AMD Radeon RX 6900 XT "Big Navi 21 XT/XL" Ostensibly Confirmed to Run @ 2.2-2.4GHz Clock Speeds + 255W (TGP) & 16GB GDDR6 (Updated)

[BiG StroOnZ]

We are just 9 days away from AMD's Radeon RX 6000 "Big Navi" reveal and AMD appears to have something potentially groundbreaking up its sleeve; to not only challenge the GeForce RTX 3080, but also possibly even the mighty GeForce RTX 3090. Patrick Schur, who has been fairly reliable with past leaks, is suggesting that the Navi 21 XT (a la Radeon RX 6900 XT) will feature a staggering game clock of 2.4GHz. Details of the Navi 21 XT and XL GPUs have leaked too; these details include base, game, and boost clocks of both GPUs, plus the TGP (total graphics power) and memory capacity of the Navi 21 XT.

 

Quote

 

 

 

For perspective, the Radeon RX 5700 XT (RDNA) has a base clock of 1.6GHz, game clock of 1.75GHz and a boost frequency of 1.9GHz. At 2.4GHz, that is the highest clock speed that we've seen for any consumer graphics cards stock from the factory. And considering that just the game clock is mentioned, we could be looking at even higher speeds for the maximum boost clock with this new RDNA 2 architecture. Based on previous rumors, the Radeon RX 6900 XT has a total of 80 CUs and 5120 stream processors. With a game clock of 2.4GHz, we're looking at around 24.5 TFLOPs compute compared to 9.754 TFLOPs for the Radeon RX 5700 XT. In other words, this is going to be a performance monster.

 

Apparently, the Navi 21 XL, which is expected to be the lesser version of the Navi 21 GPU, will come with a base clock ranging from 1350MHz and 1400MHz, a game clock between 1800MHz and 1900MHz, and capable of boosting up to 2100/2200MHz. On the other hand, the Navi 21 XT is higher clocked, featuring a base clock of 1450-1500MHz, a game clock ranging from 2000MHz to 2100MHz, and a boost clock of up to 2200-2400MHz. The reference cards should be closer to the bottom of the clock range, while AIB cards are expected to be closer to the top of it. This also suggests that we can expect even higher Turbo frequencies. These days, the thing with AMD cards is that the turbo boost and actual gaming frequency are two different things. So we're not quite sure what to make of that.

 

The GPU is also said to feature a TGP (total graphics power) of 255W. This does not mean all the power of the card, but the GPU itself. The GeForce RTX 3080 has a TDP of 320W, and the GeForce RTX 3070 has a TDP of 220W. Navi 21 XT, which is built on an enhanced 7nm process node, is reported to have a TDP (total design power) of 225 watts compared to a TDP (total design power) of 320 watts for the GeForce RTX 3080. The GeForce RTX 3090 has a TDP of 350 watts. Finally, Schur is alleging that the Radeon RX 5900 XT is packing 16GB of GDDR6 memory compared to 10GB GDDR6X for the GeForce RTX 3080 and 24GB GDDR6X for the GeForce RTX 3090.

 

Source 1: https://www.eteknix.com/amd-navi-21-xt-graphics-card-specs-leak/

Source 2: https://hothardware.com/news/amd-radeon-rx-6900-xt-big-navi-24ghz-16gb-of-ram

Source 3: https://www.techpowerup.com/273490/amd-navi-21-xt-seemingly-confirmed-to-run-at-2-3-2-4-ghz-clock-250-w

Source 4: https://www.guru3d.com/news-story/rumor-amds-navi-21-gpu-has-255-watts-tgp-and-boost-up-to-2-4-ghz.html 

Source 5: https://videocardz.com/newz/amd-navi-21-xt-to-feature-2-3-2-4-ghz-game-clock-250w-tgp-and-16-gb-gddr6-memory

Source 6: https://www.kitguru.net/components/graphic-cards/joao-silva/amd-navi-21-xt-xl-leak-suggests-an-over-2-0ghz-boost-clock/

 

 

For now we will have to wait until 10/28/2020 to really see what AMD has in store for us. However, from what has been seen thus far, is that AMD is looking to shake up the dGPU market with its Radeon RX 6000 family of GPUs. Additionally, besides the Navi 21 XT and XL, rumors also say that there’s an XTX GPU with more CUs and conceivably different clock speeds. Regardless, in the meantime; take all of these leaks/rumors with a grain of salt, as currently, there is nothing really to back up these specs. 

