AMD Radeon RX 7900 XTX RDNA3 Prototype Leaked, Confirms Reference Cooler Design

[phanbuey]

PCI-e slot should be able to provide 75W at least. But with 2x 8pin, I believe it is capable of delivering more than 300W. Assuming its sub 400W while delivering good rasterization performance, I think it will be pretty decent alternative. I don't think AMD will catch up with Ada in terms of RT performance, but the fact that Nvidia had to cook up some frame inserting tech may prove that performance gap may not be that big between the 2 camps.

I am hoping that this is the case! (so i don't have to get a new case)

 

[SL2]

It looks so slim lol, but yeah, it doesn't look great.

Give us a Radeon VII style shroud anodized in gun metal color.

 

[jesdals]

It looks so slim lol, but yeah, it doesn't look great.

Give us a Radeon VII style shroud anodized in gun metal color.

That cooler was terrible please dont - I still rember the heat out put from mine

 

[dj-electric]

Hey, AMD, can we please have red PCBs back with some designs?

Thanks.

 

[Xuper]

beautiful Fan.

 

[rv8000]

2 power connectors? So, the 400+ TBP claims are not true for the reference cards? Doesn't give me much hope for it competing with the 4090 even in rasterization then.


I don't think AMD will get anywhere close to that level of efficiency, but to be fair, benchmarks of 4090 with 300W power limit gets about 91% of the performance relative to "full" performance at 450W. If AMD has 10% efficiency advantage, they'd be able to match 4090 full-power with 300W.

However, AMD is slightly behind nvidia in terms of fab process this gen, and MCM will have at least a small performance penalty compared to monolithic, I simply don't see the winning on pure efficiency this gen.

If the specs are accurate, AMD overall will have a larger increase in traditional rasterization based hardware. Depending on the penalty for going MCM, I feel like there’s actually a very good chance AMD wins when it comes to rasterization performance. Considering how close the the 6900/6950XT came with a 256-bit bus, an improved infinity cache and 320/384-bit bus will be a nice thing to see.

Im actually excited to see what the 7000 series has to offer. Nvidias 4000 series has me snoring atm, a lot of odd choices despite how much performance the 4090 has to offer.

 

[claster17]

Other than that, too much plastic. I hate plastic.

It probably is all metal like the 6900XT already was.

 

[SL2]

That cooler was terrible please dont - I still rember the heat out put from mine

Well I said shroud so I wasn't talking about performance.

 

[Punkenjoy]

That cooler was terrible please dont - I still rember the heat out put from mine

On a thermodynamics stand point, if you felt the heat output from it, it meant it was good at extracting the heat.

But anyway, it isn't the rad that create the heat but the GPU. they seems pretty much the same except the cover but we can't really judge if we don't see the cold plate.

 

[Bomby569]

red PCB?!

 

[TheinsanegamerN]

The 7000 series is actually interesting. The 6950xt is 60% the relative performance of a 4090, with 5120 cores, at 4k. This new GPU has 12288. With a 384 bit bus and three times the infinity cache. Even if there was a 10-15% penalty for being MCM that would still put it's rasterization at 4k roughly equal to the 4090, and at 1440p significantly ahead.

It wouldnt be the first time AMD has dont this. The 5870 obliterated the 480 in performance and efficiency, while being significantly smaller, on the same TSMC process. The RTX 4000s are surprisingly efficient for the performance offered, so I doubt this will be fermi 2 GPU cooler boogaloo, but it's entirely possible that AMD may have made an ADA killer here.

If you are making the card 3 slots high, you should design it so it can accept a 120x25mm fan. Those wimpy 80x12mm fans barely do the job.
Scythe used to have this monstrous 5-heatpipe gpu block that was 2 slots tall with a slim 120x12 fan, but you could put a 25mm one there (making it 2.5 slot tall). It made the card silent and have significantly better temps than the crap they put even on modern cards. It was only 200W though, but it was also single fan...

Jeez, you trying to turn GPUs into skyscrapers? A GPU that tall is going to have a hard time fitting into cases that are not gigantic.

 

[TheDeeGee]

The PCIE 8-pin power is rated 150W for PCIE spec.

But the actual capability of those things can go up to 360W per connector if using HCS rated components.

Corsair pulls 300 Watt through them, with their 12VHPWR cable, as it has only two 8-Pin connectors on the PSU side.

Thats 20mm to long for Meshify C setup - damn

View attachment 267967

Btw @Wizzard could you include in your upcomming test if thermalpads on the back plate is needed like on the reference 6900/6950XT and how thick it need to bee.

Just get a Noctua NH-D15 or NH-U12A, problem solved.

 

[Nopa]

That cooler was terrible please dont - I still rember the heat out put from mine
View attachment 267968

Either of these with Rose-Gold color would be stylish as hell.

 

[jesdals]

Corsair pulls 300 Watt through them, with their 12VHPWR cable, as it has only two 8-Pin connectors on the PSU side.


Just get a Noctua NH-D15 or NH-U12A, problem solved.

Well I have a D15, but the Artic II 360 gives more headroom for my 5950x

 

[evernessince]

2 power connectors? So, the 400+ TBP claims are not true for the reference cards? Doesn't give me much hope for it competing with the 4090 even in rasterization then.


I don't think AMD will get anywhere close to that level of efficiency, but to be fair, benchmarks of 4090 with 300W power limit gets about 91% of the performance relative to "full" performance at 450W. If AMD has 10% efficiency advantage, they'd be able to match 4090 full-power with 300W.

