GhostofWar
Regular
Sigh, waiting on nvidia prices mean it's the same song and pony dance again. If they were going to aggressively price these parts they wouldn't need nvidias prices to slot behind.
AMD RDNA4 Architecture Speculation
- Thread starter Frenetic Pony
- Start date
Bondrewd
Veteran
what did I miss
Nvidia. also praying to whatever deity you fancy for AMD to revive C thingies.
a-ha. 9950X3D is the leadership product. Different rules, different ballgame.
Waiting for RDNA5.
Upgrading high-end GPU on every generation makes very little sense.
For consumers/gamers, there are two sane strategies, to keep the computer at relatively good performance level
1) buying high-end components (or totally new high-end computer) rarely
2) buying cheaper components more often
But for those who have huge amount of extra money and want to buy a super-high-end gaming machine during 2025 or beginning of 2026, nvidia is the only option. But this is a very small market.
For professional use where all extra power gives productivity increase the situation is of course different.
Correction. Make that $400.
DegustatoR
Legend
We don't know the comparative performance yet.
9070 will probably be below 5070 but not by much.
9070 XT will probably be at about 5070.
Considering that there is feature parity now and they have +4GB of VRAM $400 seems a bit too low.
I'd say $400 for 9070 and $500 for XT are best bets.
This of course could mean that their margins will suffer if their plan was to price these higher.
That's how I'm interpreting this as well. If they were confident in having decided on some 'aggressive' pricing, they wouldn't have held off. They probably got word of Nvidia's 50 series pricing, and then realized they were not nearly as well positioned as they thought they'd be.
It's telling that in their interviews they did after the event, it really sounded like their talk of the importance of good pricing and learning lessons and all that was all in relation to their competition. That's probably obvious from a business standpoint, but from a consumer standpoint, that's still very much a "We're not really trying to provide great standalone value, only make it seem like good value when compared against our super high priced competition" giveaway here, and goes against what consumers are thinking when we talk about aggressive pricing.
EVGA complained they've been getting prices from NVIDIA practically when Jensen is on stage already, what makes you think AMD gets the prices early?
We're only talking hours here. That's still very much what I'd consider 'last minute'.
Potato Head
Newcomer
You guys see the latest from CES?
RX 9070’s with triple slot coolers and either 2x8 power or 12V-2X6. 375W.
I fear this is going to be Vega 2.0.
RX 9070’s with triple slot coolers and either 2x8 power or 12V-2X6. 375W.
I fear this is going to be Vega 2.0.
Who is stating 375w? 2x8pin is pretty standard for mid range these days, that by itself doesn't mean high power use at all.
Regarding pricing. Since 7800XT was $499 RRP I don't see why a next gen 70 class shouldn't be cheaper than that. If they were planning another $499 RRP I don't know what they were thinking.
Potato Head
Newcomer
Gigabyte AORUS - 3x8
These things are all bigger than 5090’s.
These things are all bigger than 5090’s.
Potato Head
Newcomer
PowerColor…
It has been always like that. Market leader (especially with market share of near 90%) defines the features that game studios will implement just due the huge user base. If underdog comes up with something different and new, it has to be in scale of R300 vs nv25 to get it pushed thru and that is huge risk which probably no one is willing to take any more.
if nVidia loses market share, even if it would be in scale of 10-15 %, has really almost no effect at all the market situation. AMD again can't afford losing much more, so Intel's actions are 10 times bigger threat to AMD GPU division than nVidia just because the market situation is what it is.
https://videocardz.com/newz/amd-nav...rx-9070-pictured-may-exceed-nvidia-ad103-size
~390mm2 ? Quite chunky
~390mm2 ? Quite chunky
Reference is 2x8-pin.
To quote TPU, "The card calls for three 8-pin PCIe power connectors. We've only seen one other custom RX 9070 XT come with three connectors, and that is the XFX RX 9070 XT Merc 319 Black. ... Most other cards, including the PowerColor Red Devil, come with just two 8-pin connectors (375 W)"
Yes, there's cards (more than those 2 they had seen at that point) with 3x, but there's nothing new in AIBs putting extra power connectors, heck, there's even one with 12V-2x6, but it doesn't mean the card is drawing 600W
That would be strange.https://videocardz.com/newz/amd-nav...rx-9070-pictured-may-exceed-nvidia-ad103-size
~390mm2 ? Quite chunky
Navi 32 with 60CU's had total die area of 346mm², and that obviously included four 6nm MCD's totaling about 146mm².
Navi 48 is supposed to be 64CU's, but there should be die savings from moving to 4nm and putting the memory bus and L3 IC on the main 4nm die as well(scaling here isn't great, but it's still something).
Perhaps 64CU is wrong and they actually upped CU count a fair bit instead of going for narrow+faster clockspeeds? Or maybe RDNA4 CU's are a really sizeable chunk wider to accommodate better RT/AI? I dont know, but that doesn't strike me as the sort of architectural or area efficiency gains they'd have liked, especially if the aim is to be able to price this GPU more aggressively.
DegustatoR
Legend
Probably this. Less about RT and more about adding MM ALUs.
Still would lead to a die size similar to GB203, with a similar power draw.