If AD103 doesn't come to desktop I will really dislike that line up. There's a huge gap without it. And it would be more efficient than clocking AD104 to the moon.
If AD103 doesn't come to desktop I will really dislike that line up. There's a huge gap without it. And it would be more efficient than clocking AD104 to the moon.
Yes, I suppose so. But is the 4080 going to be a cut down 102 like Ampere or 104 like Turing? I suppose we will see but AD104 seems too small for that.It'd certainly make it easier to swallow paying $1500+ if even the deepest cut AD102 is way faster.
Yes, I suppose so. But is the 4080 going to be a cut down 102 like Ampere or 104 like Turing? I suppose we will see but AD104 seems too small for that.
104 chip in Older xx80 cards used to be 2/3 of the big 102 chip.
Yeah, Nvidia wants to be sure that the 4090 defeats the 7900.
That's why I expect the price of the 4090 to be rather extreme as well, but for the other cards to have fairly reasonable MRSPs.
No way that RTX 4080 will get AD104, too small die. At worst it will get full or cut down AD103 84-80SM.
Remember, AD104 has only 192Bit bus.
I don't think that RTX 4060 will use AD106 die, this would have been downgrade in memory from 12GB to 8GB compared to 192Bit RTX 3060.
Yep, the target is 2.5GHz boost. Let's see how good is their new AI EDA toy in circuit layout optimization...As for the clock speeds, I feel like 2.2-2.3 is way too low. nVidia is getting 1.9-2 on the crappy SS8. I feel like fill rate is a bit of a bottleneck, and while nVidia could up the ROPs/TMUs per SM, I somehow doubt that will happen. Hell even Intel is getting more than that.
Where did you got that info? Any source?Yep, the target is 2.5GHz boost. Let's see how good is their new AI EDA toy in circuit layout optimization...
2.5 GHz? At 600W of power?
What do you make out of this?
nVIDIA said they'll keep selling RTX 3000 series in parallel after RTX 4000 are released, so they're not going to cannibalize RTX 3070 and 3080 with smaller memory sizes, I think we're not going to see 12GB..I don't see the point in putting 20GB in the 4080 Ti, if you care about more than 16 GB, you'd want the 24 GB.
Putting 12 GB in the 4060 might be a good move against an 8 GB 7600, so there is something that can be used to justify buying it over the 7600, assuming that the AMD card will beat it, as expected.
RTX 4080 will not get AD104 or lower, too big gap between AD104 and AD102
nVIDIA said they'll keep selling RTX 3000 series in parallel after RTX 4000 are released, so they're not going to cannibalize RTX 3070 and 3080 with smaller memory sizes, I think we're not going to see 12GB..
You are assuming that they are going to sell the entire range, which makes little sense. I expect them to keep selling the 3060 (and perhaps 3050), just like they kept selling the 2060.
Also, they never actually promised they would, but said that they might, depending on the situation at the time.
Just sayin, last time before Ampere that the x80 was not using x104 was Kepler 2, almost 9 years ago. It could still be a very deep cut AD102 but that seems unlikely because N5/4 yields are too good.
Yields don't matter since they will still bin everything to hit performance targets and not all functional silicon can do that. We could easily see a similar situation with Lovelace where the 4080 is an ~80% 102 die if AMD offers strong enough competition with RDNA3. Nvidia doesn't want the 4080 being compared with mid-range AMD cards. That makes AMD look like a winner more than anything else.