, which are aimed
at high-end markets.
The GF114 and the GF104
are similar in many aspects. In fact,
Walker says, the basic architecture
in both chips is identical. However,
Nvidia has increased the clock speed
in the GF114 over the GF104, and it
has fully enabled the SM (streaming
multiprocessor) units, improving
The magic behind the GTX 560 Ti
is the GF114 GPU. Nvidia released
the GF114 as an upgrade to last July’s
GF104, following a similar pattern to
its release of the GF110 in late 2010,
which was an upgrade to the GF100.
All four of these chips are part of the
Fermi family of Nvidia GPUs, but the
GF104 and GF114 have what Walker
calls a streamlined design compared to
the GF100 and GF110, which are aimed
at high-end markets.
The GF114 and the GF104
are similar in many aspects. In fact,
Walker says, the basic architecture
in both chips is identical. However,
Nvidia has increased the clock speed
in the GF114 over the GF104, and it
has fully enabled the SM (streaming
multiprocessor) units, improving
performance. The GF114 enables all
eight of the GPU’s SMs, while the
GF104 enabled only seven. Leaving one
SM disabled caused the GF104 to lose
a little bit of its geometry, shading, and
texture processing power.
Although the GF104 and GF114
each contain 1.95 billion transistors,
Nvidia has optimized the GF114’s
transistors. This procedure allows
Nvidia to enable all of the chip’s
functional units, as well as raise clock
speeds. Essentially, Walker says,
Nvidia has tweaked the distribution
of the transistors to place fasterperforming
transistors in areas of
the GPU where speed is of utmost
importance, even though such transistors
tend to leak. Nvidia placed
the slower, low-leakage transistors on
portions of the chip where premium
speed isn’t as important, allowing
for better power savings, less heat
generation, and better overall performance
per watt.
“What we’ve done is re-engineer the
GPU,” Walker says. “We looked at areas
on the die that are critical to speed. In
that way, we’re optimizing performance
and performance per watt.”
The GF114 and GF104 each use the
same pin structure, meaning video card
makers could place the GF114 GPU into
older video card designs. However, some
older card designs may not be able to
handle the slightly higher power demands
eight of the GPU’s SMs, while the
GF104 enabled only seven. Leaving one
SM disabled caused the GF104 to lose
a little bit of its geometry, shading, and
texture processing power.
Although the GF104 and GF114
each contain 1.95 billion transistors,
Nvidia has optimized the GF114’s
transistors. This procedure allows
Nvidia to enable all of the chip’s
functional units, as well as raise clock
speeds. Essentially, Walker says,
Nvidia has tweaked the distribution
of the transistors to place fasterperforming
transistors in areas of
the GPU where speed is of utmost
importance, even though such transistors
tend to leak. Nvidia placed
the slower, low-leakage transistors on
portions of the chip where premium
speed isn’t as important, allowing
for better power savings, less heat
generation, and better overall performance
per watt.
“What we’ve done is re-engineer the
GPU,” Walker says. “We looked at areas
on the die that are critical to speed. In
that way, we’re optimizing performance
and performance per watt.”
The GF114 and GF104 each use the
same pin structure, meaning video card
makers could place the GF114 GPU into
older video card designs. However, some
older card designs may not be able to
handle the slightly higher power demands
GF114 Architecture Explained
Single SM From GF114
Nvidia’s GF114 GPU is included with the recently announced
GeForce GTX 560 Ti. With the GF114, Nvidia hasn’t changed the
chip’s basic architecture from the GF104, but the company has
optimized and activated all of the components. The GF104 GPU
appears in the GTX 460 video card.
“Compared to the GF104 and GTX 460, we’re using the same
basic microarchitecture,” Nvidia product manager Justin Walker says.
“The block architecture is going to look very similar to the 460.”
As seen in the GF114 Architecture Overview image, you’ll find
four SMs in each of the two GPCs (graphics processing clusters,
outlined in white) for the GF114. The GF104 architecture also
contained two GPCs.
Within each of the eight SMs, you’ll find two dispatch units per
warp scheduler.
Each SM in the GF114 contains 48 CUDA cores, 16 load/
store units, eight SFUs (special function units), and eight textureprocessing
units.
