At times, NVIDIA may be a rather predictable business. When they announced the fantastic GeForce GTX 580 last month, we knew it was only a matter of time before further GTX 500 series cards joined their product selection. It’s practically unheard of for them to offer just one product based on a high-end GPU.
Now, less than a month after the GTX 580’s debut, that moment has here. The GeForce GTX 570, the second card to use NVIDIA’s new GF110 GPU, is being released today. The GTX 570, a less expensive and less effective graphics card, is the literal and spiritual successor of the GTX 480 and the spiritual successor to the GTX 470. Additionally, at $350, it bridges the substantial gap between the current GTX 470 and GTX 580 cards from NVIDIA.
How does NVIDIA’s newest and second-best perform, and can the GTX 580 consider it a worthy sibling? Let’s investigate.
|GTX 580||GTX 570||GTX 480||GTX 470|
|Texture Address / Filtering||64/64||60/60||60/60||56/56|
|Memory Clock||1002MHz (4008MHz data rate) GDDR5||950MHz (3800MHz data rate) GDDR5||924MHz (3696MHz data rate) GDDR5||837MHz (3348MHz data rate) GDDR5|
|Memory Bus Width||384-bit||320-bit||384-bit||320-bit|
|FP64||1/8 FP32||1/8 FP32||1/8 FP32||1/8 FP32|
|Manufacturing Process||TSMC 40nm||TSMC 40nm||TSMC 40nm||TSMC 40nm|
The GTX 570, or any other GF100 card for that matter, is most likely the closest thing we’ll see to a GF110 version of the GTX 480. We’ve previously seen NVIDIA choose for a fully functional GF110 architecture in the GTX 580, so the GTX 570 builds on that. The greater yields provided by the GF110 design and TSMC’s manufacturing upgrades allowed for this. The final product combines the shader count from the GTX 480 with the ROP count and memory bus from the GTX 470. As a result, performance is much more similar to the GTX 480 than the GTX 470 due to the higher clockspeed.
The GTX 570 matches the GTX 480 at 480 active CUDA Cores and 60 texture units when 15 of 16 SMs are activated. In order to compensate for the smaller ROP/memory blocks and to benefit from the GF110’s lower leakage at higher rates (a side note, why the weird clocks lately? ), the core clock is 732MHz, 32MHz (4.5%) faster than the GTX 480. Consider the PLL). However, this time the memory clock is up to 950MHz (3.8GHz data rate), which is 113MHz (13.5%) over the GTX 470. The memory system is still using the same 320bit (64bit x 5) memory bus and 10 memory chip architecture we saw on the GTX 470. Although memory rates are 26MHz faster than the GTX 480, this is not nearly enough to compensate for the smaller memory bus. It has the same 60 ROPs and 640KB of L2 cache as the GTX 470, but because the ROPs operate on the core clock, they are running 125MHz (20.5%) faster than the GTX 470. Lastly, we have the ROPs, which share an existence with both the core and memory subsystems and split the difference.
The GTX 570 shares the same architectural improvements we originally saw in the GTX 580 because it is built on GF110. This means that GTX 570 has a better Z-culling system and can retire twice as many FP16 texels per clock as GTX 480. This aids in bridging the performance gap between the GTX 570 and GTX 480, which may exist due to the lower ROP count and smaller memory bus. The difference on the GTX 570 will be greater on shader-bound games and applications and smaller when we’re memory bandwidth constrained, unlike the GTX 580 and its balanced approach. Do keep in mind, however, that the overall improvements are asymmetrical when compared to the GTX 470: we’re looking at around a 30% theoretical improvement in shading/compute/texture performance, but only a 13.5% improvement in memory bandwidth.
The GTX 570, which succeeds the GTX 470, generally has similar power and noise characteristics. The fact that NVIDIA chose to maximize performance within their chosen power profile for the GTX 570 rather than boosting performance while also reducing power consumption to the GTX 580 is highlighted by the fact that NVIDIA pegs the TDP at 219W, just 4W more than the GTX 470. The GTX 570 has the same PCB, vapor chamber cooler, and shroud as the GTX 580, hence the card is otherwise similar to the GTX 580.
The MSRP for the card has been set by NVIDIA at $349. In previous weeks, this pricing had been left unoccupied because neither NVIDIA nor AMD had a product to place between the 480/470 and the 5970/5870, respectively. Coming from the top end of the market, this is essentially a nice price drop for performance similar to that of the GTX 480, but it also means that the Radeon 5870 and GTX 470 are the GTX 570’s value threats because they are nearly $100 cheaper and offer performance that is comparable to that of the GTX 480. The GTX 460 1GB SLI and the Radeon HD 6850 CF will currently be the GTX 570’s only rivals.
