AMD’s Radeon HD 5770 & 5750: DirectX 11 for the Mainstream Crowd
by Ryan Smith on October 13, 2009 12:00 AM EST- Posted in
- GPUs
4 chips in 6 months.
This is the schedule AMD’s GPU engineering teams committed themselves to for the launch of the Evergreen family. The entire family from top to bottom would be launched in a 6 month period.
Last month AMD took the first step of that plan with the launch of Cypress, the forebear of the family and the source of the Radeon HD 5870 and 5850. Today AMD is taking the next step in the launch of the Evergreen family by delivering the 2nd and final Evergreen chip of the year: Juniper. Or as the products based off of them are known as, the Radeon HD 5770 and 5750.
| ATI Radeon HD 5870 | ATI Radeon HD 5850 | ATI Radeon HD 5770 | ATI Radeon HD 5750 |
ATI Radeon HD
4870
|
ATI Radeon HD
4850
|
|
| Stream Processors | 1600 | 1440 | 800 | 720 | 800 | 800 |
| Texture Units | 80 | 72 | 40 | 36 | 40 | 40 |
| ROPs | 32 | 32 | 16 | 16 | 16 | 16 |
| Core Clock | 850MHz | 725MHz | 850MHz | 700MHz | 750MHz | 625MHz |
| Memory Clock | 1.2GHz (4.8GHz data rate) GDDR5 | 1GHz (4GHz data rate) GDDR5 | 1.2GHz (4.8GHz data rate) GDDR5 | 1.15GHz (4.6GHz data rate) GDDR5 | 900MHz (3600MHz data rate) GDDR5 | 993MHz (1986MHz data rate) GDDR3 |
| Memory Bus Width | 256-bit | 256-bit | 128-bit | 128-bit | 256-bit | 256-bit |
| Frame Buffer | 1GB | 1GB | 1GB | 1GB / 512MB | 1GB | 1GB / 512MB |
| Transistor Count | 2.15B | 2.15B | 1.04B | 1.04B | 956M | 956M |
| TDP | 188W | 151W | 108W | 86W | 150W | 110W |
| Manufacturing Process | TSMC 40nm | TSMC 40nm | TSMC 40nm | TSMC 40nm | TSMC 55nm | TSMC 55nm |
| Price Point | $379 | $259 | $159 | $129 / $109 | $140-$160 | $109-$129 |
In our 5800 series launch article, we briefly discussed Juniper and the other members of the Evergreen family. With Cypress a bit too big and a bit too expensive to hit mainstream prices, a new chip was introduced in to AMD’s usual 3 chip stack to cover that segment of the market, and that chip was Juniper.
What’s Juniper? In a nutshell, it’s all of Cypress’ features with half the functional units (and no Double Precision for you scientist types). DirectX 11, Eyefinity, angle-independent anisotropic filtering, HDMI bitstreaming, and supersample anti-aliasing are all accounted for. For more information on these features, please see our Radeon 5870 launch article from last month.
With half of the functional units left behind, we’re left with 10 SIMDs, giving us 800 stream processors and 40 texture units, while the ROP count has also been cut in half to 16, in turn giving us a 128-bit memory bus. If Cypress was 2 RV770s put together, then Juniper is the closest thing you’re going to see to RV770 coming out of the Evergreen family.
Juniper
With the reduction in functional units, Juniper becomes a leaner and meaner core. The transistor count is 1.04 billion, a little less than half of Cypress and about 100 million more than RV770. The die size of this resulting core is 166mm2, significantly less than both Cypress and RV770, the latter due to the smaller process size. RV770 for comparison was 260mm2.
From Juniper we are getting the 5770 and the 5750. The 5770 is a full Juniper, with all of Juniper’s functional units enabled and the card running at what amounts to a full speed of 850MHz (the same as 5850). The 5750 is slightly cut down, much like 5850 is compared to 5870. Here we have 1 SIMD disabled, and the core clock reduced to 700MHz. This is a notable departure from how AMD handled the 4870/4850 split, where 4850 was differentiated using a slightly slower core and much slower RAM, without the need to disable any SIMDs.