CUDA is an Nvidia technology that allows for parallel processing
by making use of the GPU’s processing power in tandem with the
CPU. The CUDA cores can work simultaneously to process data,
and by using more CUDA cores in the GF114, the processor’s
overall efficiency increases, Walker says.
The basic architecture of each SM in the GF104 is the same
as the GF114, but one of the GF104’s SMs is disabled, meaning the GF104 has only 336 active CUDA cores. In the
GF104 Architecture Overview image, you can see the disabled SM is grayed out.
“With the addition of the polymorph engine and with higher clock speeds, we’re boosting our tessellation performance,”
Walker says. “As we add more cores, we can add more tessellation performance.”
In the GF114 (just as with the GF104), Nvidia has included a 512KB L2 cache, four 64-bit memory system controllers,
and a 256-bit memory bus connection to the GPU from local graphics memory. ■
Source: Nvidia
GeForce GTX 560 Ti. With the GF114, Nvidia hasn’t changed the
chip’s basic architecture from the GF104, but the company has
optimized and activated all of the components. The GF104 GPU
appears in the GTX 460 video card.
“Compared to the GF104 and GTX 460, we’re using the same
basic microarchitecture,” Nvidia product manager Justin Walker says.
“The block architecture is going to look very similar to the 460.”
As seen in the GF114 Architecture Overview image, you’ll find
four SMs in each of the two GPCs (graphics processing clusters,
outlined in white) for the GF114. The GF104 architecture also
contained two GPCs.
Within each of the eight SMs, you’ll find two dispatch units per
warp scheduler.
Each SM in the GF114 contains 48 CUDA cores, 16 load/
store units, eight SFUs (special function units), and eight textureprocessing
units.
CUDA is an Nvidia technology that allows for parallel processing
by making use of the GPU’s processing power in tandem with the
CPU. The CUDA cores can work simultaneously to process data,
and by using more CUDA cores in the GF114, the processor’s
overall efficiency increases, Walker says.
The basic architecture of each SM in the GF104 is the same
as the GF114, but one of the GF104’s SMs is disabled, meaning the GF104 has only 336 active CUDA cores. In the
GF104 Architecture Overview image, you can see the disabled SM is grayed out.
“With the addition of the polymorph engine and with higher clock speeds, we’re boosting our tessellation performance,”
Walker says. “As we add more cores, we can add more tessellation performance.”
In the GF114 (just as with the GF104), Nvidia has included a 512KB L2 cache, four 64-bit memory system controllers,
and a 256-bit memory bus connection to the GPU from local graphics memory. ■
Source: Nvidia
manufacturers and gamers more aware of the
overclocking capabilities of the GTX 560 Ti.
“[The board manufacturers] love
overclocking as much as the gamers do,”
Walker says. “A lot of the stuff we hear
back from [the board manufacturers]
has to do with overclocking. . . . With
the 460, this is the first time we had
introduced something with such a high
amount of overclocking headroom. With
the 560, since they had already seen it
before, they’ve really jumped on it.”
overclocking capabilities of the GTX 560 Ti.
“[The board manufacturers] love
overclocking as much as the gamers do,”
Walker says. “A lot of the stuff we hear
back from [the board manufacturers]
has to do with overclocking. . . . With
the 460, this is the first time we had
introduced something with such a high
amount of overclocking headroom. With
the 560, since they had already seen it
before, they’ve really jumped on it.”
More Performance, Minimal Cost
Nvidia has been able to implement
these changes with only a very small
power increase. Despite a stock core clock
increase of over 20%, the GTX 560 Ti
manages a TDP of 170 watts—only 10
watts higher than the GTX 460.
“The overall performance gains far outweigh
the power increases,” Walker says.
Walker says the GTX 560 Ti’s improvements
in clock speeds and the full
complement of enabled SMs at the GTX
560 Ti’s disposal result in an overall
performance boost of about 33% over
the GTX 460. With the additional power
requirements calculated, he says Nvidia
measures the improvement in performance
per watt with the GTX 560 Ti at about 21%.
“The performance improvement is
about what we expected,” Walker says.
increase of over 20%, the GTX 560 Ti
manages a TDP of 170 watts—only 10
watts higher than the GTX 460.
“The overall performance gains far outweigh
the power increases,” Walker says.