It should be a challenging launch today. We didn’t think NVIDIA could have that many GF110 devices available so quickly before the GTX 580 launch, but they were able to disprove our hypothesis there, so we’re happy to accept their word for it here. We would anticipate some cards to sell out based on their own forecasts and the cheaper cost of the GTX 570, but supply shouldn’t be an issue.
Finally, NVIDIA’s lineup will change with the release of the GTX 570. Since GF110 is a very good replacement for GF100, NVIDIA will want to phase out GF100 cards as soon as they can. The GTX 470 will continue to be available for quite some time (all signs point to NVIDIA still having a large number of GF100 chips available), but the GTX 480’s time is up.
Meet the GTX 570
The GTX 570 is a perfect redistribution of the GTX 580 design, as we briefly mentioned in our introduction. As with the GTX 580, NVIDIA employed the same PCB, cooler, power-throttling chips, and shroud. Our reference card even sports virtually the same livery. The GTX 580 has 6+8pin PCIe power sockets, whereas the GTX 570 uses 6+6pin PCI power sockets, which is the only externally visible hardware change between the two cards.
While having the same design as the GTX 580 undoubtedly made it possible for NVIDIA to release the GTX 570 swiftly and affordably, it also means that the GTX 570 now benefits from all of the design advancements that were made to the GTX 580. This indicates that a vapor chamber-based aluminum heatsink, supported by NVIDIA’s reinforced blower, provides cooling for the GTX 570. This also suggests that it is utilizing NVIDIA’s angled shroud, which is intended to provide greater cooling in confined locations for SLI users. It shouldn’t come as a surprise that the GTX 570 is similarly capable because this design has previously proven to be competitive with the GTX 470 on the GTX 580. Overall, the only significant drawback to this design is that because NVIDIA is employing the GTX 580 design, the GTX 570 is an inch longer than the GTX 470 due to the GTX 580’s 10.5″ length.
The situation with customized GTX 570s will be hazy, much like with the GTX 580. As part of its increased oversight of the GTX 500 series, NVIDIA will only approve designs that are on par with or better than the standard design. This isn’t necessarily a negative thing, but it limits the scope for unique designs, especially if someone wants to attempt cutting an inch off the card to make it the same length as the GTX 470. We learned last week that NVIDIA is opposing triple-slot coolers for the GTX 580 (and undoubtedly the GTX 570) because they can’t be used in SLI mode with most motherboards. As a result, any custom designs that do emerge will undoubtedly be more conservative than what we’ve seen with the GTX 400 series.
I/O is identical to the GTX 580 because NVIDIA is using the GTX 580 PCB. The standard NVIDIA arrangement of two DVI ports and one mini-HDMI connector is used here, with the card’s exhaust taking up the second slot. Additionally, this implies that the card can only drive two of the three ports at once, necessitating the use of SLI in order to utilize NVIDIA/3DVision Surround. The card’s mini-HDMI connector, however, supports HDMI 1.4a for 3D video applications, but audio bit streaming is not supported, restricting audio output to LPCM and DD+/DTS.
NVIDIA GTX 570 REVIEW TEST
We’ll only be examining single-GPU performance today due to a last-minute mishap with our second GTX 570. Once we obtain a replacement card, we’ll update our article to reflect the performance of the GTX 570 in SLI.
Launch drivers for the GTX 570 are 263.09, and they function exactly the same as the GTX 580 launch drivers from last month, which were 262.99. On the AMD side, we are using Catalyst 10.10e, which is a more recent revision than 10.11 despite the name.
|CPU:||Intel Core i7-920 @ 3.33GHz|
|Motherboard:||Asus Rampage II Extreme|
|Chipset Drivers:||Intel 126.96.36.1995 (Intel)|
|Hard Disk:||OCZ Summit (120GB)|
|Memory:||Patriot Viper DDR3-1333 3 x 2GB (7-7-7-20)|
|Video Cards:||AMD Radeon HD 6870|
AMD Radeon HD 6850
AMD Radeon HD 5970
AMD Radeon HD 5870
AMD Radeon HD 5850
AMD Radeon HD 5770
AMD Radeon HD 4870
NVIDIA GeForce GTX 580
NVIDIA GeForce GTX 570
NVIDIA GeForce GTX 480
NVIDIA GeForce GTX 470
NVIDIA GeForce GTX 460 1GB
NVIDIA GeForce GTX 460 768MB
NVIDIA GeForce GTS 450
NVIDIA GeForce GTX 285
NVIDIA GeForce GTX 260 Core 216
|Video Drivers:||NVIDIA ForceWare 262.99|
NVIDIA ForceWare 263.09
AMD Catalyst 10.10e
|OS:||Windows 7 Ultimate 64-bit|
As usual, Crysis: Warhead, one of our benchmark suite’s most challenging games, gets things started. Even now, two years after the first Crysis was released, the issue “but can it run Crysis?” is still significant, and the answer is still “no.” Even while we’re getting closer than ever, no single card can currently handle full Enthusiast settings at a playable framerate.