The smaller Juniper core also affords these cards lower power usage than the 5800 series. The 5770 is 108W at load and 18W at idle, meanwhile the 5750 is 86W at load and 16W at idle.
As an interesting aside, when AMD started sampling Evergreen cards to game development houses and other 3rd parties, they were Juniper based, and not Cypress based. The Juniper team was rather proud of this, particularly since Juniper came back from TSMC second. They also had less time to get their GPU up and working than the Cypress team did, since they had to wait on Cypress before being able to finish work on some elements. This is what makes AMD’s 6 month rollout all the more impressive, since it means the non-Cypress teams had less time to get their work done than they have in previous product cycles.
Facebook
Twitter
RSS
117 Comments
View All Comments
flipmode - Tuesday, October 13, 2009 - link
What in the world is going on with this game? 8800 GT beats the 4850? No, sorry, I don't buy that. Something is wrong here. The 5770 beats everything? If that is the case, then this game should immediately be removed from the bench suite - games in the bench suite should help us understand the general performance characteristics of the hardware and a game that returns such erratic results actually distorts that understanding.Griswold - Tuesday, October 13, 2009 - link
On page 13 you say:"The 3870 beats it by 14W at the cost of a significant degree of performance, while the 8800GT is neck-and-neck with the 4770, again with a decent-sized performance gap."
You certainly meant 5770 there. But this brings me to a question: Why isnt the 4770 included here? As an owner of that card, I'm very much interested in the performance/power/noise difference - just ditch one of the relatively irrelevant SLI or CF combos. I dont think too many care about comparing high-end multi-GPU with performance parts such as the 5770 and 5750, even if its 57xx in CF.
flipmode - Tuesday, October 13, 2009 - link
Ryan - thanks so much for the review. Nice job. It does seem like a 5750 Crossfire would be an interesting value - moreso than the 5770 since the latter is overpriced.And, Anand, I love your site, and don't take this personally, but, PLEASE GET A COMMENT SYSTEM THAT DOES NOT TOTALLY SUCK!
Check out TechReport for an example of the awesomest comment system in the universe.
PLEASE!
Ryan Smith - Tuesday, October 13, 2009 - link
AMD only sent out 1 5750, so I don't have a second one to Crossfire at this time.Roland00 - Tuesday, October 13, 2009 - link
It makes no sense (beside bad drivers) for the 5770 to lose to the 4850. The 5770 has more memory bandwidth (76.8) compared to the 4850 (63.55 gb/s), due to the 4850 sticking with ddr3, even with the 128 bit bus. The 5770 is also clocked 36% faster than the 4850 (850 vs 625).Yet the 5770 underpeforms the 4850 being almost tied?
Zool - Tuesday, October 13, 2009 - link
Maybe the 4*64bit memmory controlers on the perimeter of the chip keep up the data better than 2*64bit controlers with higher bandwith.I think that they could make it at least 192 bit (3*64bit).
Zool - Tuesday, October 13, 2009 - link
Actualy where the hell are the Cypress and Juniper die shots ?I cant find a single one on net.
Ryan Smith - Tuesday, October 13, 2009 - link
AMD is not releasing die shots.dgz - Tuesday, October 13, 2009 - link
Looks to me like AMD is trying to lure people into buying the remaining 48** cards. Once the old chips are cleared, the price of 57** will no doubt drop.GrizzlyAdams - Tuesday, October 13, 2009 - link
What really has me concerned is how the 5870 is scaling in these tests.The 5870 core is essentially two 5770s strapped together, and you would hope scaling would be near linear. When two 5770s in crossfire match or even beat a 5870 I'm left scratching my head.
Somewhere there is a significant bottleneck in the 5870's design, and I'm wondering where that is. Anyone have any clue?
Hopefully a driver update will fix these issues, because if not there is a lot of wasted silicon on each of these chips...