ATI RV740 review/preview

[mczak]

Sapphire HD4730 have the same specs. It's interesting - core 7% slower than HD4770, but GDDR5 modules are 12,5% faster than HD4770 (exactly as fast as HD4890). I expect it will be about as fast as HD4770...

Nope. If you see it on the shelf, you better start running.
Looks like it lost half the ROPs.
Review http://ht4u.net/reviews/2009/powercolor_ati_radeon_hd_4730/ (in german), but in short slower than HD4770, (slightly) slower than HD4830 even, but with twice the power draw of the HD4770 (especially idle power draw is completely unacceptable for a card of this performance category).

 

[no-X]

Hmm, halving the ROPs seems to be the limit here

Anyway, it's very nice illustration of the ROPs influence on performance. Stalker, FarCry 2, COJ, AC and COD: WaW seems to be pretty limited by ROPs performance - 16xx*10xx MSAA 4x performance difference is 20-25% (HD4730 with 8 ROPs vs. HD4770 with 16 ROPs).

Hypotetical RV790 with 32 ROPs would be at least as fast as GT200b...

//edit: I'm beginning to believe, that if the 180mm2 RV8xx GPU has some ROPs, it could be 24 ROPs + 192 bit memory interface (or 28 ROPs + 224 bit)...

 

[CarstenS]

It's the way of plaguing the market with the old generation mainstream cards based on power hungry last gen. monsters chips [...]

WRT there was a slide on the launch presentation for HD 4770 and it's advanced tech, that made me smile. It was something about some company putting new stickers on old hardware.

Except that said company has also failed to introduce their next-gen-cards yet.

 

[AlexV]

It was something about some company putting new stickers on old hardware. .

IIRC that was about putting new stickers on the same old SKU. In this case you get differences between differing SKUs beyond the nomenclature change. Whether or not this "flood the market with all kinds of non-sensic" SKUs is a good thing or not is a wholly different discussion alltogether.

 

[CarstenS]

Actually the first item was a HD 3850 compared to 8800 GT - the former entitled to a new architecture, the latter to an old architecture with a process shrink. Which is IMO quite arguable.

 

[keritto]

WRT there was a slide on the launch presentation for HD 4770 and it's advanced tech, that made me smile. It was something about some company putting new stickers on old hardware.

Except that said company has also failed to introduce their next-gen-cards yet.

Was it on rv740 presentation or recent evergreen one? Anyway you nicely cut out the real issue here and that's that they just put power hungry monster into a new series as "mainstream" so RV740 could easily service ATi as some dx10.1 mainstream part just with a new name. Let's hope they don't put (relabel as nVidia ) RV790 in HD5000 lineup if they're cheaper to produce cause old 55nm is simply cheaper than 40nm process (like nVidia did) ... I really didn't wrote that as some Krieglager (?) statement. But you nicely d-tour me there W

 

[kemosabe]

Comes as no surprise that TSMC is apparently compensating AMD for poor 40nm yields with preferential pricing.

 

[Vincent]

Comes as no surprise that TSMC is apparently compensating AMD for poor 40nm yields with preferential pricing.

I doubt that ........

 

[ShaidarHaran]

I doubt that ........

Why? Sounds logical and even probable.

 

[kemosabe]

With such crappy yields and Global Foundries knocking at IHVs' doors, can't imagine how TSMC would consider it good business sense to do otherwise.

 

[keritto]

Comes as no surprise that TSMC is apparently compensating AMD for poor 40nm yields with preferential pricing.

Well they'll compensate them but not in preferential pricing but on preferential treatment when their 40nm/32nm will be in full shine they get more capacities befor nvidia few month before and that can meant a much jus like in rv670 vs gt92 case. Probably they extend that even to some LowPower 28nm in 6-12 month. ATi doesn't need refunds but they wouldn't be eager to see how nVidia makes huge profit on their backs with preferential on TSMC 40-32nm on something they invested to TSMC.

 

[entity279]

The thing is, AMD is virtually a lost costumer for TSMC because of GF. So they wouldn't have too much interest in pleasing them, i guess.

But then again, they may have contracts to observe

 

[Kaotik]

The thing is, AMD is virtually a lost costumer for TSMC because of GF. So they wouldn't have too much interest in pleasing them, i guess.

But then again, they may have contracts to observe

TBH I don't think ATI is moving to GF, they have long lasting relationship with TSMC, and designing for other companys process might not be worth it, even if GF will eventually move to new processes faster than TSMC

 

[ShaidarHaran]

To reinforce the idea that AMD will not rely on GF for GPU production in the immediate future - GF has zero experience producing GPUs. If and when AMD does utilize GF for GPU production it will start with a trickle, likely IGPs (possibly Fusion).

 

[entity279]

Yes you have a point, it might take more for ATI to get there.

Now that I think of it, it would really be ironic if Nv were to sitch to GF faster.

 

[keritto]

To reinforce the idea that AMD will not rely on GF for GPU production in the immediate future - GF has zero experience producing GPUs. If and when AMD does utilize GF for GPU production it will start with a trickle, likely IGPs (possibly Fusion).

Well don't think they'll have problems developing GPU indahouse, but for now more problems for them lies in limiting capacities for global deployment of their gpu, cause GPUs as we see in the last few years were deployed in much shorter time than CPU are (usually 6-8 month respin while ATi was developing their CPUs for 18-24 month)

The market would punish them no matter how advanced tech they have. Cause people would buy chips/cards that are in abundance and with more chips produced they cut down r&d prices and chip cost at the end like we could see in RV770 becoming almost value chip selling bigger die for less price than RV670 was ever sell in retail. And they can afford it probably because RV770 outsells RV670 2:1 or maybe more, with virtually no wasted chips

 

[ShaidarHaran]

Yes capacity would likely be an issue as well. At least until they get their fab in upstate New York online.

 

[Vincent]

Yes capacity would likely be an issue as well. At least until they get their fab in upstate New York online.

Conversion from TSMC to Global foundries could be the issue as well. (Library)

 

[rjc]

Semiaccurate on why there are no RV740s:

There have been a lot of claims about low yields for RV740, some as low as 20%. Having heard much more exact figures, lets just say those numbers aren't even close, yields are much higher than that. ATI ran a full production run of wafers, and got tens of thousands of working parts back. The reasoning behind running the wafers was simple, TSMC 40nm has many problems, some of which aren't seen until you are in a volume run, was to figure all out the problems out.

In the end, there are basically no more R740 based parts for a bit. We hear production will start up again in late summer or early fall. TSMC is promising tolerable yields for July, likely wafers in for July, not out, so September seems like a good bet. Until then, there are a lot of R770 chips to bleed off before the Evergreen family obsoletes them.

 

[trinibwoy]

Ah, I see bizarro Charlie has finally made his way to earth If only he could evenly apply his sugarcoating and cheery perspective on bad situations.