However it fits?
Where can I read up on this?
AMD Radeon RDNA2 Navi (RX 6500, 6600, 6700, 6800, 6900 XT)
Where can I read up on this?
Prices in Russia are currently atrocious - the cheapest 6800 is about $1130, the cheapest 6800xt is $1330 (which is ironically Red Devil), mining craze takes its toll. RTX cards are mostly delisted or are even crazier in prices ($1860 for the only RTX3080 that is available through biggest online price aggregator)
That's cheaper than here in Finland
You can't really, it's just observations made during the fall to this day on our most popular shops by myself and other finns, they don't have similar pages up as Proshop does
D
Deleted member 90741
Guest
https://videocardz.com/newz/tesla-m...tfs-of-power-can-play-cyberpunk-and-witcher-3
Rumor of RDNA2 in Tesla Model S
Rumor of RDNA2 in Tesla Model S
So it's not enought that automotive gets priority from (taiwanese) semiconductor manufacturers, now they start stealing our GPUs too?
neckthrough
Regular
I like the "Performance Target" and "Notes" fields in that chart
@CarstenS do you habe any update on this value? Im already confused if RDNA 2 has now 4 or 8 scan converter?
Navi21 has 4.
@3dcgi The Front end is not really clear. In the driver there are 8 Scan Converter (Rasterizer).
Also we have the twitter post wher it mentiond that navi 21 has 80% more polygonoutput than navi10.
0x22h on Twitter: "@Locuza_ @Digidi20 In the polygon throughput related tests, I can see up to 80% higher performance than 5700 XT." / Twitter
Also we have the twitter post wher it mentiond that navi 21 has 80% more polygonoutput than navi10.
0x22h on Twitter: "@Locuza_ @Digidi20 In the polygon throughput related tests, I can see up to 80% higher performance than 5700 XT." / Twitter
If you are checking mangos and Linux driver the Packer are coming after the scan converter. I think packer are taking the pixel from the rasterizer and send them to the shaders?
https://www.pcgamer.com/a-linux-update-may-have-let-slip-amd-big-navis-mammoth-core/
This might explain why NVIDIA's 3000 series is outselling AMD's 6000 series by ~30:1 (at least in Denmark):
1 Million Ryzen 5000 CPUs were Sold in Q4 2020: That's Just 10-12% of AMD's 7nm Capacity at TSMC; 3% for Big Navi [Update] | Hardware Times
Console CPU/GPU: ~7 million units
Ryzen CPU's: ~1 million units
6000 series: ~250-300k units.
1 Million Ryzen 5000 CPUs were Sold in Q4 2020: That's Just 10-12% of AMD's 7nm Capacity at TSMC; 3% for Big Navi [Update] | Hardware Times
Console CPU/GPU: ~7 million units
Ryzen CPU's: ~1 million units
6000 series: ~250-300k units.
arandomguy
Veteran
I would question the methodology and conclusion you draw from that article though. Well at least the specifics, I don't think anyone at this point is unaware that shipments are supply constrained as opposed to demand constrained in general for GPUs.
Unless I'm misunderstanding it the article extrapolates the number of Navi 21 GPUs sold via the ratio of scalped GPUs to AMD CPUs. Assuming a near similar ratio of scalped to non scalped sales for the two is a big assumption. I'd actually argue that reasonably you'd expect the scalp ratio to be higher for GPUs. If so the actual amount of sold GPUs would be lower then their estimates. The wafer allocation also seems to be extrapolated based on the amount of GPUs sold, as opposed to having a source for wafer allocation and therefore using that as an upper limit for Navi 21 supply. The calculation itself also makes a number of other assumptions.
Unless I'm misunderstanding it the article extrapolates the number of Navi 21 GPUs sold via the ratio of scalped GPUs to AMD CPUs. Assuming a near similar ratio of scalped to non scalped sales for the two is a big assumption. I'd actually argue that reasonably you'd expect the scalp ratio to be higher for GPUs. If so the actual amount of sold GPUs would be lower then their estimates. The wafer allocation also seems to be extrapolated based on the amount of GPUs sold, as opposed to having a source for wafer allocation and therefore using that as an upper limit for Navi 21 supply. The calculation itself also makes a number of other assumptions.
Jawed
Legend
Here's 36+ 6800XTs sat on the shelves (add them to your basket) :
AMD RX 6800 XT - Graphics Cards AMD | Ebuyer.com
And currently there's 7 of these:
AMD RX 6900 XT - Graphics Cards AMD | Ebuyer.com
4+ have sold since I last looked at 10:30am. They've been there for a few days now.
5950X seems to have more of a supply problem right now than 6000 series. I suppose Milan is gobbling them up.
The ASUS TUF card gets a glowing review from Hardware Unboxed, for what it's worth.
Meanwhile:
Week 5 Mindfactory.de Total Gaming GPU Sales 2190 Units Nvidia 62.10% and Radeon 37.90% : realAMD (reddit.com)
1) RX 6900XT = 215 Units.
2) RX 6800 = 215 Units.
3) RX 6800XT = 210 Units
They still ignore the fact that those are not all the 7nm products AMD is making.
In addition to those, they're making current and next gen Epycs, Threadrippers, older Ryzen CPUs (that >1 million, not ~1 million, from Mercury Research is only 5000 series), two generations of Ryzen APUs, CDNA, RDNA1 and possibly even Vega20 - all on TSMC 7nm
edit: oh, and their 6000 series number is based on some bot tracking scalpers
edit2: also added couple things I forgot from the 7nm list
The answer I gave you is crystal clear and not a guess.
This is not really clear like I told you. Navi 10 has also 4 Rasterizer at 2 GHz but is outperformed by navi 21 with 80% and the driver is also clear with 8 scan converter instead of 4 in navi 10. also @CarstenS had strange readings. That’s why I want to know the values to find out what Frontend Navi21 has.....
If AMD's Radeon division employee gives you crystal clear answer, you need to either accept it or prove to him he is wrong with facts and not guesses
I'm saying that in case you didn't know who 3dcgi is.
Probably because these 4 rasterizers in Navi21 can rasterize 2x pixels per cycle, i.e. 32 pixels with color per rasterizer?
FP16 blending has been full speed relative to int8 for a long time on Radeons, so in order to blend 2x more FP16 pixels with Navi21 (which sinthetic tests seem to show) raster front end must have 2x throughput in comparison with Navi10.
32 rasterized pixels per Rasterizer look like overkill to me because average triangle size these days seem to be way below 32 pixels in 1080 and 1440p, so such wide rasterizers won't be utilized to full extend on triangles with less than 32 pixels area.
