Two weeks ago we saw the paper launch of the Radeon HD 6450, the low-end member of AMD’s Northern Islands family of GPUs. It was a solid product for HTPC use and a very notable improvement over the 5450 it replaced, but it was an uncharacteristically delayed launch for AMD. At the same time we noted that the Northern Islands family had one more GPU we had not seen: Turks.

As it turns out, Turks-based video cards will be launching alongside the 6450 today, delivering all of the remaining Northern Islands products in a single push. Turks will be powering the Radeon HD 6670 and Radeon HD 6570, replacing the Redwood-based Radeon HD 5670 and Radeon HD 5570 respectively. Considering that we saw AMD deliver a solid update for their low-end lineup with the 6450, will we see the same with Turks and the 6670/6570? Let’s find out.

	AMD Radeon HD 6670	AMD Radeon HD 5670	AMD Radeon HD 6570 (GDDR5)	AMD Radeon HD 5570
Stream Processors	480	400	480	400
Texture Units	24	20	24	20
ROPs	8	8	8	8
Core Clock	800MHz	775MHz	650MHz	650Mhz
Memory Clock	1GHz (4GHz data rate) GDDR5	1GHz (4GHz data rate) GDDR5	1GHz (4GHz data rate) GDDR5	900MHz (1800MHz data rate) DDR3
Memory Bus Width	128-bit	128-bit	128-bit	128-bit
VRAM	1GB	1GB / 512MB	512MB	1GB
FP64	N/A	N/A	N/A	N/A
Transistor Count	716M	627M	716M	627M
TDP	66W	61W	60W	42.7W
Manufacturing Process	TSMC 40nm	TSMC 40nm	TSMC 40nm	TSMC 40nm
Price Point	$99	$65-$85	$79	$50-$70

Turks is the 4^th and final member of the Northern Islands family, or to be more practical the 3^rd and final member of the revised Evergreen family. Based on Redwood before it, Turks implements the Northern Islands series’ architectural improvements while also implementing a larger number of SIMDs in order to give AMD a mid-cycle performance boost on the same TSMC 40nm process.  With the 6450 dragging up the low-end of the market and competitive pressure from NVIDIA, AMD needed a product above the 5670 but still smaller/cheaper than the 5750, and that is what Turks will provide.

Architecturally, Turks is very close to Redwood. Compared to Redwood it has 6 SIMDs instead of 5 SIMDs, giving it 480 SPs and 24 texture units versus 400 SPs and 20 texture units on Redwood. The ROP count remains unchanged at 8 ROPs, while the memory bus is still 128 bits wide. Of course being a Northern Islands GPU, Turks implements all of the common improvements we see with NI: UVD3, improved texture filtering, HDMI 1.4a (e.g. Blu-Ray 3D), a revised tessellation unit, and DisplayPort 1.2 support.

These improvements have of course added some bulk to the die; Redwood was comprised of 627 million transistors while Turks is 716M. This in turn has increased the die size slightly to 118mm2, coming from 104mm2 on Redwood. Power consumption has also gone up accordingly, but less so than the transistor count or die size did. For the 6670 the TDP is 66W, versus 60W for the 5670. The 6570 GDDR5 meanwhile is rated for 60W versus 43W for the 5570; this is largely a consequence of switching to GDDR5. Meanwhile idle power consumption is 10W for the 6570 DDR3, 11W for the 6570 GDDR5, and 12W for the 6670.

As far as performance is concerned, the 6670 is close to the 5670 on paper. At 800MHz the 6670 has a 25MHz (3%) core clock advantage, while the 1GHz (4GHz data rate) memory clock is identical to the 5670. The big advantages for the 6670 are any gains in architectural efficiency, combined with the additional SIMD; the SIMD alone gives the 6670 a 20% shading and texturing improvement over the 5670. With the same 8 ROPs between the two cards, the 6670 should excel in shading and texturing bound situations, while ROP or memory bandwidth bound situations will be much closer.