However, AMD is slightly behind nvidia in terms of fab process this gen, and MCM will have at least a small performance penalty compared to monolithic, I simply don't see the winning on pure efficiency this gen.

Chiplets allow AMD to take the best performing dies and put them in a single product. It's 100% possible that like the 5950X AMD is binning the 7900 XTX (the model in this article) specifically to achieve the best performance per watt. The regular XT model might not have the same binning.

8-Pin is rated for 342 Watts

Yes but to follow the spec you can only draw 150w continuously max. I don't see the point in AMD breaking spec when they could just add a 3rd connector if they need it. The PCB appears to be full length.

 

[ARF]

The 7000 series is actually interesting. The 6950xt is 60% the relative performance of a 4090, with 5120 cores, at 4k. This new GPU has 12288. With a 384 bit bus and three times the infinity cache.

Not 3 times the Infinity Cache. Navi 21 has 128 MB, while Navi 31 is rumoured to have only 96 MB. That's actually a negative increase.

 

[Punkenjoy]

Not 3 times the Infinity Cache. Navi 21 has 128 MB, while Navi 31 is rumoured to have only 96 MB. That's actually a negative increase.

32 MB less infinity cache, but they changed it so a cache miss is less penalizing than with RDNA2. Also, the rumours is they also increased quite a bit the cache bandwidth. Overall those GPU will perform better at 4K since they will have much more bandwidth available.

We don't know yet about the L0, L1 and L2 cache too.

 

[TheinsanegamerN]

Not 3 times the Infinity Cache. Navi 21 has 128 MB, while Navi 31 is rumoured to have only 96 MB. That's actually a negative increase.

Is that confirmed? I've seen rumors of 192MB per MCM floating around the internet, usually connected to the 12288 core rumor. Even at 96MB, that's a total of 192MB of cache, more then the 128MB of navi 21, and that is coupled to a larger 320/384 bit interface with faster DDR6 VRAM.

 

[ARF]

We will understand on Thursday, hopefully.

AMD's main Navi 31 GPU reportedly has 96MB of Infinity Cache, Navi 32 is up to 7680 cores - VideoCardz.com

 

[medi01]

The 5870 obliterated the 480 in performance and efficiency, while being significantly smaller, on the same TSMC process.

There were some very VERY specific reasons for that (AMD got what was wrong with the process and took that into account when designing). It is unreasonable to expect that sort of advantage this time.

However, "chiplet" design might.

The 6950xt is 60% the relative performance of a 4090, with 5120 cores, at 4k. This new GPU has 12288.

I don't trust these numbers.
NV went 10752 => 16384 so, plus 52%.
AMD going 5120 => 12288 would be plus 140%

Most likely AMD simply got into NV marketing "business" of claiming twice the number of shaders they really have (since they could do fp+fp is the formal reason for this lie).

So in real numbers, AMD is going from 5120 to 6140 real cores, an increase of 20%.

 

[claster17]

I don't see the point in AMD breaking spec when they could just add a 3rd connector if they need it.

They have broken spec before and not just by a bit:

AMD Radeon R9 295X2 8 GB Review

Today, AMD is launching their Radeon R9 295X2, a dual-GPU card based on two fully unlocked, fully clocked Hawaii graphics processors. As a result, the card delivers impressive numbers in 4K and EyeFinity. But with a price of $1500, it is certainly not cheap, no matter how you look at it.

www.techpowerup.com

 

[ARF]

They have broken spec before and not just by a bit:

AMD Radeon R9 295X2 8 GB Review

Today, AMD is launching their Radeon R9 295X2, a dual-GPU card based on two fully unlocked, fully clocked Hawaii graphics processors. As a result, the card delivers impressive numbers in 4K and EyeFinity. But with a price of $1500, it is certainly not cheap, no matter how you look at it.

www.techpowerup.com

I trust the good old 8-pin connectors much more than the new and proved to be a fail design 12+4-pin hybrid 12VHPWR connector for PCIe 5.0.

The card requires two 8-pin PCI-Express power connectors. This configuration would normally be good for up to 375 W of power draw. AMD, however, has chosen to trump those specifications and cites a minimum of 28A (=336 W) for each connector, which brings the total to around 750 W. I think AMD exaggerates a bit, but you had better be safe than sorry. In actual usage we see power draw peak at 650 W, which makes 25A per connector enough. Nevertheless, this configuration greatly exceeds the PCI-Express specifications, so you should not use a weak or poorly built PSU with this card.

AMD Radeon R9 295X2 8 GB Review - A Closer Look | TechPowerUp

 

[evernessince]

They have broken spec before and not just by a bit:

AMD Radeon R9 295X2 8 GB Review

Today, AMD is launching their Radeon R9 295X2, a dual-GPU card based on two fully unlocked, fully clocked Hawaii graphics processors. As a result, the card delivers impressive numbers in 4K and EyeFinity. But with a price of $1500, it is certainly not cheap, no matter how you look at it.

www.techpowerup.com

Buildzoid pointed out that example as well when doing his review of Nvidia's new cable. That's more the exception than the rule.

 

[cvaldes]

red PCB?!

A traditional color for prototypes. Nothing new about this, the electronics industry has been doing this for decades.

 

[Chane]

Only two and a half slots, nice.