True, but with watercooling and some serious overvolting these may OC like crazy. Would they still be unimpressive if they were OC beasts?
Maybe they will suck at OCing, but im betting these will hit 1000mhz(+?) with water and the right voltage.
So, do we know anything about RV670 yet?
- Thread starter nicolasb
- Start date
True, but with watercooling and some serious overvolting these may OC like crazy. Would they still be unimpressive if they were OC beasts?
Maybe they will suck at OCing, but im betting these will hit 1000mhz(+?) with water and the right voltage.
Let's discover a new concept:IMHO, it's unimpressive. I should've pointed that out. Because I really dont care all that much about power consumption, noise&heat(watercooling is my good friend). I'm more into performance types of things...or features-the RV670 brings neither(DX10.1 solving Global Illumination-yeah, sure, now why didn't we figure out ages ago that it was that simple). Probably there exists a niche for those like myself.
It's also unimpressive in terms of the about a billion five hundred million posts:OMG, it's going to OWNORRZ, it is ATi's comeback etc. Call me odd, but fanboysm pisses me off considerably(from either side of the fence), and the fact that it's simply a middle ended not excessively sexy refresh gives me a limited ammount of satisfaction. I'd still like to get my grubby mitts on a pair of 1Gb 2900XTs from Diamond, just for shits and giggles...hmm, time to sell the 512MB ones.
You're likely to be correct in that from a purely performance-oriented perspective it's unlikely to be impressive, excepting price to performance ratio.
Sound_Card
Regular
Let's discover a new concept:IMHO, it's unimpressive. I should've pointed that out. Because I really dont care all that much about power consumption, noise&heat(watercooling is my good friend). I'm more into performance types of things...or features-the RV670 brings neither(DX10.1 solving Global Illumination-yeah, sure, now why didn't we figure out ages ago that it was that simple
It's also unimpressive in terms of the about a billion five hundred million posts:OMG, it's going to OWNORRZ, it is ATi's comeback etc. Call me odd, but fanboysm pisses me off considerably(from either side of the fence), and the fact that it's simply a middle ended not excessively sexy refresh gives me a limited ammount of satisfaction. I'd still like to get my grubby mitts on a pair of 1Gb 2900XTs from Diamond, just for shits and giggles...hmm, time to sell the 512MB ones.
I would not expect a nasty come back till R700, but in the mean time, I fail to see how RV670 does not impress. Comparing it to the 1gb XT that is... half the price, less than half the power consumption(means less heat, and hopefully silent cooling solutions), and UVD. I'm pretty sure thanks to 55nm that this chip could yeild higher OC'ing results than the 80nm power leaking mammoth R600 is, water cooling or not. Considering that 825mhz is around 10% faster than 750mhz, makes me wonder what the performance would be at 1ghz(ya I said it, 1ghz, I think this chip could reach it) assuming your not bandwith limited.
Sound_Card
Regular
Sometimes twice get's the message across...:smile:
Yes but its not competing against the r600, thats the problem, its against a 8800 gt, so there is more then just AMD's cards you have to look at.
Sound_Card
Regular
That's where price comes to play.
I understand that the OC potential is there but like I asked before, what does that mean in performance? If I can OC a 3850 to 850MHZ what will I gain in performance? Is it an additional 5 FPS or 10 FPS across the board (for example)"? Right now a big increase in OC performance is the only thing that will allow this card to compete with the GT IMO! This is what I am looking for.
Last edited by a moderator:
Silent_Buddha
Legend
Or maybe you need 2 of HD3870 1GB DDR4 cards
Just only a week left for the RV670 parts and the INQ added more to it...
AMD tells partners to focus on 3850 with GDDR3
If that is true it looks like AMD is doing a couple things this winter...
1. Conceding the high end and performance segment to Nvidia with only the 3870 to keep at least a presence in the performance sector.
2. Aiming to dominate the high volume budget, value, and mid-range sector. And by association the OEM market as that's where the majority of OEM contracts are.
It'll be interesting to Nvidia's response to a HD 2400 and 2600 undercutting their respective NV parts even more. As well as AMD positioning the 3850 as direct competition to the 8600 GTS.
From an enthusiast perspective this would be a major downer. From a business standpoint this is brilliant considering the technology position AMD/ATI is in currently with regards to R(v)6xx compared to G8x/G9x.