Top: Radeon HD 6670. Bottom: Radeon HD 6570

For the 6570 on the other hand, the performance advantage is quite a bit larger. The original 5570 – and the one we continue to test today – launched with DDR3 memory. AMD did introduce a GDDR5 variant in the summer of 2010, but it never gained much traction. So compared to the 5570, the GDDR5 6570 with its 1GHz (4GHz data rate) GDDR5 has 222% the memory bandwidth of the DDR3 5570 and its 900MHz (1.8GHz data rate) RAM. Now there’s a catch in all of this: similar to how the 5570 ended up, AMD will be launching cards with both GDDR5 and DDR3. The GDDR5 cards like the one we’re reviewing today will come with 512MB of RAM, while the DDR3 cards will come with 1GB of RAM. The extra RAM has its advantages in some edge cases, but our advice always has been and remains to be that you should pick the GDDR5 versions of most video cards over the (G)DDR3 versions. The only notable downside to the GDDR5 card in this case is that GDDR5’s power consumption is much higher, which is why the GDDR5 6570 is a 60W card while the DDR3 6570 is 44W.

RAM aside, the 6570 has the same advantages over the 5570 as the 6670 has over the 6570. With identical core clocks of 650MHz, the performance difference comes down to the RAM, followed by the 20% increase in the SIMD count. The GDDR5 6570 should outperform the DDR3 5570 by quite a bit at all times, but it’s going to absolutely shine in memory bandwidth or shader-bound situations, while anything that’s ROP-bound will be closer.

AMD will be launching the 6670 at $99 and the 6570 at $79, putting both cards right in the middle of the highly-competitive sub-$100 market. Competing cards include the Radeon HD 5670, the frequently discounted to $99 Radeon HD 5770, the similarly discounted GeForce GTS 450, and the GeForce GT 440 – a higher clocked GF108 (GT 430) part with GDDR5 that NVIDIA quietly released back in February. As was the case with the 6450 and most other sub-$100 launches, without a die shrink new parts cannot compete with discounted parts from a higher tier, so it shouldn’t come as a surprise that the 6670 in particular is easily beaten on performance at $99.

Thankfully unlike the 6450, the 6670 and 6570 are hard launching today. They will be appearing at e-tailers later today alongside the 6450.

April 2011 Video Card MSRPs
NVIDIA	Price	AMD
GeForce GTX 590	$700	Radeon HD 6990
GeForce GTX 580	$480
GeForce GTX 570	$320	Radeon HD 6970
	$260	Radeon HD 6950 2GB
GeForce GTX 560 Ti	$240	Radeon HD 6950 1GB
	$200	Radeon HD 6870
GeForce GTX 460 1GB	$160	Radeon HD 6850
GeForce GTX 460 768MB	$150	Radeon HD 6790
GeForce GTX 550 Ti	$130
	$110	Radeon HD 5770
	$99	Radeon HD 6670
GeForce GTS 450	$95-$110
	$79	Radeon HD 6570
GeForce GT 430	$50-$70	Radeon HD 5570
GeForce GT 220	$55	Radeon HD 6450
GeForce G 210	$30-$50	Radeon HD 5450

 

Meet The Radeon HD 6670 & Radeon HD 6570

With the Redwood cards (5670 & 5570), we saw two distinct reference designs even though both used the same GPU. The 5670 was a full-profile card while the 5570 was a low-profile card intended to fill a higher performance low-profile niche. With the Turks cards however, this kind of stratification is gone.

Look closely and you'll see the ATI logo. AMD is using spare 5570 coolers for the 6570 reference design, which means these coolers predate the AMD rebranding. Retail cards won't be like this, however.

Instead both reference cards are low-profile cards. The PCB for both cards is quite similar, but not identical. In practice with only a 6W difference between the two cards, we can’t imagine too many partners will roll different PCBs for the cards. The difference between the two cards really comes down to the cooler used – the 6570 uses the same AMD reference cooler we’ve seen on the 5570 and 6450 (it is in fact a spare 5570 cooler), while the 6670 uses a new low-profile double-wide cooler. This is how AMD managed to get the 6670 out as a low-profile card as opposed to a full-profile card like the 5670: it’s shorter, but it’s wider.

Our 6670 is equipped with 4 2Gb 5GHz Hynix GDDR5 memory modules. This marks the first time we’ve seen 2Gb GDDR5 on anything other than a high-end card, which is a good sign that we should expect to see more cards using 2Gb GDDR5 in the near future. Both the 6670 and 6570 measure 6.61” long, the same length for that matter as the 6450.