Crysis ultimately serves as the backdrop for the remainder of this study. The GTX 570 is essentially tied with the GTX 480 at 2560 and 1680, and only at 1920 does it lag behind by a paltry 4%. While it is 20% faster (and 40% more expensive) than the GTX 470, it is over 10% slower than the GTX 460 1GB SLI. Overall, the GTX 570 is almost on pace with the GTX 480 and should be able to play almost any game at 1920p.
When it comes to AMD’s cards, the 5870, which was never far behind the GTX 480, is nipping at the heels of the GTX 570. The GTX 570 is occasionally only 5% faster and rarely better than 15% faster, demonstrating why the 5870 poses a value threat to the GTX 570. At 1920, the 6850 CF is far and away the best option here, and it costs only $20 to $30 more than the GTX 570. A pair of inferior AMD cards will provide superior gaming performance in exchange for the limitations of a multi-GPU arrangement, as was the case with the GTX 580.
When compared to our minimum framerates, the narrative is largely the same. Except for 1920, where the GTX 570 lags behind the GTX 480 a little more than we’d expect, the GTX 570 and GTX 480 are close to one another except for 2560, where the extra memory gives the NVIDIA cards a clear advantage. The 460SLI/6850CF setups are still well ahead even though the 5870 isn’t nearly as dangerous in this situation as it is with typical framerates.
The next game is Metro 2033 by 4A Games, a tunnel shooter that was launched earlier this year. The game finally received a significant patch in September that fixed some persistent image quality issues, allowing us to include it in our benchmarking programmed. We were able to accurately benchmark far more demanding conditions than we could with FRAPS at the same time a dedicated benchmark mode was created to the game. The underground equivalent of Metro would be Crysis, the tropical GPU killer.
While Metro was a fantastic game for the GTX 580 to demonstrate its performance advantage, the GTX 570’s scenario is very different. Here, it performs its function as a GTX 480 replacement once more, but it is much more frail when compared to other cards. The 5870 and GTX 570 are more competitive the higher the resolution, with the GTX 570 only holding a 10% advantage at 1920. The SLI/CF cards, however, are in a comparable position and only outperform the GTX 570 by about 10%.
The GTX 570 is once more 25% faster at 1920 than the GTX 470, which is respectable for this shader-intensive game.
Battlefield: Bad Company 2
One of the key DX11 titles in our benchmark suite continues to be the most recent Battlefield game, Bad Company 2. Since BC2 lacks an integrated benchmark or recording option, we ran a FRAPS run of the first act’s jeep pursuit, which is an on-rails section of the game and offers spectacular explosions, trees, and other visual effects.
Another game where the GTX 570 significantly outperforms the GTX 480 is Bad Company 2. It is obviously exceeding the GTX 570’s theoretical clock speed advantages and moving toward architectural optimizations. Unfortunately, the GTX 570’s advantage only becomes noticeable at lower resolutions, so by the time we reach 2560, the gap has vanished. We are once again on par with the GTX 480 because the advantage is hardly noticeable at 1920.
That advantage is that the Radeon 5870 gets uncomfortable close at higher resolutions, which may have been better for NVIDIA. The GTX 570 advantage is minimal at 2560 and barely 10% at 1920. NVIDIA is in a difficult situation here due to AMD’s powerful CF scaling, since the 6850 CF is a staggering 32% quicker than the GTX 570 in Bad Company 2.
But with our Waterfall benchmark, which acts as a stand-in for minimum framerates, NVIDIA manages to flip the tables. The GTX 570 is still tied with the GTX 480 here, but it is also now at parity with the 6850CF and over 50% faster than the 5870, showing unmistakably that the 5870 and GTX 570 aren’t quite as close as they were at first glance if you’re worried more about minimums than averages in Bad Company 2. Here, AMD would likely gain greatly from having more RAM.
