NVIDIA GeForce GTX 570 Review

[SKYMTL]

If we look back on the shape of the GPU market not two months ago, not many would have predicted that 2010’s Christmas shopping season would be one of the most exciting in recent memory. A few weeks ago NVIDIA surprised everyone by preempting AMD’s HD 6900 series launch with a flagship card of their own called the GTX 580. To make matters even more interesting, this new card was widely available at launch with plenty of stock in the retail channels and featured improved thermals and acoustics over the product it usurped. The remainder of this year will also see the launch of the aforementioned HD 6900 series as well as yet another enthusiast-grade NVIDIA card: the GTX 570.

Not satisfied to simply go quietly into the night while AMD’s Cayman products creep ever closer, NVIDIA has decided to attack another price point as quickly as possible. Performance will naturally be paramount in order to fend off whatever Radeons are waiting in the wings so the GTX 570 comes packing some incredible specifications into a sub-$400 card. It also incorporates all of the features from the higher end GTX 580 such as a vapor chamber-based heatsink and architectural improvements which lead to an overall increase in rendering efficiency. The result is a card which is roughly 25% faster than the GTX 470.

NVIDIA has been very, very quiet about this launch and for once there have been very few leaks. This is all to keep AMD guessing which will in turn give them less time to react with any changes to upcoming architectures. However a plan like this can also act as a double edged sword since actual availability can suffer if board partners are kept in the dark. Nonetheless, it looks to us like there will be decent numbers of GTX 570 SKUs making their way onto store shelves as you read this. We just have to remember that due to the $350 price of this card, it will surely be a best seller during the Christmas shopping season.

With NVIDIA quickly rolling out their 500-series, it seems increasingly like they have headed off AMD’s upcoming juggernaut at the pass. Considering the GTX 570 uses the same DNA as the GTX 580, we are definitely expecting great things from it.

 

[SKYMTL]

GF110; Something Old, Something Borrowed….

GF110; Something Old, Something Borrowed….

At the heart of the GTX 500-series beats a revised GF100 core which has been christened GF110 for this somewhat new generation of products. Since most of the features are the same between the two cores, we invite you to visit our dedicated GF100 article where we go into all of the advancements the Fermi architecture brings to the table. This section meanwhile will used to discussing the new items NVIDIA has built into the GF110 which allow it to outpace the older series on a number of levels.

One of NVIDIA’s main focuses for the GF110 was to increase the overall performance per watt over past iterations of Fermi. As it stood the GTX 480 was and still is a massively powerful card but it hit up against a power consumption wall due to a number of inefficiencies within the architecture.

To address some of the inherent power consumption issues Fermi exhibited, NVIDIA basically rearranged their layout so more of the faster, higher leakage transistors were placed on the critical rendering paths instead of being used for periphery tasks. Meanwhile, the slower low leakage transistors were placed where speed wasn’t a primary concern.

Strategically distributing the transistors in this way allows for a small speed-up in overall rendering performance. More importantly it also means the fastest transistors will now be fully utilized instead of being used for non critical tasks and thus lowering overall performance per watt.

From a high-level architectural standpoint, not much has changed between the GF100 and the GF110. The layout still features 512 CUDA cores, 64 Texture Units, 48 ROPs and six 64-bit GDDR5 memory controllers along with a hefty count of 3 billion transistors.

Where the GF110 differs from the GF100 is its ability to feature higher clock speeds while enabling all 16 Streaming Multiprocessors without pushing astronomical power consumption figures. This means the flagship GTX 580 will finally feature the full 512 cores many were hoping to see from the GTX 480 while the GTX 570 will have a single SM disabled.

By enabling this extra SM, the higher end GTX 580 gains more than just 32 CUDA cores. Four additional texture units along with their associated cache structure are also added to the equation but the most important addition-for DX11 apps at least- is the single PolyMorph Engine which is attached to the SM. The extra tessellator contained in this unit could go a long way towards helping out with the higher tessellation workloads which second generation DX11 games will feature.