Display connectivity is the same as with other AMD low-profile cards. On the PCB are a single DVI-DL port and a DP 1.2 port, while on the bracket is a VGA port attached with a ribbon cable. With partner cards we’d expect a number of cards to drop the DP1.2 port for an HDMI 1.4 port on both cards. But on that note, one thing we’re still trying to get confirmation on from AMD is whether Turks can drive more than one DVI-type display at once. We have no specific reason to believe this, but Barts did drop a DVI-DL port for a DVI-SL port, so if anyone is going to drop down to supporting a single DVI-type display at once, it’s AMD. In any case through the magic of DP1.2 MST hubs, the 6670 can drive four displays. Meanwhile the 6570 can drive three displays.

As was the case with the 6450, AMD’s reference design is only for engineering and sampling purposes – it’s not going to be used in retail cards. We don’t have any pictures of retail cards as of press time, but we’d expect 6670 and 6570 designs to resemble 5570 designs with a larger cooler more suitable for Turks. As such our temperature/noise results are a decent approximation of the cards that will be on the market, but they won’t match any specific retail cards.

The Test

For the AMD DX11 lineup including the 6570 and 6670, we’re using the Catalyst 11.4 preview driver. For NVIDIA’s lineup we’re using the release 270 driver.

Sub-$100 cards have rarely been able to play our games at 1680 at our usual settings, so we’re once again breaking everything down into both 1280 at lower settings as our primary point of reference, and 1680 for comparison with our faster cards. The 6670 generally is too slow to play at 1680, but there will be a few exceptions.

CPU:	Intel Core i7-920 @ 3.33GHz
Motherboard:	Asus Rampage II Extreme (X58)
Chipset Drivers:	Intel 9.1.1.1015 (Intel)
Hard Disk:	OCZ Summit (120GB)
Memory:	Patriot Viper DDR3-1333 three x 2GB (7-7-7-20)
Video Cards:	AMD Radeon HD 6970 AMD Radeon HD 6950 2GB AMD Radeon HD 6870 AMD Radeon HD 6850 AMD Radeon HD 6790 AMD Radeon HD 6670 AMD Radeon HD 6570 AMD Radeon HD 6450 (GDDR5) AMD Radeon HD 5970 AMD Radeon HD 5870 AMD Radeon HD 5850 AMD Radeon HD 5830 AMD Radeon HD 5770 AMD Radeon HD 5670 AMD Radeon HD 5570 (DDR3) AMD Radeon HD 5450 (DDR3) AMD Radeon HD 4870X2 AMD Radeon HD 4870 NVIDIA GeForce GTX 580 NVIDIA GeForce GTX 570 NVIDIA GeForce GTX 560 Ti NVIDIA GeForce GTX 550 Ti NVIDIA GeForce GTX 480 NVIDIA GeForce GTX 470 NVIDIA GeForce GTX 460 1GB NVIDIA GeForce GTS 450 NVIDIA GeForce GT 430 NVIDIA GeForce GTX 295 NVIDIA GeForce GTX 285 NVIDIA GeForce GTX 260 Core 216 NVIDIA GeForce GT 240 (GDDR5) NVIDIA GeForce GT 220 (DDR3)
Video Drivers:	NVIDIA ForceWare 262.99 NVIDIA ForceWare 266.58 NVIDIA ForceWare 270.51 Beta AMD Catalyst 10.10e AMD Catalyst 11.1a Hotfix AMD Catalyst 11.4 Preview
OS:	Windows 7 Ultimate 64-bit

 

Crysis: Warhead

Kicking things off as always is Crysis: Warhead, still one of the toughest games in our benchmark suite. Even three years since the release of the original Crysis, “but can it run Crysis?” is still an important question, and for three years the answer was “no.” Dual-GPU halo cards can now play it at Enthusiast settings at high resolutions, but for low-end cards even Mainstream/Medium quality is about the best that can be achieved.

It takes a 5770 to crack 30fps in Crysis at 1680x1050, so the fact that the 6670 is only a bit over 20fps isn’t a big surprise. As far as performance at 1680 is concerned Crysis is a good summary of what we’ll see: too slow for 1680, and well behind the next tier of cards represented by the GTS 450 and the Radeon HD 5770/5750. However the 6670 does have a notable advantage here: with 1GB of VRAM it handles 1680 much better than the 512MB 5670, even if it’s ultimately too slow to be playable at these specific settings.

