AMD Radeon RDNA2 Navi (RX 6500, 6600, 6700, 6800, 6900 XT)

3060 has more TF but it targets practically the same market. It's not a matter of teraflops only, it is a matter of performance bracket with a due transistor budget and power target. Quite probably AMD's profiling team established this was the best compromise among these factors. Also, computing needs are still growing as games evolve, so 10TF may seem "too much" now but not in a couple of years.
 
I really want to see better low-end, but I am afraid the lure of margins and fear of second hand flood might prevent it at scale.
 
It is pretty damned weird, especially being so utterly close to the Navi 22 die. The only possible target market I can imagine for this is very highly targeting gaming laptops. With that many CUs you can cut clockspeed drastically, think to like 1.1/1.2ghz while still getting performance roughly that or above like, an rx580/GTX 1060 in no doubt quite a low power envelope. Thus making a potentially incredibly power efficient GPU. That a lot of gaming laptops target 1080p already fits nicely.

I think few N23 parts will have all the 32CUs clocked at 2.5GHz, as the GPU isn't really designed to be balanced for running games at that compute throughput (too much compute for bandwidth).
The 50-100W part called RX 6600M enables 28 CUs and runs at 2.1GHz or less. It's mostly a ~6.5 - 7.8 TFLOPs part.

I'm guessing the fully enabled N23 @ 2.5GHz going into the Teslas' infotainment is pushing well above 100W, and so will any desktop part doing so. But at those settings the gaming performance is probably not going to scale linearly with power consumption and compute throughput.
 

It would be the wiser of the strategies.

The surplus of GPU's not being sold to miners (due to Ethereum going PoS in December) will easily satisfy the GPU gaming market share AMD aimed at (~20%). It does not imply Radeon is weakening or going out of business.

There isn't a win/win choice when the manufacturing capacity is limited. You have to chose your battles quarter by quarter.
 
The wiser of the strategies would be to use another crypto hangover to capture more GPU market share than they had for the last several years.
That's not money.
Now Milan?
Effing cash galore.

It's a CPU house that makes excellent GPU IP, not the other way around.
 
The wiser of the strategies would be to use another crypto hangover to capture more GPU market share than they had for the last several years.

That is not how a crypto hangover works.

In addition to the surplus of GPUS not being sold to miners that are now accessible to gamers, there will be an enormous wave of second hand GPUs. There isn't a viable strategy to win market share in these conditions unless you want to sell new Silicon with zero profit and still not match the used market prices.

Hence, switching wafers to the datacenter is the wiser of the choices.
 
The surplus of GPU's not being sold to miners (due to Ethereum going PoS in December) will easily satisfy the GPU gaming market share AMD aimed at (~20%). It does not imply Radeon is weakening or going out of business.

There isn't a win/win choice when the manufacturing capacity is limited. You have to chose your battles quarter by quarter.
Yes, they should all brace for the flood of 2nd hand GPUs that are coming into the market.
 
That's not money.
Now Milan?
Effing cash galore.

It's a CPU house that makes excellent GPU IP, not the other way around.
I fully understand the reasoning.
But that's where the previous point on AMD being way more limited than Nv due to using TSMC comes into play.

That is not how a crypto hangover works.

In addition to the surplus of GPUS not being sold to miners that are now accessible to gamers, there will be an enormous wave of second hand GPUs. There isn't a viable strategy to win market share in these conditions unless you want to sell new Silicon with zero profit and still not match the used market prices.

Hence, switching wafers to the datacenter is the wiser of the choices.
It's exactly how crypto hangover works.
The better perf/price products will capture more market share during a surplus of supply.
A higher production output would allow to lower costs of production resulting in an option of offering lower prices and capturing the market - as a first or second hand sales (used market prices are controlled by the retail market prices at their max).
So you could use that to capture the market - but you'd be loosing margins of course.
So if the strategy is to get higher margins and not more market share then yes it's a "wiser" strategy.
 
AMD has spent about 15 years proving that going after market share by "lowering prices" doesn't work.

For pretty much all of those 15 years: "Please AMD, save us from the ridunkulous prices NVidia is charging, we want NVidia to lower its prices so that we can buy more NVidia stuff".
 
But that's where the previous point on AMD being way more limited than Nv due to using TSMC comes into play.
No, even with infinite supply it would still be heavily weighted against GPUs.
Because we need Milan.
It we could get ALL the Milan we'd get ALL the Milan.
AMD has spent about 15 years proving that going after market share by "lowering prices" doesn't work.
Bingo.
There's a very, very good reason why N31 is a giant oversized overly expensive fuckoff stick and not what you'd expect Radeon GPU to usually be.
 
It's exactly how crypto hangover works.

lol. No. You have a problem for sure.

There is a wave of product coming into the market whose prices are out of the manufacturer hands. You cannot, never, control a market in such conditions.

That is crypto for you, and it happened twice before.
 
AMD already got burned by the strategy proposed here, they flooded the market during the first crypto craze (well, second, if you count the 2010-2011, but it was an amateur era) with gazillion Tahiti GPUs which never sold well because miners sold their used cards even cheaper. I distinctly remember the time when you could buy 7950 GE at $200 or so, so people did not buy Hawaii as it costed twice more and wasn't initially mindblowingly better. And the zombified adherents still bought 780 / 780ti (some people even bragged how they got their 780 for only $800 or so) because they needed PhysX / Gameworks etc.
 
There's a very, very good reason why N31 is a giant oversized overly expensive fuckoff stick and not what you'd expect Radeon GPU to usually be.

And it doesn't matter if it's giant oversized and overly expensive. As long as it gets the longest bar in the reviews then lots of people just want to get a graphics card of the fastest brand.
 
Built for mobile, yes.

Bottom two parts are mobile first yes.

AMD likes their fast just too much.
oh well, for the kind of games I like the most -strategy, HoMM like, some fps, a bit of racing, and a bit of everything really, but mostly the first two- it could be more than enough, but yeah, I wonder how are you going to put 10teraflops to good use.ç

The 6600/6600XT is allegedly in the same ballpark of the 5700XT which is a very good 1440p performer, the resolution I want to play games at.

It only occurs to me that you need 10 teraflops for RT.

A colleague of mine has the 6700XT and at 1440p a game which runs at 90 fps without RT, drops to 35fps with RT on.

We shall see, but the most efficient it is the better -at least for me- and with the help of FSR it could be an amazing bang for the buck GPU. Just imagine running games at 240fps/1080p on it...
 
Back
Top