 

[SKYMTL]

…And Many Things New

…And Many Things New

While NVIDIA was prioritizing critical rendering paths with higher speed transistors, some additional improvements were made to the underlying Fermi architecture as well. Most of these were relatively minor in the grand scheme of things but when compounded they can lead to some impressive performance increases in certain scenarios.

While the GF100 architecture fully supported FP16 HDR texture formats, the GF110 now has the capability to process these textures at a higher throughput. With this “full speed” FP16 texture filtering, the TMUs within cards like the GTX 580 are able to operate at a slightly higher efficiency than the previous generation.

In addition to the increased FP16 performance, two other areas were prioritized: z-cull efficiency and the L1 cache. An overall bump in z-cull efficiency will mean quicker processing of certain draw calls which are sent through the GF110’s Raster Engine. In addition, the 64KB of shared memory and L1 cache now has additional configuration options which can in theory augment texture performance by speeding up the texture units’ communication with the rest of the rendering pipeline.

Some may remember that NVIDIA’s GF104 core handled instructions in a slightly different manner than the higher end Fermi GPUs. This was mostly due to the larger number of cores per Streaming Multiprocessor. However, since the GF110 maintains the same 32 cores per SM, it maintains the same instruction issuance as the GF100.

GTX 580 performance increases illustrated for demonstration purposes only​

All of the advances we have been talking about can amount to some significant improvements in framerates across a cross section of games. Since the architectural enhancements are specifically targeted to certain areas of the rendering process, the impact they have varies from one game to the next. However, in DX11 apps or games that use a large amount of texture data, the focus upon high level geometry performance does prove to be beneficial.

The focus upon transistor layout also allowed for higher clock speeds so when the butcher’s bill is tallied, some games can see a performance jump of about 30%.

 

[SKYMTL]

NVIDIA’s Focus upon Acoustics and Heat

NVIDIA’s Focus upon Acoustics and Heat

One of the main critiques leveled against the GF100-based cards was their rampant heat production leading to high fan speeds. Using a GTX 480 in particular was never what we would call a completely silent experience but many of the critiques were blown a bit out of proportion in our opinion. Nonetheless, NVIDIA went back to the drawing board in order to completely redesign their high-end cooling solutions in order to focus upon lowering heat and shrinking the acoustical profile of their flagship products.

The first step was to discard the inefficient and massive heatpipe setup which came with the GTX 480. The replacement is a vapor-chamber cooler that quickly whisks away heat from the core in order to distribute it evenly across a large aluminum fin array.

A vapor chamber design is broken into a number of components. There is a copper base-plate that makes direct contact with the core so the water contained therein will quickly vaporize on the evaporator and make its way through the chamber to the condenser. The condenser wick is placed up against the upper cover in order to disperse the heat generated as the hot water vapor condenses. Finally, there is a transportation wick for liquid return that is used to move the condensed water back to the vaporization wick. Think of this like a massive flat heatpipe which is infinitely more efficient at dispersing heat over a large surface area.

In order for this method to be effective, the entire chamber needs to be sealed and put under vacuum. This is due to the fact that water vaporizes much easier in an environment with extremely low air pressure. Thus, it is very important that a vapor chamber cooler is well made without any manufacturing defects or the air pressure within the vapor chamber will decrease and this will result in lowered heat dissipation capabilities.

Other than the custom vapor chamber, there are additional features packed into the GTX 580 what are used to lower its overall noise profile.

Through testing, NVIDIA found the GTX 480’s fan had blades that were vibrating slightly which created certain tonal characteristics that could be annoying to some people. To rectify this, a tension ring was added around the fan’s edge and hub’s bottom was filled in to reduce friction. The whole setup was then placed closer to the fin array to allow the fan to spin at lower speeds when pushing cool air into the heatsink. For SLI systems, a small indent was added shroud’s rearmost portion which should allow for slightly better airflow.