So for the kind of high settings we like to test at, it’s 1280 that’s going to be playable on the 6670, 6570, and similar cards. Once we drop down to 1280 and Mainstream quality the 6670 can crack 60fps, and with a bit of fiddling it should be possible to increase the quality of a setting or two without significantly impacting performance. Overall the 6670 has a 10% lead over the 5670 it replaces, which is a reasonable outcome for a game that’s largely (but not completely) shader-bound.

As for the 6570, it’s well-entrenched near the top of the pack. At 95% of the performance of the 5670 it does quite well, and the next-fastest modern card is the GT 240 which is quite some distance off. Against the 5570 it has a 35% lead, thanks largely to the use of GDDR5 instead of DDR3.

We have also thrown in the 8800GT into our 1280 results, just to offer a different look at performance relative to NVIDIA’s retired champion. It’s a good reference point for where we’re at now, versus what a $300 card did 3 years ago; and once we get to power/temperature/noise testing how better process technology and smaller GPUs have improved those metrics.

Looking at the minimum framerates, the relative ranking of our cards remains unchanged. The 6670 cements its lead over the 5670 here at 1680 thanks to its 1GB of VRAM, while at 1280 the gap narrows significantly to 5%. Meanwhile the 6570 loses ground on the 5670, with the 5670 taking a 10% lead.

BattleForge

Up next is BattleForge, Electronic Arts’ free to play online RTS. As far as RTSes go this game can be quite demanding, and this is without the game’s DX11 features.

As with Crysis, neither Turks card can reach 30fps at our usual 1680 settings and is more or less crushed by the 450/5770, so we’ll have to take a look at 1280. As BattleForge is a particularly shader-bound game, we’re seeing the best case scenario for the 6670 and 6570. The 6670 has a 14% lead on the 5670 and a 15% lead on the 6570. Meanwhile the 6570 is within 1% of the 5670, and 25% ahead of the 5570. If BattleForge didn’t show a tendency to also be ROP-bound at times, I’d expect to see the Turks cards take an even greater lead.

Metro 2033

The next game on our list is 4A Games’  tunnel shooter, Metro 2033. Metro is quite a resource intensive game, and if Crysis is a tropical GPU killer, then Metro would be its underground counterpart.

Being a GPU killer, the 6670 and 6570 don’t even stand a chance at 1680, so it’s to 1280 we go. Normally we’d classify Metro as being shader-bound, which is why we’re surprised the 6670 only takes a 6% lead over the 5670 and an 11% lead over the 6570. At these settings we’re looking at either being ROP or memory bandwidth bound, with the former being the more likely outcome given that our results closely match the differences in clockspeed. With that said the 5570 is certainly vulnerable to both ROP and memory bandwidth limitations, which is why the 6570 takes a sizable 33% lead.

HAWX

Ubisoft’s 2008 aerial action game is one of the less demanding games in our benchmark suite, particularly for the latest generation of cards. However it’s fairly unique in that it’s one of the few flying games of any kind that comes with a proper benchmark.

As one of our less demanding games, HAWX is the first title in our test suite that’s playable at 1680 with the 6670. At 46fps you may want to turn off anti-aliasing to ensure smooth performance, but it should be doable. In this case the 6670 has a 9% lead over the 5670, while the 5670 in turn still has 5% on the 6570. However all of these cards are still well behind the likes of the GTS 450 and Radeon HD 5770.

Meanwhile at 1280 the story is much the same. Even without sacrificing anything beyond resolution and AA a number of cards crack 60fps, including both Turks cards.

Civilization V

Civilization V is the latest incarnation in Firaxis Games’ series of turn-based strategy games. Civ5 gives us an interesting look at things that not even RTSes can match, with a much weaker focus on shading in the game world, and a much greater focus on creating the geometry needed to bring such a world to life. In doing so it uses a slew of DirectX 11 technologies, including tessellation for said geometry and compute shaders for on-the-fly texture decompression.

At 1680 the 6670 can almost – but not quite – hit 30fps. As a turn based game this should be playable, but it definitely won’t be smooth if you’re scrolling around. It is interesting to note though that the 5750 doesn’t have its usual advantage here: it’s only ahead by 6%. Compared to AMD’s other cards this is a good showing for the 6670, but thanks to NVIDIA’s implementation of command lists, they have a significant advantage in this game.

