Radeon HD 7770 and 7750 Specs surface

[bladerash]

VLIW4/5 was and is very efficient for game workloads. GCN never was about improving on that in the first place! Because GCN is not worse than VLIW for these types of workloads is achievement on it's own. If you want to compare GCN to VLIW and see the difference just look at compute

GCN is quite a bit of a step in the right direction

Really? Cuz 1,5Billion transistors for it seems a lot to me. Juniper has only 1Billion transistors and Barts has 1,7Billion.
Cape Verde has about 12% less transistors than Barts, but Barts has double the bit bus, double the ROPS and almost double the amount of shaders, even though they are VLIW5.
So.. GCN better be more efficiënt and while at it, better double the performance per core too since a GCN shader seems to be twice as big on average compared to a VLIW5 shader.

[kuroikenshi]

Really? Cuz 1,5Billion transistors for it seems a lot to me. Juniper has only 1Billion transistors and Barts has 1,7Billion.
Cape Verde has about 12% less transistors than Barts, but Barts has double the bit bus, double the ROPS and almost double the amount of shaders, even though they are VLIW5.
So.. GCN better be more efficiënt and while at it, better double the performance per core too since a GCN shader seems to be twice as big on average compared to a VLIW5 shader.

what workloads are you actually referring to when you say "efficient" and "performance"

[-Boris-]

Really? Cuz 1,5Billion transistors for it seems a lot to me. Juniper has only 1Billion transistors and Barts has 1,7Billion.
Cape Verde has about 12% less transistors than Barts, but Barts has double the bit bus, double the ROPS and almost double the amount of shaders, even though they are VLIW5.
So.. GCN better be more efficiënt and while at it, better double the performance per core too since a GCN shader seems to be twice as big on average compared to a VLIW5 shader.

That's the drawback with shaders capable of good compute performance.
For gaming today I guess VLIW4 is the most efficient per mm² and per watt.

[Apokalipse]

VLIW is very efficient in terms of density, but isn't necessarily going to be utilized to its full potential (especially with games which are very difficult or virtually impossible to make specialised VLIW code for)

If all games were capable of utilizing VLIW5 shader clusters to near 100%, the 5870 would be significantly ahead of the 480.

Hopefully GCN is capable of near 100% utilization without needing to be programmed for specifically - I think that is one of it's design goals.

[Alex-Ro]

Tomorrow wil be launch day and we are still not sure about complete specs

[Lanek]

Tomorrow wil be launch day and we are still not sure about complete specs

Peoples are enough busy with Kepler rumors lol for care about a low-middle gamer card.

[Lightman]

Really? Cuz 1,5Billion transistors for it seems a lot to me. Juniper has only 1Billion transistors and Barts has 1,7Billion.
Cape Verde has about 12% less transistors than Barts, but Barts has double the bit bus, double the ROPS and almost double the amount of shaders, even though they are VLIW5.
So.. GCN better be more efficiënt and while at it, better double the performance per core too since a GCN shader seems to be twice as big on average compared to a VLIW5 shader.

First of all I meant architecture and therefor per shader performance, not transistor count per work done. If you want to compare it that way then with almost every step back you will see less transistors used (but more specialized) needed for particular task. You can't really add more features like improved tesselation, UVD encode block, DX11.1 extension, better caching, vastly improved compute and power gating (big transistor hog) with same transistor budget as older design.
What is important is how well new architecture is aligned with todays tasks as well as manufacturing process.
For me GCN is placed very well in both. Die sizes for Thaiti and Cape Verde are smaller than chips they replace and are much more efficient at dealing with todays tasks we expect them to run. Besides power consumption is just great on these parts compared to older gen!

[AFQ]

Tomorrow wil be launch day and we are still not sure about complete specs

Lol thats how its supposed to be. Leaks ruin it. :p

[cold2010]

XFX 7770 Black ED





http://translate.google.com/translat...%2F%3Fp%3D3851

[ilkkahy]

Memory bandwidth on same level with ~5 years old HD3870.

[kuroikenshi]

Memory bandwidth on same level with ~5 years old HD3870.

not really sure if 5 years ago GDDR5 was in middle range cards...IF ANY... but hey I'm no expert.

[bladerash]

No one said anything about GDDR5.. he said bandwidth.. and yes.. bandwidth is on same level since bitbus is halved.

[Alex-Ro]

3870 was highend back then....7770 is not even middle,or something like that....

[kuroikenshi]

My apologies, but I still fail to see why you'd need more than 1 GB on these cards.

[informal]

If you manage to OC 7770 to 1.15Ghz you may end up pretty close to 5850(or even faster). How's that for a "mid-low" segment card. Even @ stock it will be faster than 6850.

[kuroikenshi]

I say mid range since there is Tahiti and New Zealand taking the upper tier in the performance class.

[informal]

I say mid range since there is Tahiti and New Zealand taking the upper tier in the performance class.

I was referring to Alex-ro (nonsense) post, not yours .

[AffenJack]

If you manage to OC 7770 to 1.15Ghz you may end up pretty close to 5850(or even faster). How's that for a "mid-low" segment card. Even @ stock it will be faster than 6850.

You're really sure about that? I don't think 7770 will be faster than 6850, except in benches , which don't like the vliw architecture.

[kuroikenshi]

I was referring to Alex-ro (nonsense) post, not yours .

lol phew!... owe you one bud

[kuroikenshi]

You're really sure about that? I don't think 7770 will be faster than 6850, except in benches , which don't like the vliw architecture.

I think you might be in for a surprise

http://elchapuzasinformatico.com/201...os-benchmarks/

[SKYMTL]

As everyone (hopefully) already knows, 3DM11 is a poor indicator of in game performance.

[Alex-Ro]

I was referring to Alex-ro (nonsense) post, not yours .

If you tell me i am nonsense,then you are an idiot because you do not know 7770 performance,you do not know specs or if it can beat 6850

and 5850 is already old card,a mid-low 2 generation above it should beat it.But what do i know?

[informal]

Yes Alex, you usually post nonsense comments (in bad English too). The card will have very good price/perf. after the initial price gouging ends. It will be around 6850's level of performance,maybe some 5% faster or slower depending on the game. 6850 is already pretty close to 5850,so it will be quite an achievement to be able to come close to it. On top of this comes OCing potential. If you think this is "bad" performance for the money then nobody is forcing you to buy it.

[Alex-Ro]

Ok,you sound pretty dumb so i will not mess with you anymore.But i like the fact that you know so many things about 7770 eventhough you did not touch/tested it.

But again,you are no genius are you informal?

[informal]

Keep the personal insults coming Alex,you are good at that . With everything else- better luck next time .