 

Smallish update:

 

A custom AMD Radeon RX 6800XT board partner card allegedly features a 2577MHz boost clock

 

Quote

 

 

 

 

 

According to Igor’sLAB, a ‘special board partner’ card based on Navi 21 features a boost clock of 2577 MHz. Before we all get too excited, Igor did not mention which SKU this is (XTX, XT or XL). He did say that this is a boost clock, meaning this is not the ‘actual’ clock speed of the graphics card. This Radeon RX 6000 model will likely stay at 2.3-2.4 clocks, where AMD’s ‘gaming clock’ is more representative of what may be seen during gameplay. The data came from a BIOS that Igor got access to. 

 

Source 7: https://videocardz.com/newz/amd-radeon-rx-6800xt-board-partner-card-allegedly-features-a-2577-mhz-boost-clock

[Moonzy]

here's to hoping they don't screw up as bad as nvidia did this time around

hope they are more conservative with clock speeds, so people wont crash on stock

and OC headroom is always fun

 

might finally be a uno reverse situation.

[Bad5ector]

I'm eyeballing my wife's FreeSync / G-Sync compatible monitor and thinking I might just go team red this time around... she can have my Gsync certified monitor in trade  

[TetraSky]

The wait for the announcement is getting unbearable... For the release date and the price... And for actual stocks to force used prices down.

[DildorTheDecent]

Hope those performance graphs vs the RTX 3080 were from one of the shit tier big navi cards.

 

Hope that big boi starts pulling bus lengths the faster it's clocked.

 

AMD pls. 7nm dream again pls.

[Bombastinator]

22 minutes ago, TetraSky said:

The wait for the announcement is getting unbearable... For the release date and the price... And for actual stocks to force used prices down.

I’m worried used cards may have serious problems depending on how this whole storage thing works out.  Numbers are great, but what about the data latency thing?

[dizmo]

It'd be interesting if this is the case, but a lot of the leaks for the CPUs were wrong and I don't see this being any different. I wonder how long the cards will actually be able to keep the game clocks up for. 

1 hour ago, Bombastinator said:

I’m worried used cards may have serious problems depending on how this whole storage thing works out.  Numbers are great, but what about the data latency thing?

If you're referring to stuff like RTX I/O, we're years away from seeing that in PC games. 

[Loote]

May I ask what is the date when independent reviews of them can be published? Is it before or after they get into shops?

I've been telling my friend wait wait and wait for the past two months and after RTX3000 premiere he's thinking of going for Navi cards on the premiere day, whereas I strongly hope they will have much better availability and Nvidia is going to put out much more of their cards too.

[Random_Person1234]

4 minutes ago, Loote said:

May I ask what is the date when independent reviews of them can be published? Is it before or after they get into shops?

I've been telling my friend wait wait and wait for the past two months and after RTX3000 premiere he's thinking of going for Navi cards on the premiere day, whereas I strongly hope they will have much better availability and Nvidia is going to put out much more of their cards too.

Usually on the day they get into shops.

[Bombastinator]

21 minutes ago, dizmo said:

It'd be interesting if this is the case, but a lot of the leaks for the CPUs were wrong and I don't see this being any different. I wonder how long the cards will actually be able to keep the game clocks up for. 

If you're referring to stuff like RTX I/O, we're years away from seeing that in PC games. 

Not ray tracing.  They’ve got to do I suspect a minimum of two more gpu iterations for that.  I mean the high speed storage thing.  There was an article about a wolfemstein thing going for it.

[Loote]

14 minutes ago, Bombastinator said:

Not ray tracing.  They’ve got to do I suspect a minimum of two more gpu iterations for that.  I mean the high speed storage thing.  There was an article about a wolfemstein thing going for it.

RTX I/O mentioned by dizmo is Nvidia's implementation of DirectStorage

[S w a t s o n]

5ghz confirmed

[Bombastinator]

8 minutes ago, Loote said:

RTX I/O mentioned by dizmo is Nvidia's implementation of DirectStorage

Ah. So nothing to do with ray tracing.  they just stuck the branding on it. 

[CarlBar]

The memory i'd belie. But either that top clock speed or the TDP is nonsense. That kind of clock-speed increase from the previous generation requires that the TDP per CU goes up and there's more CU's than the 5700XT, yet it supposedly has the same TDP on a node thats only slightly more efficient. Thats not happening. This is complete and utter nonsense.