Walker says the GTX 560 Ti’s improvements
in clock speeds and the full
complement of enabled SMs at the GTX
560 Ti’s disposal result in an overall
performance boost of about 33% over
the GTX 460. With the additional power
requirements calculated, he says Nvidia
measures the improvement in performance
per watt with the GTX 560 Ti at about 21%.
“The performance improvement is
about what we expected,” Walker says.
“If you . . . scale [the numbers] up,
if you take that as a general measurement
of horsepower, you get what
we expected.
“Really, [our] goal for the GF114 was
to fine-tune the chip for performance and
performance per watt.”
Nvidia has also tweaked the physical
design of the GTX 560 Ti. The
GTX 560 Ti measures 9 inches in
length, while the GTX 460 is a bit
shorter, at 8.25 inches. The GTX 560
Ti uses a slightly larger heatsink and
fan combination than its predecessor,
allowing for better cooling. You’ll find
an extra heatpipe in the GTX 560 Ti,
too, giving this card three heatpipes.
if you take that as a general measurement
of horsepower, you get what
we expected.
“Really, [our] goal for the GF114 was
to fine-tune the chip for performance and
performance per watt.”
Nvidia has also tweaked the physical
design of the GTX 560 Ti. The
GTX 560 Ti measures 9 inches in
length, while the GTX 460 is a bit
shorter, at 8.25 inches. The GTX 560
Ti uses a slightly larger heatsink and
fan combination than its predecessor,
allowing for better cooling. You’ll find
an extra heatpipe in the GTX 560 Ti,
too, giving this card three heatpipes.
“On the board level, we added new
thermals and a slightly larger power
supply,” Walker says. “It helps enable
more overclocking. That’s why we’re seeing
such a nice overclocking performance.”
thermals and a slightly larger power
supply,” Walker says. “It helps enable
more overclocking. That’s why we’re seeing
such a nice overclocking performance.”
The Sweet Spot
Nvidia released its GTX 580 and
GTX 570 video cards in November and
December 2010, respectively, and both
cards are based on the GF110 GPU. Both
are high-end cards—about $500 for the
GTX 580 and about $350 for the GTX
570. With the price of the GTX 460
dropping below $200, Nvidia has filled
the $250 space with the GTX 560 Ti
“We get excited about all the cool
stuff we introduced in the 580, but not
everybody can afford it,” Walker says.
“The great thing about this pricing area is
you can get a card now that can.”
Walker says Nvidia doesn’t want to
limit overclocking to a certain segment
of the video card market. Users
wanting to overclock aren’t limited to
those who will buy only top-end video
cards, he says, which is why Nvidia
likes to include good overclocking
capabilities in video cards like the
GTX 560 Ti.
Walker also points to the GTX 560 Ti’s
support for DirectX 11, eight tessellationengines, and a CUDA core clock speed to
1.645GHz as key features Nvidia built into
the video card to ensure long life.
“The 560 Ti is meant to have a bit of
legs,” he says. “We built in some futurelooking
technologies.”
Building “future-looking” technologies
into a video card that Nvidia named after
cards from the past gives the GTX 560
Ti a nice balance. Throw in the 1GHz
overclocking capabilities, and it’s easy
to see why Nvidia thinks its latest $250
video card will fit into the sweet spot of
the market.
stuff we introduced in the 580, but not
everybody can afford it,” Walker says.
“The great thing about this pricing area is
you can get a card now that can.”
Walker says Nvidia doesn’t want to
limit overclocking to a certain segment
of the video card market. Users
wanting to overclock aren’t limited to
those who will buy only top-end video
cards, he says, which is why Nvidia
likes to include good overclocking
capabilities in video cards like the
GTX 560 Ti.
Walker also points to the GTX 560 Ti’s
support for DirectX 11, eight tessellationengines, and a CUDA core clock speed to
1.645GHz as key features Nvidia built into
the video card to ensure long life.
“The 560 Ti is meant to have a bit of
legs,” he says. “We built in some futurelooking
technologies.”
Building “future-looking” technologies
into a video card that Nvidia named after
cards from the past gives the GTX 560
Ti a nice balance. Throw in the 1GHz
overclocking capabilities, and it’s easy
to see why Nvidia thinks its latest $250
video card will fit into the sweet spot of
the market.
Article From Computer Power User Magazine
0 comments:
Post a Comment