Mass Effect 2
Our Unreal Engine 3 game is a space-faring RPG from Electronic Arts. Although it has a built-in benchmark, we can enable anti-aliasing through the driver control panels, which helps us understand how well UE3 performs at higher quality settings. We utilize FRAPS to capture a brief run because we are unable to use a recording or benchmark in ME2.
Mass Effect 2 is the only other game where the GTX 570 has any kind of notable disadvantage compared to the GTX 480; albeit slight, the GTX 570 behind the GTX 480 by about 4% here, which is a bit of a curveball from our previous titles. This is another of the GTX 570’s less significant benefits over the GTX 470, with a gain of only 17%. It appears that the game in our suite that will most likely experience a memory bandwidth constraint is based on what we’re seeing with the GTX 500 series.
The GTX 570 doesn’t stand a chance of outpacing its AMD rivals because of that memory limitation. The performance advantage over the 5870 at 1920 is only 13%, while the SLI/CF cards perform admirably by outperforming it by more than 35%.
The 2009 off-road racing game from Code masters continues to hold the title of token racer in our benchmark collection. DIRT 2, the first DX11 racing game, utilizes the DX11’s tessellation capabilities very extensively and continues to be the best-looking racing game we have ever seen.
Another game in which the GTX 570 has an edge over the GTX 480 beyond just core clock differences is DIRT 2. The performance advantage increases to 4-6% whenever we go up to less CPU-limited resolutions, giving the GTX 570 a slight but observable lead. Additionally, because of this advantage over the GTX 470, the difference is now greater than 35%.
In terms of the opposition, the Radeon 5870 can’t even come close to closing the gap in this game; in fact, it performs somewhat worse than the GTX 470. The SLI/CF cards perform better in this situation, especially the NVIDIA cards due to their improved scaling in this game.
Compute & Normalized Numbers
Following our examination of gaming performance, we turn to our normal examination of compute performance, which is combined with an examination of theoretical tessellation performance. Everything here should be determined by the core clock and SMs, with the GTX 570’s modest core clock advantage over the GTX 480 defining the majority of these tests, in contrast to our gaming benchmarks where NVIDIA’s architectural upgrades could have an impact.
In order to decompress textures instantly, Civilization V employs Direct Compute as our initial compute benchmark. The texture decompression method used by Civ V is put to the test by continually decompressing the graphics needed for one of the game’s leader scenes.
The GPU ray tracing branch of the open source Lux Render renderer, Small Lux GPU, serves as our second GPU compute test. Despite being in beta, Small Lux GPU recently achieved a breakthrough by integrating a full ray tracing engine in OpenCL, enabling them to completely offload the procedure to the GPU. The ray tracing engine is what we’re evaluating.
Small Lux GPU, the GPU ray tracing division of the open source Lux Render renderer, serves as our second GPU compute test. Small Lux GPU recently achieved a milestone by building a full ray tracing engine in OpenCL, enabling them to completely offload the procedure to the GPU, even though it is still in beta. We are trying this ray tracing engine.
The prerequisites for Small Lux GPU are rather simple: computing, and plenty of it. The GTX 570’s higher core frequency than the GTX 480’s results in a relatively simple performance gain of 4%, which is broadly in line with the theoretical maximum. Small Lux GPU does not appear to be affected by the decreased memory bandwidth and L2 cache. The GTX 570 is still 27% faster than the GTX 470, however the advantage doesn’t quite reach its theoretical maximum.
The GTX 570, thanks to AMD’s compute-intensive VLIW5 design, which Small Lux GPU excels with, is only slightly slower than the 6850 and 6870 in performance, as was the case with the GTX 580. As a result, the Radeon 5870 is the best card, and the GTX 570 performs 27% worse than it should compared to it.
A Folding @ Home benchmark serves as our last computation test. Cards like the GTX 580 can be especially intriguing for distributed computing fans, who are typically seeking for the fastest card in the nicest package, given NVIDIA’s concentration on compute for Fermi and in particular GF110 and GF100.
Once more, the GTX 570’s performance advantage is equal to its core clock advantage. The GTX 570 would probably hit the sweet spot for price, performance, and power/noise if it weren’t so easy to scale a DC project like F@H to multi-GPU systems.