Overall the 6670 enjoys a 12% lead over the 5670 at 1680, and 11% at 1280; and at the same time we see a 20% lead over the 6570, one of the biggest gaps between the two cards that we’ll ever see. Meanwhile the 6570 can’t quite keep up with the 5670, falling behind by 8% at 1280. Indeed even with its massive memory bandwidth advantage, it’s only 11% ahead of the 5570 at this resolution. Ultimately Civ V is bottlenecked by more than just shaders and memory bandwidth, and as a result it’s difficult for the 6570 to break 30fps even at 1280 with medium quality settings.

Battlefield: Bad Company 2

Now approaching a year old, Bad Company 2 remains as one of the cornerstone DX11 games in our benchmark suite. Based on the Frostbite 1.5 engine, it will be replaced in complexity by the DX10+ only Frostbite 2 engine (and Battlefield 3) later this year.  As BC2 doesn’t have a built-in benchmark or recording mode, here we take a FRAPS run of the jeep chase in the first act, which as an on-rails portion of the game provides very consistent results and a spectacle of explosions, trees, and more.

Bad Company 2 is another game where the 6670 delivers technically playable performance at 1680. Given that it’s a first person shooter it might be reasonable to drop down in resolution and/or quality to keep the entire experience smooth, but it’s playable. Indeed of the cards we test at 1680, it’s the slowest card we’d call playable.

In any case, depending on the resolution the 6670 has a 12-14% lead over the 5670, while the 6570 trails the 5670 by as little as 2%. Everything here is well ahead of the GT 430 and other slower cards, but well behind the GTS 450 and faster cards. Turks fits in very well between GF108 and GF106, it would seem.

STALKER: Call of Pripyat

The third game in the STALKER series continues to build on GSC Game World’s X-Ray Engine by adding DX11 support, tessellation, and more. This also makes it another one of the highly demanding games in our benchmark suite.

As with Crysis and Metro, while the 6670 does very well compared to the 5670 at 1680 due to the extra VRAM, ultimately there’s no hope for a playable framerate at 1680 at high settings for the 6670. Once we go down to 1280 with medium quality settings the performance of the 6670 shoots way up, hitting 46fps. Here it still has a 13% advantage over the 5670 and a 20% advantage over the 6570, while the 6570 trails the 5670 by 6%.

DIRT 2

Codemasters’ 2009 off-road racing game continues its reign as the token racer in our benchmark suite. As the first DX11 racer, DiRT 2 makes pretty thorough use of the DX11’s tessellation abilities, not to mention still being the best looking racer we have ever seen.

DIRT 2 is another game where the 6670 could be playable at 1680 if only it was a bit faster; it can crack 30fps, but it’s certainly not a smooth experience even without AA. The 5750 is 22% ahead here, and it only gets worse from there.

Looking at 1280 however and the performance is much, much higher. Here the 6670 has an 11% lead on the 5670 and 13% on the 6570. Meanwhile the 6570 once again closely trails the 5670, this time by 2%. DIRT 2 has a number of different quality settings, so it should be possible to run it at a higher quality without significantly impacting the performance of the Turks cards.

Mass Effect 2

Electronic Arts’ space-faring RPG is our Unreal Engine 3 game. While it doesn’t have a built in benchmark, it does let us force anti-aliasing through driver control panels, giving us a better idea of UE3’s performance at higher quality settings. Since we can’t use a recording/benchmark in ME2, we use FRAPS to record a short run.

The 6670 once again cracks 30fps at 1680, this time on Mass Effect 2. Even the 6570 does well here, hitting 33fps and tying with the 5670. ME2’s framerate can fluctuate wildly depending on the level, but if you were to disable anti-aliasing it should be playable at 1680 most of the time.

Overall the 6670 has a lead between 9% and 12% on the 5670 depending on the resolution, while the 6570 is effectively tied with the 5670 at all resolutions. Memory bandwidth plays a huge part of this, which is why the 6570 has a shocking 50% lead over the 5570 with ME2.

Wolfenstein

Finally among our benchmark suite we have Wolfenstein, the most recent game to be released using the id Software Tech 4 engine. All things considered it’s not a very graphically intensive game, but at this point it’s the most recent OpenGL title available. It’s more than likely the entire OpenGL landscape will be thrown upside-down once id releases Rage later this year.