The fact that Rv670 is such a small 55nm chip means that ATI should still be able to attain decent margins at a low price point.
That in turn will force Nvidia to either 1) concede the budget/mid-range to ATI or 2) lower prices thus cutting into their margin until they can get a proper chip out in response.
Considering how large G92 is, I don't think it could compete effectively (from an economic point of view) in the sub 200 USD range.
It does suck though that from how things look right now, the 8800 GT is the only card putting price pressure on the 8800 GTX. Although a rumored 3870 X2 might be able to do so, I'm an avid hater of dual GPU solutions until they come up with an elegant solution to rendering that doesn't involve AFR.
Regards,
SB
Don't trust the naysayers, it could. What matters is revenue/wafer, not some ridiculous metric like revenue-for-a-given-SKU-of-a-given-size.Silent_Buddha said:
And chip revenue for a given SKU is hardly proportional to the final board price. The amount of money that goes to DRAM and the PCB must be substracted to what you can get for your chip.
NVIDIA's strategy for the ~$150 segment is perfectly obvious: 12 ROPs/6 Clusters G92 SKU(s) with 384MiB of GDDR3 (192-bit saves money on the PCB, and you only need 6 DRAM chips). For around ~$175, a 256MiB 700MHz GDDR3 8800 GT is also a fincancially viable proposition once PCB production ramps up. It's hard to say exactly how performance and margins will compare to the 3850, but I would tend to believe NV should be quite competitive.
That doesn't change the fact that, this time around, it looks like ATI will be very very cost-competitive (and price-competitive). It's certainly good for the industry, and for the consumers.
P.S.: NVIDIA might prefer not to *have* to sell a lot of 192-bit SKUs though, because if yields are high enough then they'd have to sell fully functional chips for less money. As such, I would not exclude the possibility that these SKUs will not be present in both the channel and at OEMs. My guess would actually be channel-only, too, since margins are better there and you don't have contractual obligations.
I'm not so sure I buy that, given your recent discussion w/DK on RWT regarding this very same subject (NV x86 thread)
If NV can make money selling 8800 GT 512MB cards @ $200 or less, ATi can make more money doing the same. NV's only answer here is smaller frame buffers (which ATi can match) or a new, more cost-effective mid-range GPU.
Probably ATI's own catalyst driver-software overdrive utility might allow to hit 800Mhz+ for GPU.
DegustatoR
Legend
112SPs for GT instead of 96?
I would imagine those things are hard to change one the chips have left the foundry.
Geeforcer: The chip probably always had 128 SPs; what changed would be the number of SPs activated for this specific SKU.
ShaidarHaidan: Pfff, not my fault if those damn engineers and technically oriented people don't understand economics, now is it?
However, AMD doesn't seem to be able to have a 192-bit SKU (as they don't have that kind of granularity). So NVIDIA should be able to compete around ~$150 with that advantage (the PCB is cheaper, and you only need 6 chips for 384MiB versus 8 chips for 256MiB on a 256-bit bus).
On the other hand, ATI might be able to generate a 256MiB 128-bit SKU out of RV670 for less than $150; that's a fair bit of redundancy, but it's better than having nothing in that price segment or wasting chips with a half-defective memory controller. And since their die is so small, that's financially viable, while it definitely wouldn't be for NV.
ShaidarHaidan: Pfff, not my fault if those damn engineers and technically oriented people don't understand economics, now is it?
However, AMD doesn't seem to be able to have a 192-bit SKU (as they don't have that kind of granularity). So NVIDIA should be able to compete around ~$150 with that advantage (the PCB is cheaper, and you only need 6 chips for 384MiB versus 8 chips for 256MiB on a 256-bit bus).
On the other hand, ATI might be able to generate a 256MiB 128-bit SKU out of RV670 for less than $150; that's a fair bit of redundancy, but it's better than having nothing in that price segment or wasting chips with a half-defective memory controller. And since their die is so small, that's financially viable, while it definitely wouldn't be for NV.
No doubt, the question is when (if ever) you would have to change number of enabled SPs to make the launch date. Mid-September seems far too late for that.
And how fast did the 112SP G80 GTS showed up? Even faster. IMHO you are over-estimating the time Nvidia needed to make the change.