The result of all these changes is of course a temperature reduction but additional benefits have been reaped in the acoustical department as well. According to NVIDIA, the GTX 500-series will be quieter than the GTX 285 without any sacrifices in the thermal department due to “new” adaptive fan control. Basically, this mean the fan will ramp up in a much more linear fashion rather than the sudden jumps we saw with past cards. Since it adjusts fan speed in real time based on core temperature it never really needs to run at high speeds since temperatures are kept in check from the get-go.

Most ultra high end cards exhibit absolutely epic power consumption numbers and the GF100 was no different. However, in order to better manage voltage fluctuations and power spikes, NVIDIA has implemented support for real-time monitoring of 12V voltage readings. If the reading on any of the PSU’s rails drops or increases by a set amount, the VRM modules on the card will immediately compensate. This should not only lower overall power consumption but also protect the card against “dirty” power from cheap PSUs.

In addition, the hardware monitoring modules are able to work hand in hand with the GF110 core to dynamically adjust performance to keep consumption within NVIDIA’s specifications. Overclocking won’t be affected by this unless you want to push things to obscene limits with exotic cooling but power “viruses” like Furmark which stress the GPU beyond its design limits will see their effectiveness reduced.

 

[SKYMTL]

The GTX 570 in NVIDIA’s New Lineup

The GTX 570 in NVIDIA’s New Lineup

With the recently released GTX 580 taking up the flagship role in NVIDIA’s current product stack; it was up to the GTX 480 and GTX 470 to set up shop between the $250 and $475 price points. Below them came the massively popular GTX 460 and of course the rest of the lower end and much more affordable cards. One of the main issues with this approach was the massive gap between the cards allowed AMD to slot their cards into specific portions of the market and sales soared as a result.

Now that NVIDIA has begun rolling out their 500-series, it was only natural that the GF110 core would eventually be adapted for lower price points. The GTX 570 is meant to eventually replace the GTX 470 and its introduction also signals the GTX 480’s demise.

The GTX 570 is actually quite a bit different from the GTX 580 in a number of ways even though both feature the 3 billion transistor GF100 core. While the GTX 580 comes with the full bore 512 CUDA cores, this new card takes elements from the GTX 480 and the GTX 470 to create a real beast of a product at a relatively low $350 price. Much like the GTX 480, it has one Steaming Multiprocessor disabled which effectively eliminates 32 cores and four texture units. However, due to the higher clock speeds and other architectural improvements of the GTX 570 we should see slightly better performance than the outgoing card in situations that aren’t bandwidth limited.

This card’s high level architecture may be a carbon copy of the GTX 480 but the memory allotment resembles that of a GTX 470 with 1.28GB of GDDR5. This also means a single 64-bit memory controller and one ROP array are also eliminated while the L2 cache count goes down to 640KB.

To see NVIDIA hitting the $350 price point with a card that should compete blow for blow against the GTX 480 really is a great thing to see. It looks like cuts were made in logical places to bring the overall cost of this card to a point that is more than affordable for the vast majority of gamers.

It should be more than obvious that NVIDIA’s GTX 570 is meant to be a true leader in the price / performance field. However, this does open up a relatively large gap between the two 500-series products while AND will likely be more than happy to fill with the upcoming Cayman cards.

In addition to some great GPU releases, the latter part of this year has also been rife with accusations from both camps regarding image quality, optimizations and other hot topic items. We here at Hardware Canucks have listened to both sides of every argument and for the most part have not been able to replicate any image quality reducing optimizations in the games we use in our benchmarking suite. Granted, these optimizations may still be running even in the games you will see in this review but at not one time did we see a perceptible difference between NVIDIA and AMD cards within any of these games.

 

[SKYMTL]

A Closer Look at the GTX 570

A Closer Look at the GTX 570

​

From an outside perspective, the GTX 570 is a spitting image of its bigger brother; the GTX 580. It’s a total of 10.5” long and uses a full-length fan shroud that will usually have a manufacturer’s decal applied to it.

​

Along with the indent around the fan to increase overall airflow when the card is in SLI, there aren’t any outwards signs that this is a GTX 570 until you get a look at the side. Instead of an 8-pin + 6-pin setup, the lower power requirements of this card have allowed NVIDIA to use a dual 6-pin setup. Meanwhile, there is still support for triple SLI through a pair of SLI connectors.