As our final game Wolfenstein is another game that’s pretty easy to run, leading to both Turks cards easily exceeding 30fps at 1680, and 60fps for the 6670 at 1280x1024. Regardless of the resolution the relative performance of each card is rather consistent, with the 6670 enjoying an 11% lead over the 5670 and 17% lead over the 6570, while the 6570 is within 5% of the 5670. Once again showcasing how these cards are outclassed by higher-tier cards on sale, the 5750 is the next-slowest modern card at 1680, and it’s ahead by 30%.

Compute & Tessellation Performance

Our first compute benchmark comes from Civilization V, which uses DirectCompute to decompress textures on the fly. Civ V includes a sub-benchmark that exclusively tests the speed of their texture decompression algorithm by repeatedly decompressing the textures required for one of the game’s leader scenes. Note that this is a DX11 benchmark.

While Turks and the 6670/6570 are quite a bit faster than the 6450, both are still so-so as compute cards. Although the unit of measurement (fps) is effectively arbitrary, what isn’t arbitrary is that the 5750 is nearly 33% ahead of the 6670. Juniper (5750/5770) seems to be the floor for good compute performance on AMD cards.

Our second compute benchmark of the day is SmallLuxGPU, the GPU ray tracing branch of the open source LuxRender renderer. While it’s still in beta, SmallLuxGPU recently hit a milestone by implementing a complete ray tracing engine in OpenCL, allowing them to fully offload the process to the GPU. It’s this ray tracing engine we’re testing. AMD in turn recently hit a milestone by offering the OpenCL runtime with their drivers, so SmallLuxGPU will now run out of the box without the AMD Compute SDK.

At 3200K rays/sec, the 6670 actually beats the Radeon HD 4870 by about 10%, showcasing just how far AMD’s compute performance has come compared to the 4000 series. The 6670 also enjoys a similar lead over the 5670 thanks to its additional shaders, but the 720SP 5750 looms large above it.

At the other end of the spectrum from GPU computing performance is GPU tessellation performance, used exclusively for graphical purposes. Barts’ tessellation improvements are inherited by Turks, which should give it an edge over the 5670.

What ended up surprising us the most with our DX11 Detail Tessellation Sample test wasn’t so much the raw performance of the 6670, it was the performance relative to some other AMD cards. At a high tessellation factor the Radeon HD 5870 isn’t all that far ahead, and meanwhile the 5750 is edged out by the 6670 in all cases. Given that our most tessellation-intensive games struggled at 1680x1050 on the 6670 it may not be powerful enough to really use its tessellation abilities in full on top of a heavy graphical workload, but the basis for respectable tessellation performance is there.

Power, Temperature, and Noise

Last but not least as always is our look at the power consumption, temperatures, and acoustics of the Radeon HD 6670 and Radeon HD 6570. As we’ve noted previously, these reference designs won’t be sold in the retail market, so while our performance data and power data should be consistent with other cards, temperature and noise data is going to be very dependent on the specific cooler used.

Radeon HD 6570/6670 Voltage
6570 Idle	6670 Idle	6570 Load	6670 Load
0.9v	0.9v	1.1v	1.1v

Both the 6670 and 6570 have the same load voltage of 1.1v. This caught us a bit off guard, since lower clocked parts normally enjoy a lower load voltage. This will be of importance when we’re looking at load power consumption.

At idle we’re running up against the limits of what a 1200W PSU can do.                In reality the 6450 should be consuming at least a few watts less than the 6670 and 6570. The fact that all 3 cards are among the lowest power consuming cards at idle is not a mistake, however.

Our Crysis power data surprised us some, and we’re still trying to get to the whys. Both the 6670 and 6570 ended up being sub-200W; this is notable because the 5670 draws 214W in the same test and has a slightly lower TDP than the 6670. In spite of what the specs say, for whatever reason the 6670 is clearly well ahead of the 5670 in both power consumption and performance when it comes to Crysis.

FurMark confirms what we’ve seen with Crysis: the 6670 and 6570 consume less power than the 5670 even with their similar TDPs. In fact the 6570 consumes 5W more than the 6670 even though the latter is the higher clocked and should be the higher consuming card. Given that both cards use fully enabled GPUs at the same load voltage, the difference looks to come down to minor variations in the quality of the individual chips. In this case our 6670 appears to use a better chip than our 6570, hence the difference.

In any case the load power consumption looks very good. The 6670 edges out even the much slower GT 430, while the 6570 similarly edges out the GT 220 DDR3. AMD has done a lot of work on power consumption for Northern Islands and it shows. Meanwhile the fact that the 5750 draws around 25W more is both a curse and a blessing – 25W is quite a lot when Turks cards only draw around 60W in the first place, on the other hand that 25W comes with a lot more performance.

