Technical data in comparison
You really have to ask yourself why AMD leaves such a big performance gap unused in the world instead of plugging it competitively at the moment. We did not manage to reach the desired performance midpoint stably and sensibly with an overclocked HD 7770 even at a clock of 1.15 GHz. More shaders for Cape Verde? Also rather unlikely, which actually rules out a potential card above the HD 7770.
Now, one could speculate that AMD’s Pitcairn yield is apparently so high that there simply aren’t enough partially defective chips that could simply be recycled cheaply on a card between the HD 7700 and HD 7800 series. However, it would certainly not be unwise to turn to a hungry target group with an artificial restriction, as has happened often enough in the recent past, which is exactly waiting for cards with this then emerging performance level.
Let’s take another look at the initial situation:
| HD 7770 |
Engineering Sample HD 7850 768SP |
HD 7850 |
HD 7870 |
|
|---|---|---|---|---|
| Stream processors | 640 | 768 |
1024 | 1280 |
| Texture units | 40 | 48 |
64 | 80 |
| Full Color ROPs | 16 | 32 |
32 | 32 |
| GPU clock | 1000 MHz | 860 MHz |
860 MHz | 1000 MHz |
| Texture fill rate | 40 Gtex/s | 41.3 Gtex/s |
55 Gtex/s | 80 Gtex/s |
| Memory clock | 1125 MHz | 1200 MHz |
1200 MHz | 1200 MHz |
| Memory bus | 128-bit | 256-bit |
256-bit | 256-bit |
| Memory throughput | 72 GB/s | 153.6 GB/s | 153.6 GB/s | 153.6 GB/s |
| Video RAM | 1 GB GDDR5 | 2 GB GDDR5 | 2 GB GDDR5 | 2 GB GDDR5 |
| Transistors (billions) | 1,5 | 2,8 | 2,8 | 2,8 |
| Structure size | 28 nm | 28 nm | 28 nm | 28 nm |
| Power connections | 1 x 6-pin | 1 x 6-pin | 1 x 6-pin | 2 x 6-pin |
| Power consumption (max) | 80 watt | 120 Watt* |
130 watt | 175 watt |
| Street price (from approx.) | 110 euros | 160 Euro (desired price) |
210 Euro | 290 Euro |
*) please note that it is not a final GPU
Teardown
Let’s first look at the front side of the stripped test card. We see four phases for the GPU and one for the 2 GB of memory in addition to the one 6-pin power connector, which is divided into a total of 8 modules.
Actually, this starting position is quite plausible, because 25% less shader units infer about 20-25% less performance in the end, depending on the application and situation. Since this is an engineering sample and not an end customer chip, some points like power consumption or temperature development can of course not be weighted as heavily as we would otherwise. Despite everything, we can also hand out a reassurance pill to AMD in this regard – there is absolutely nothing that is unacceptable if you classify it correctly.
So even at this stage, our test sample is not a disappointment. Therefore, let us now turn to our benchmark run, which corresponds to that of our charts.
8 Antworten
Kommentar
Lade neue Kommentare
Moderator
Veteran
Neuling
Moderator
Urgestein
Mitglied
1
Alle Kommentare lesen unter igor´sLAB Community →