The normalized data view that we initially saw with our examination of the GTX 580 is back for us to examine the performance of the GTX 570. The GTX 570 has the same number of SMs as the GTX 580, but it has fewer ROPs and a smaller memory bus. The GTX 580 had identical memory/ROP capabilities to the GTX 480, but it had more SMs. As a result, whereas a normalized dataset for the GTX 580 demonstrates the benefit of the architectural improvements made to the GF110 and the higher SM count, the normalized dataset for the GTX 570 demonstrates the architectural improvements as well as the impact of reduced memory bandwidth, ROPs, and L2 cache.
Certainly an intriguing picture is painted by the results. The loss of memory bandwidth, L2 cache, and ROPs eventually has an impact on almost everything; if the GTX 570 didn’t typically have an edge due to its higher core clock, it would typically fall short of the GTX 480 by minor margins. STALKER, Mass Effect 2, and BattleForge stand out among the titles that are most memory-intensive in this list.
On the other hand, despite the fact that our normalized GTX 570 is inferior to a GTX 480 in all respects other than architectural advantages, DIRT 2 and HAWX both display an improvement of 4%. These are undoubtedly some of the games that NVIDIA had in mind when they modified GF110.
Power, Temperature, & Noise
As always, we will look at the GTX 570’s power usage, temperature range, and sound level. While NVIDIA’s improvements to these characteristics were primarily aimed at bringing the GTX 480’s direct replacement to more realistic levels, the GTX 570 will undoubtedly benefit as well.
With VIDs, we can only draw a limited amount of data because we only have 1 card. Although the VID of our GTX 570 sample is lower than that of our GTX 580, NVIDIA uses a range of VIDs for each card, as is common knowledge. Since NVIDIA is likely using ASICs for the GTX 570 that have damaged ROPs/SMs and/or ASICs that wouldn’t pass muster at a suitable voltage with all functional units turned on, we would anticipate that the average GTX 570 VID will be higher than the average GTX 580 VID based on what we have seen with the GTX 470. This would prevent us from constantly realizing the full power benefit of lower clock speeds and fewer functional units.
|GeForce GTX 500 Series Voltages|
|Ref 580 Load||Asus 580 Load||Ref 570 Load||Ref 570 Idle|
The GTX 570 is the first GF1x0 card to run our test system at less than 170W at idle thanks to fewer functional units than the GTX 580 and a less leaky manufacturing process than the GTX 400 series. However, despite their greater size, NVIDIA’s larger chips still perform worse than AMD’s smaller processors here, which is highlighted by the fact that the 6850 CF uses just one additional watt when idle. However, this is a 4W upgrade from the GTX 470 and a 21W upgrade from the GTX 480; the former in particular highlights the process advances since there are more functional units yet power consumption has decreased.
Under load, we once more witness the effectiveness of TSMC’s process advancements and NVIDIA’s architecture. Power consumption has decreased by 5W even though the GTX 570 is over 20% quicker than the GTX 470 in Crysis. Even more impressive is the comparison to the GTX 480, which uses 60W less power for the same level of performance—a figure far greater than the power savings from deleting two GDDR5 memory chips. It’s amazing how a lot of seemingly insignificant adjustments may add up to so much.
On the other hand, we now have access to Fur Mark once more because W1zzard of Tech Power Up was able to successfully disable NVIDIA’s power throttling measures on the GTX 500 series. After benchmarking the GTX 570, GTX 580, and GTX 580 SLI, we now have numerical results that are completely comparable.
It’s rather intriguing that, while performing so well in Crysis, the GTX 570 performs worse in this test than the GTX 470. The GTX 470 benefits from the official TDP difference, but not by this much. I value the Crysis scores more than the Fur Mark scores, but it’s important to remember that the GTX 570 might occasionally perform lower than the GTX 470. The GTX 570 is still outperforming the GTX 480 by 33W, despite the fact that we’ve already shown that their performance is comparable.
Naturally, AMD’s GPUs shine out because they benefit from smaller, more energy-efficient chip designs. In every circumstance, the Radeon HD 5870 consumes 50W or less less power than the GTX 570, and even the 6850 CF fluctuates between being slightly worse and somewhat better than the GTX 570. The GTX 570’s power consumption is comparable to earlier cards thanks to advances in the architecture and manufacturing process, but NVIDIA is still unable to surpass AMD in terms of power efficiency.
The idling temperatures come next. It should come as no surprise that the GTX 570 performs exactly like a GTX 580 here: 37C, given that the GPUs are almost comparable and the platform is the same. The GTX 570 surpasses the 5870 while crushing the SLI and CF cards to join the ranks of some of our coolest idle high-end GPUs.