Idle temperatures are roughly where we expect them to be. The 6570 is consistent with the other cards using that model cooler, while the 6670’s new double-wide cooler makes it competitive with all but the coolest of our cards.

Since we’ve already established that the 6670 and 6570 have similar levels of power consumption, temperatures are almost entirely a product of the cooler in use. The 6670 has a larger cooler that can move more air, and as a result it’s much cooler than the 6570 under Crysis. In fact the 6570 is probably at its limits for what AMD’s single-wide low-profile cooler can do; dissipating the roughly 40W the GPU produces is no small task for such a small cooler.

Coming from Crysis our temperatures rise slightly, as is normally the case. At 69C the 6670 is extremely well cooled, while the 6570 is quite warm at 82C. Though we won’t see either cooler at retail, this should be a good idea of the range we should expect to see. A double-wide cooler should make for a very cool card, while a single-wide cooler delivers reasonable performance but it will definitely be warm.

For idle noise, both the 6670 and 6570 are where we’d expect them to be, which is to say almost silent. The 6570 does end up being a bit louder than our noise floor though, which is an interesting outcome given that the 6450 and 5570 are at the noise floor using the same cooler.

We’re not going to read into our noise results since partner cards will be so different, but if they’re anything like the AMD reference cards then things look pretty good. We’ve had quieter cards, but nothing pushing 60W in a low-profile form factor. If partners go with a full-factor design, then they should be able to meet (or beat) the GTS 450 in noise levels at the cost of some space.

Meanwhile we expect to see passively cooled cards, which should be quite attainable with a full-factor card to hold a larger heatsink.

Final Words: It’s Nobody’s Business But The Turks

At the high-end of the market we’re used to seeing AMD and NVIDIA trade blows with similarly priced, similarly performing cards. In the sub-$100 market this is not the case however. At least among the non-crippled versions of various GPUs, NVIDIA and AMD’s GPUs seem to do a better job fitting in between each other than they do going head-to-head with each other. Pricing is what makes these cards compete in the market, rather than their GPUs performing similarly.

In this case Turks seems bound to fit in between NVIDIA’s GF106 (GTS 450) and GF108 (GT 430) GPUs, a position that Redwood (5670) has similarly occupied up to this point. Between those NVIDIA GPUs there’s a very small slice of space that AMD can call their own, at least before pricing is taken into account.

The fact of the matter is that the sub-$100 retail market is extremely crowded, with the space being shared by last generation cards on close-out, current generation cards on sale, and newly introduced cards like the Radeon HD 6670 and Radeon HD 6570 that are designed for that price range in the first place. Compared to the 5600/5500 series cards that the 6670 and 6570 replace, both cards are nice mid-cycle updates. Performance is up by over 10% for the 6670, while the 6570 is very close to the 5670. Without a die shrink, this is probably the best AMD can do to iterate on Redwood.

The problem of course is that based solely on performance, the sub-$100 market is too crowded. As long as power consumption and a low-profile form factor are not concerns, the Radeon HD 5770 and GeForce GTS 450 are both regularly on sale for under $100 and are easily 30% faster than the 6670. Cards like the 6670 and 6570 have their place, but it’s not as performance kings. For that, higher-tier cards on sale have and will continue to be the better buy.

So where to the 6670 and 6570 fit in? That’s hard to say. The 5570 was the ultimate HTPC card, but the 6450 has dethroned it. Unless you can snag a higher-tier card for $80 the 6570 is a good deal – or at least no worse than the 5670 – but the 6670 isn’t as well defined.

The best qualities of both cards are that they’re low-profile cards that don’t need an external power source, and that this is a reference quality we should see in partner cards. With the exception of a couple of one-off non-reference designs like the much more expensive PowerColor Go Green 5750, the 6670 and 6570 are going to be the fastest cards available that don’t require external power. In the OEM market that AMD sold these cards to first, that’s a significant advantage. For the retail market however this is only of particular use for HTPC users that need a bit more gaming horsepwer. For every other use in this price segment, time will tell if it is enough.

Ultimately Turks and the 6670/6570 are technically superior, but at $99 and $79 respectively they won’t have that same superior position on the open market.