 

The only way i can make sense of this is if the clock is for when the CU's are running in RT mode, depending on how thats done it might well be that it consumes less power per CU at higher clocks compared to when running in raster mode. But those clocks at that TDP in raster mode are just flat out impossible IMO, i don't think they could get that TDP even if they had TSMC 5nm to work with.

[GDRRiley]

3 minutes ago, CarlBar said:

The memory i'd belie. But either that top clock speed or the TDP is nonsense. That kind of clock-speed increase from the previous generation requires that the TDP per CU goes up and there's more CU's than the 5700XT, yet it supposedly has the same TDP on a node thats only slightly more efficient. Thats not happening. This is complete and utter nonsense.

and your point? this could be on a 7nm+ node

[WereCat]

7 minutes ago, CarlBar said:

The memory i'd belie. But either that top clock speed or the TDP is nonsense. That kind of clock-speed increase from the previous generation requires that the TDP per CU goes up and there's more CU's than the 5700XT, yet it supposedly has the same TDP on a node thats only slightly more efficient. Thats not happening. This is complete and utter nonsense.

 

The only way i can make sense of this is if the clock is for when the CU's are running in RT mode, depending on how thats done it might well be that it consumes less power per CU at higher clocks compared to when running in raster mode. But those clocks at that TDP in raster mode are just flat out impossible IMO, i don't think they could get that TDP even if they had TSMC 5nm to work with.

I'm assuming that TGP stands for Total Graphics Power so 255W is just for the GPU die. If you add in the GDDR6 we are looking at 300W+ card. Seems fine.

[SpaceGhostC2C]

Big Navi is getting so hyped that you might as well put it alongside your Ampere card, pair them with your Zen N PCIe 4.0 DDR5 CPU and play Star Citizen with it. 

 

 

 

 

Spoiler

(don't worry about cross-vendor mGPU support, I'm sure it will be added to the roadmap through feature creep) 

 

 

[Loote]

Is TGP for any clock, or the base clock?

Here they say that XSX peaked at 211W https://www.eurogamer.net/articles/digitalfoundry-2020-xbox-series-x-power-consumption-and-heat-analysis

Due to binning and being a newer design, I can believe that big navi consumes at 2GHz what Xbox does at 1825MHz, if you then remove power supply efficiency and other elements, and then extrapolate from 52 CUs to 80*, you could fit in 250W TGP, now as for the boost... It's hard to imagine, but not impossible.

 

* I just assumed that from those 211W 150 were used by GPU, including memory, 61W for CPU, efficiency, storage etc. seems like going low even, then if you multiply by 1.5 it's 225W for 78CUs

 

I think someone could come up with those numbers based on what we know so far, I am not advocating for this leak being true, just wondering if we can write it off so easily.

[porina]

https://www.igorslab.de/en/amd-radeon-rx-6000-the-actual-power-consumption-of-navi21xt-and-navi21xl-the-memory-and-the-availability-of-board-partner-card-exclusive/

 

Haven't seen above posted yet.

 

tldr:

Difference between AMD chip level power consumption and nvidia board level, if you account for this, they're same ball park

Sapphire, Asus, limited cards mid-November. Other companies, end November.

PCB design similar to old navi, implies 8x 2GB chips, not 16x 1GB, don't need major board redesign.

[VanayadGaming]

I've seen this trend a lot lately that even if it is as good as the 3080, people will choose that one over the amd card. Even though the 3080 consumes ~70-100w more power. Last time this was the other way around, people were screaming that their electricity bills will be insane, that they will die of heat exposure and basically the apocalypse will come. 

[BlackManINC]

Meh, its all rumors until confirmed by AMD. I'm tired of rumors, its time we start getting some real news. 

[porina]

1 hour ago, VanayadGaming said:

Even though the 3080 consumes ~70-100w more power.

It wont, that misunderstanding comes from differences in how AMD and nvidia state power. See my previous link for a better description.

[Doobeedoo]

So far on it's looking very good for Big Navi though. How much they managed to improve architecture wise is yet to be seen. Well soon enough.

[LAwLz]

Quote

CONFIRMED

No it isn't confirmed. It's a "leak" that might as well just be a rumor someone came up with.

Don't get suckered on to the hype-train. Just sit back, relax and wait for third party reviews.

 

The best thing to do is expect it to be shit because then you will have the smug feeling of being right if it turns out to be shit, or be pleasantly surprised if it turns out that it's good.

If you expect it to be great then you have more to lose. Hell, even if it turns out to be decent you might get disappointed if you have set your expectations too high.

[Morgan Everett]

Without indications of price and availability, this isn't of much interest.