With the same cooling system as the GTX 580 and less power usage, we first anticipated the GTX 570 to defeat the GTX 580 in every scenario, but it turns out things are more complicated than that. The 82C GTX 570 is 3C warmer than the GTX 580 in Crysist, putting it on par with other cards like the GTX 285 (82C). However, more significantly, it runs 11–12°C cooler than the GTX 480/470, emphasizing the significance of the card’s lower power usage and the vapor chamber cooler. However, since the 5870 and SLI/CF cards are between 0C and 5C cooler, it is not quite as competitive with them.
Fur Mark’s temperatures are in the mid- to upper 80s, which puts it in good company with the majority of other top-tier cards. The GTX 570 behind the 5870 by 2C, with everything else being on par with or even hotter. Particularly, the GTX 480/470 end up being 4C warmer and, as we’ll see, a lot louder.
When we examine our noise data, it becomes evident that NVIDIA has been modifying the fan specifications of their design for the GTX 570 because in this area, the GTX 570 falls short of the GTX 580. For instance, the GTX 570 emits 1.5 dB less noise while idling than the GTX 580, 480, or 470, just about reaching the noise floor of our system. It operates as a silent card in our test setup when not in use for our tests.
Finally, under load, it becomes clear that NVIDIA modified the GTX 570 to be quieter rather than cooler. The GTX 570 is 2dB quieter than the GTX 580, 3.6dB quieter than the GTX 470, and 6.2dB quieter than the similarly performing GTX 480 at the slightly higher temps we previously observed. The GTX 570 practically tops the charts for its performance level here; the next quietest cards are the 5850 and 6870, which have a respectable heritage to be compared to. The 5870 is 1.4 dB louder while the SLI/CF cards are between 1.2 and 2.4 dB louder on the other side of the graph.
Our concluding thoughts would be nearly identical if we copied the conclusion from our GeForce GTX 580 post and changed 580, 480, and 6870CF with 570, 470, and 6850CF, respectively. But the GTX 580 and GTX 570 are also remarkably similar.
The GeForce GTX 570 adopts a far more focused strategy than the GTX 580 did in order to raise the bar for GPU performance while simultaneously lowering power consumption. It offers gaming performance that is essentially identical as the GTX 480 at a lower price, with less power consumption, lower temperatures, and less noise. In essence, it is the new GTX 480. The GTX 570, a lower-tier GF110 card, won’t blow everyone away with its performance, but like the GTX 580, it represents a significant improvement. In this instance, it’s a good step toward offering today’s performance on the market for less money and with power/thermal/noise characteristics that are more compatible with a variety of systems. At the very least, NVIDIA has successfully transferred the GTX 580’s great performance and noise balance to a more affordable, lower performing tier.
Additionally, for single-GPU cards, NVIDIA is currently the only player in town at a price of $350. The GTX 470 did a commendable job of bridging the gap between the Radeon HD 5870 and 5850 earlier this year, and NVIDIA has done a commendable job of doing the same thing between the GTX 580 and GTX 470 until an AMD rival appears. The GeForce GTX 460 1GB SLI and the Radeon HD 6850 CF are the only alternatives to single cards now available on the market. It is important not to undervalue the Radeon in particular because it can outperform the GTX 570 nearly at will. However, the current high 6850 pricing puts it at a $30+ premium over the GTX 570. The normal drawbacks of uneven performance scaling, increased noise, and a reliance on driver updates to enable the second GPU in new games are present in both multi-GPU implementations, of course. Similar to the GTX 580, we would choose the ease of a single-GPU arrangement over the potential performance benefits of a multi-GPU setup, but this is a matter of personal preference.
The GTX 570 is largely what we anticipated when we first saw the GTX 580 as a gap-filler, and we have no significant issues with it. We are disappointed by the fact that NVIDIA is being cautious with the pricing; $350 is not a high asking price. However, at $100 more than the GTX 470 and Radeon HD 5870, you’re paying a lot for that extra 20–25% in performance. The GTX 570 is fast enough to justify its position and the high-end card price premium. We’ll be ecstatic if AMD’s upcoming card lineup can exert some pressure on NVIDIA in this situation.
This leads us to AMD in its ultimate stage. The Radeon HD 6900 series is scheduled to debut by the end of the year, but time is swiftly running short. It’s still too early to recommend delaying any GTX 500 series purchases (especially if you want a card for Christmas), but if you’re not in a rush, it might be worthwhile to wait a few more weeks to see what AMD has in store. We would most certainly want to see AMD and NVIDIA engage in a holiday price war and see price reductions as a result.