​

The backplate connector features the usual dual DVI connectors as well as a single mini HDMI 1.4 output which allows for full stereo 3D support.

​

The back of the GTX 570’s PCB looks like that of its bigger brother but there are a number of soldering points that don’t have components attached to them.

 

[SKYMTL]

Under the Hood

Under the Hood

​

The vapor chamber on the GTX 570 is similar to the one on the GTX 580 but there are a number of differences as well. The overall fin count is identical but there are fewer openings in the top of the array which should give slightly more directionality to the airflow provided by the fan. We can also see the tip of a heatpipe rising through the heatsink’s side which is simply there to cap off the vapor chamber.

​

Aside from the main heatsink, there is also a full length retention bracket made out of anodized aluminum. This is supposed to disperse heat and act as a stiffener to compensate for the excess weight of the vapor chamber.

​

 

[SKYMTL]

Test System & Setup

Test System & Setup

Processor: Intel Core i7 920(ES) @ 4.0Ghz (Turbo Mode Enabled)
Memory: Corsair 3x2GB Dominator DDR3 1600Mhz
Motherboard: Gigabyte EX58-UD5
Cooling: CoolIT Boreas mTEC + Scythe Fan Controller (Off for Power Consumption tests)
Disk Drive: Pioneer DVD Writer
Hard Drive: Western Digital Caviar Black 640GB
Power Supply: Corsair HX1000W
Monitor: Samsung 305T 30” widescreen LCD
OS: Windows 7 Ultimate N x64 SP1


Graphics Cards:

NVIDIA GTX 570
NVIDIA GTX 580
NVIDIA GTX 480 (Ref)
NVIDIA GTX 470 (Ref)

ATI HD 5970 2GB (Ref)
ATI HD 5870 1GB (Ref)
ATI HD 5850 1GB (Ref)
ATI HD 6870 1GB (Ref)
ATI HD 6850 1GB (Ref)



Drivers:

NVIDIA 263.09 WHQL
ATI 10.10 WHQL


Applications Used:

3DMark 11
Aliens Versus Predator
Battlefield: Bad Company 2
DiRT 2
Just Cause 2
Lost Planet
Metro 2033
Starcraft 2
Unigine: Heaven


*Notes:

FOR MORE INFORMATION ABOUT OUR BENCHMARKING PROCESS PLEASE SEE THIS ARTICLE

- All games tested have been patched to their latest version

- The OS has had all the latest hotfixes and updates installed

- All scores you see are the averages after 3 benchmark runs

All game-specific methodologies are explained above the graphs for each game

All IQ settings were adjusted in-game

 

[SKYMTL]

3DMark 11 (DX11)

3DMark 11 (DX11)

3DMark 11 is the latest in a long line of synthetic benchmarking programs from the Futuremark Corporation. This is their first foray into the DX11 rendering field and the result is a program that incorporates all of the latest techniques into a stunning display of imagery. Tessellation, depth of field, HDR, OpenCL physics and many others are on display here. In the benchmarks below we have included the results (at default settings) for both the Performance and Extreme presets.


Performance Preset

Extreme Preset

Note: the chart above shows two GTX 570 cards; the upper one should read GTX 580. We are in the process of doing the necessary revisions.

 

[SKYMTL]

Aliens Versus Predator (DX11)

Aliens Versus Predator (DX11)

When benchmarking Aliens Versus Predator, we played through the whole game in order to find a section which represents a “worst case” scenario. We finally decided to include “The Refinery” level which includes a large open space and several visual features that really tax a GPU. For this run-through, we start from within the first tunnel, make our way over the bridge on the right (blowing up several propane tanks in the process), head back over the bridge and finally climb the tower until the first run-in with an Alien. In total, the time spent is about four minutes per run. Framerates are recorded with FRAPS.


1680 x 1050

1920 x 1200

2560 x 1600