Reviews - Sapphire Radeon HD 7770 OC

[HEXUS]

A pre-overclocked take on AMD's Cape Verde XT GPU.

Read more.

[scaryjim]

Interesting that a < 5% core clock increase, allied to a > 20% memory clock increase, yields > 10% framerate increase. I think it's fair to say that this is a memory-limited design (no great surprise there, the (5|6)770 was too). Be interesting to know what the card would do at 1000MHz Core and 6000MHz effective memory...

[cppmonkey]

Only 640 cores vs 800 if the 6770? I would have expected them to have atleast matched the under of cores.
With the 7970 they added a great deal more cores for the new gen. The trend had been adding more core with each generation but this time they choose to take some away placing it in poor competitive position.


I guess they are just trying to eke out as much power from a little silicon as possible, hence the aggressive clock speed.

The ATi Radeon HD 4770 has an almost identical configuration, just 8 fewer texture mapping units
Is the AMD Radeon HD 9770 (if they use the same chip for 2 gens again) going to have a similar configuration to the 5770. Something along the lines of 800 steam, 50 texture mapping units and 16 render out units using the GCN architecture?

Personally I will be holding out for the 78?0 and perhaps even wait to see what Nvidia bring to the board before choosing my next graphics card

[scaryjim]

... The trend had been adding more core with each generation but this time they choose to take some away ...

Completely different class of shader though. And don't forget that the 6970 had less shaders than the 5870 but still massively outperformed it due to a much more efficient architecture. And the fact that the 7770 outperforms the 5770, with 20% less shaders, just goes to how much more efficient the new architecture is.

However, I agree with your general point - I would've expected AMD to look for a bigger performance improvement in the same power envelope, not roughly the same performance in a smaller power envelope. When the launch price is the same as the outgoing card too, you expect something better from the new generation...

[CAT-THE-FIFTH]

Interesting that a < 5% core clock increase, allied to a > 20% memory clock increase, yields > 10% framerate increase. I think it's fair to say that this is a memory-limited design (no great surprise there, the (5|6)770 was too). Be interesting to know what the card would do at 1000MHz Core and 6000MHz effective memory...

The core actually is clocked 150MHZ higher,ie,15% higher than a stock card.

I really don't understand what AMD is trying to do here TBH! The OC models seem to be running at between 1100MHZ to 1150MHZ so it seems the core is artificially downclocked.

They should have launched the card at around 1100MHZ with higher rated memory as this would put it at around HD5850 1GB level performance:

http://www.techpowerup.com/reviews/X...rclock/26.html

Are they trying to make the cards look worse on purpose?? I really think they are taking the pre-overclocked SKU thing too far ATM. It seems they are underclocking stock cards to make the bling enhanced OEM versions look better. This is all fine and dandy but it should not happen at the expense of stock performance. Power consumption and cooling don't seem the issues here IMHO.

[scaryjim]

The core actually is clocked 150MHZ higher,ie,15% higher than a stock card.

I was referring to the Hexus overclocking results, not the Sapphire v Stock results - they got the core to 1200, which is < 5% above the factory OC, but still got > 10% performance boost.

I really don't understand what AMD is trying to do here TBH! The OC models seem to be running at between 1100MHZ to 1150MHZ so it seems the core is artificially downclocked. ... It seems they are underclocking stock cards to make the bling enhanced OEM versions look better.

I suspect they simply decide what performance point they want the cards to address and tune the qualified clock speeds to that. It's not in AMD's interest to qualify the stock clocks too high as they have to make sure every GPU sold can meet that speed, and by qualifying at a lower speed than bare metal they can counteract possible fluctuations in yield down the line. In this instance though, I'm at a loss as to why they qualified the memory speed so low, when the architecture obviously benefits massively from any increase in memory bandwidth. They could have qualified at 5500MHz instead of 4500MHz and got an extra - what, 5% - 10% of performance out of the stock card? That would've made it competitive with the 6850 and *still* have given OEMs the opportunity to release OC models with > 1000MHz core clocks and memory speeds nearer 6000MHz - which it appears that the cards will do.

[shaithis]

My guess is it's just like the 6950 fiasco. Pretty much all the cards are capable and therefore the best way to make money is to downgrade the majority of them so that you can charge a premium for the better cards.

Just this time they have done it with clockspeed rather then streams.

Pretty much what we have seen happen year after year. You could ask the question "Why aren't all i5 2500s 4GHz stock?"

Of course the other explanation could be we will see new drivers soon which automatically "overdrive" or "turbo".....but I see way too many issues with that.