Complete Specifications of AMD Radeon RX 9070 and RX 9070 XT Leaked

[3x0]

Is it this one?

Yup, that's the one. I mean, Maxwell had less cores than Kepler and smoked it in performance

 

[Cheeseball]

Yup, that's the one. I mean, Maxwell had less cores than Kepler and smoked it in performance

Understandable. Similar situation with what happened with the RTX 3060 vs the RTX 4060. Specs-wise the RTX 3060 is superior but is still around 10% weaker (except for the big L2 cache and newer architecture improvements). It only gets stronger at higher resolution due to the memory bus and even then its not a big lead.

 

[TumbleGeorge]

Yup, that's the one. I mean, Maxwell had less cores than Kepler and smoked it in performance

Where I read that is raster? Frame generation afmf I think is in results.

 

[3x0]

Where I read that is raster? Frame generation afmf I think is in results.

No, it's raster and raytracing. You're thinking of multi frame generation when Jensen compared 5070 to 4090

 

[TheinsanegamerN]

Yup, that's the one. I mean, Maxwell had less cores than Kepler and smoked it in performance

Maxwell smoked Kepler in clock speed, that's where its performance came from. And they OCed like a champ.

The 9070xt has a roughly 30% higher clock speed and 25% fewer cores then a 7900xt. So short of ground shattering IPC improvements, you're not catching a 7900xtx.

Something else not considered: rDNA4 is a denser design. Denser designs tend to run hotter. That clock speed gulf has no guarantee of staying then under intense 3d workloads, or with RT.

It's highly unlikely a design with 96 ROPs is going to consistently beat a design with 192 ROPs, especially at higher resolution.

 

[Bobaganoosh]

If it's actually available, they can price it at $699 and still sell tons of them because there's literally no competing cards in stock. All they have to do to sell cards is make them right now and they know it, so they'll price as high as possible.

 

[Scattergrunt]

If it's actually available, they can price it at $699 and still sell tons of them because there's literally no competing cards in stock. All they have to do to sell cards is make them right now and they know it, so they'll price as high as possible.

Because of luck. Doesn't mean its actually a good thing though; it just means that the next generation when AMD comes time to sell new cards; they're gonna repeat this, and it probably wouldn't work again, so we end up back at square one.
Its pretty evident AMD learned nothing from the RX 7000 series launch & reception from the general public.. or lack there of, for the latter one.

AMD needs marketshare in dGPU's, not another lack luster launch that only tech enthusiasts will actually care about. Sure, I'm not mad that they're pricing it at $699 or $649 (seen both numbers thrown around), but it doesnt mean its a good thing. AMD can clearly cut the prices more too, if the prices of RX 7000 cards dropping drastically after launch says anything, they just choose not too. I get that they dont have as good of margins as NVIDIA, but come on. If you can't price it well have other reasons people wanna buy it.

I know I said I'd reserve my opinion until AMD actually showcased the cards & had numbers with them but its just like; holy shit, everyone called it. Again. ffs..

HOPING To be proven wrong but also not gonna sit here and cheerleader them when I know they're gonna probably just going to further mirror that one famous quote about "insanity is doing hte same thing over and over again expecting something to change."

 

[Yashyyyk]

I don't even care about price/performance, I want ITX card (but not 3x price like workstation card lol)

No dual fan model since AMD cancelled their reference (and idk if that was even dual slot)

When ITX GPU?

I want an ITX / low length/single fan RTX 4070 with 16GB VRAM. (Or rather something like the 1080Ti mini for the 4070Ti (Super/16GB VRAM)) It's ~200W just like the numerous ITX 1080/2070(S)/3060Ti, all ~1 "Liter" in size, so ~200 "Watt per Liter" We should be getting better/more...

www.techpowerup.com

 

[Jtuck9]

- Gonna give yourself a bad case of the horse lung...

Its been internalized, for sure.

Completely off topic, your username reminded me of these

Judith Butler: To Imagine a World After This, Democracy Needs the Humanities

Many young people tell me that they fear there is no future. When they ask about the future, they are also asking: what is still imaginable or for what may we still hope? To say there is no future,…

lithub.com

Mechthild of Magdeburg: God’s Power Is Love

Scholar Wendy Farley introduces the Rhineland mystic Mechthild of Magdeburg: Little is known about Mechthild , though her name indicates that

cac.org

 

[GodisanAtheist]

Its been internalized, for sure.

Completely off topic, your username reminded me of these

Judith Butler: To Imagine a World After This, Democracy Needs the Humanities

Many young people tell me that they fear there is no future. When they ask about the future, they are also asking: what is still imaginable or for what may we still hope? To say there is no future,…

lithub.com

Mechthild of Magdeburg: God’s Power Is Love

Scholar Wendy Farley introduces the Rhineland mystic Mechthild of Magdeburg: Little is known about Mechthild , though her name indicates that

cac.org

-Facinating reading, was esp. drawn in by Mechthild of Magdeburg, non-linear thinking on the nature of power...

 

[Jtuck9]

-Facinating reading, was esp. drawn in by Mechthild of Magdeburg, non-linear thinking on the nature of power...

Purnam!

 

[3x0]

Maxwell smoked Kepler in clock speed, that's where its performance came from. And they OCed like a champ.

The GTX 980 had a 20% higher clock on average and 33% more ROPs, while the GTX 780Ti had 40% more cores, 87.5% more TMUs and 50% more memory bandwidth. It's not as simple as clock speed

 

[Tomorrow]

Too bad these new Radeon parts are UHBR13.5 instead of UHBR20 like the RTX 50 series.

...or like their own RDNA3 based Pro cards were UHBR20 years ago. Very disappointing.
Yet they add useless PCIe 5.0. Im guessing to cut the bus on some models to x8 or even x4 link speeds to save literal pennies.

 

[Ultron1337]

128 ai accelerators. Why?

This is unchanged from RDNA3, that introduced 2 AI accelerators per CU
From AMD home page: AMD RDNA™ 3 delivers over 2X more AI performance than AMD RDNA™ 2 architecture. Featuring 2 AI Accelerators per compute unit – Radeon RX 7000 Series Graphics cards deliver up to 123 TFLOPs of FP16 performance

 

[JohH]

It's highly unlikely a design with 96 ROPs is going to consistently beat a design with 192 ROPs, especially at higher resolution.

It doesn't have 96 ROPs. Hope that helps explain things.

 

[Arkz]

Too bad these new Radeon parts are UHBR13.5 instead of UHBR20 like the RTX 50 series.

Does it matter? DSC is indistinguishable from uncompressed, and with DSC UHBR13.5 could do 4k240 or 8k120, as if anyone's actually gonna use it for that.

 

[Tomorrow]

Does it matter? DSC is indistinguishable from uncompressed, and with DSC UHBR13.5 could do 4k240 or 8k120, as if anyone's actually gonna use it for that.

That's not the point.

1. These cards will most likely cost more than 700. Cost of UHBR20 implementation is not an issue.
2. Feature parity. Competitor has UHBR20 support. They strive for feature parity in everything else. In the past they were actually in the lead as 40 series did not support any 2.1 speed at all.
3. RDNA3. Pro cards in that series had full UHBR20 support two years ago. It was reasonable to expect AMD to move from UHBR13.5 on RDNA3 consumer and UHBR20 on RDNA3 prosumer cards to unified UHBR20 this time, considering that RDNA3 was two years ago.
4. Monitors. More and more monitors are being released with UHBR20 support and the cable length issue has been resolved. UHBR20 is no longer some niche feature a few professionals use.