Monday, July 17th 2023
AMD Planning September Launch for Radeon RX 7800 series and RX 7700 series
AMD is planning to plug the Atlantic gap between its mainstream Radeon RX 7600 and enthusiast-class RX 7900 XT with the RX 7800 series and RX 7700 series, with either an announcement or teaser planned for 2023 Gamescom, which is scheduled for August. There could be up to four new graphics card SKUs announced, with their product launches spread across Q3 and Q4 2023. The "Navi 32" MCM is expected to power at least three of these SKUs, while it was recently rumored that AMD could design a new GPU that has the GCD of the "Navi 31" on the package of "Navi 32" with its four MCDs, to end up with a higher CU count than what the "Navi 32" can offer.
The "Navi 32" GPU is an MCM, just like the "Navi 31" powering the RX 7900 series. It is rumored to feature a 5 nm GCD (graphics compute die) with 60 RDNA3 compute units, which work out to 3,840 stream processors, 120 AI accelerators, 60 Ray Accelerators, 240 TMUs, and possibly 128 ROPs. The four 6 nm MCDs give it 64 MB of Infinity Cache, and a 256-bit wide GDDR6 memory interface. Assuming the RX 7800 XT uses the unnamed new MCM with the GCD of the "Navi 31" that has a CU count somewhere between 60 and 72, a maxed-out "Navi 32" could power the RX 7800, while its cut-down variants power the RX 7700 XT and RX 7700.
Sources:
Moore's Law is Dead (YouTube), VideoCardz
The "Navi 32" GPU is an MCM, just like the "Navi 31" powering the RX 7900 series. It is rumored to feature a 5 nm GCD (graphics compute die) with 60 RDNA3 compute units, which work out to 3,840 stream processors, 120 AI accelerators, 60 Ray Accelerators, 240 TMUs, and possibly 128 ROPs. The four 6 nm MCDs give it 64 MB of Infinity Cache, and a 256-bit wide GDDR6 memory interface. Assuming the RX 7800 XT uses the unnamed new MCM with the GCD of the "Navi 31" that has a CU count somewhere between 60 and 72, a maxed-out "Navi 32" could power the RX 7800, while its cut-down variants power the RX 7700 XT and RX 7700.
59 Comments on AMD Planning September Launch for Radeon RX 7800 series and RX 7700 series
Board manufacturers are happy to make gigantic cards with higher TDPs and higher clocks, but the original plan for Navi31 was 3GHz at 300-350W, and that never happened. That doesn't make Navi31 a bad product, or bad silicon, only that AMD didn't reach their original goals due to unforseen design issues. Sam talked about delaying Navi32 to apply the lessons they'd learned from Navi31. I'm simply wondering whether they actually took that path or not. Talking about it in an interview doesn't make it a reality and AMD have been pretty quiet about it publicly since then.
If Navi32 is respun silicon and it's more efficient and faster, you can guarantee they'll charge more for it.
If Navi32 is the original silicon design taped-out shortly after Navi31 launched, then it'll be a bit slower and hotter, but cheaper to buy.
I don't really see a downside with either outcome; Faster/more efficient silicon would be nice, but MCD is never going to be as power-efficient as monolithic so it's unlikely to match the Nvidia 40-series either way, rending it a moot point not worth discussion.
Id check personally but PC is in pieces until noctua offset mount comes in tonight; nitro typically hovers around low 2900s stock and I’ve got it FPS capped to 240.Where do they say the limit is “2500mhz”. I’ll wait.
Other clock speeds and TDPs are available for any SKU, whether that's Nvidia, AMD or Intel. Whether that's a partner model with beefed-up cooling and power delivery, or someone pushing a manual overclock/undervolt on a reference card - but that's irrelevant.
When comparing GPUs - especially speculating against pre-launch models, the reference clocks are the defining standard. It's why most reviews (even those that test overclocked performance) always show reference clocked results too. That's what you're paying for, and that's what you're guaranteed to get.
If someone takes a 1001 horsepower Buggatti Veyron and tunes it to 1500 horsepower, that doesn't meant that all Veyrons now make 1500 horsepower. The original Veyrons are all still limited to 1001hp until they, too, are individually modified. Such is the way for "stock" GPUs; They run at stock until modified. The default TDP and clock limits in the driver and BIOS of the original 7900XTX doesn't allow it get anywhere near 3GHz, and whilst the 2500MHz reference clock is likely a lower estimate of what typical cards will opportunistically boost to, it cannot possibly be a shocking new revelation to you.
Take an AMD/Intel CPU for example. They are sold as "up to 5 GHz boost clock" and if you're lucky it might slightly exceed that by a few percent at stock settings based on silicon lottery, but it's not magically going to boost to 6GHz without some pretty extreme tinkering and modification to power limits, cooling, BIOS etc.
Default boost behavior of the reference card exceeds that on average and at max; that doesn’t even take into account the issues with the reference cooler and it’s limits.
Navi in any form is by no means limited to 2500mhz; if you want to hide behind marketing lingo that doesn’t match reality, be my guest. Clocking artifacts and bad silicon? 2.5ghz limits? They didn’t knock it outta the park and they definitely didn’t hit efficiency targets, but your claims have zero proof.
I think you think I'm arguing something I'm not.
I don't know what you think I'm arguing but I thought I'd made it pretty clear that I'm just citing AMD themselves.
I don't need to defend my opinion because I've not provided one, other than being pleased Navi32 is on the way; If you have issues with what AMD have said in the past, take it up with them, not me.
Not only does the reference model exceed that on average by 100mhz and peak 500 mhz, every single AIB model with adequate cooling and does so significantly; not to mention when allowed to draw 450w+ Navi 31 will clock in excess of 3100-3200.
Where is this magical bad silicon limiting clock speeds to 2500 and causing artifacting? All I see is a missed efficieny target.
As I said earlier, AMD aren't competing with just the Nvidia 40-series, they're competing with every last gen card still sat on store shelves, despite their discounted price. Artificially limiting Navi32 for any reason doesn't make sense to me as AMD will charge us as much as they can for the silicon that they already paid TSMC a fixed price for, a year or more ago.
Whether Navi31 had issues or not, my question was whether AMD have had a chance to respin/tweak/fix/improve on it in the last 10 months. (please select whichever word doesn't trigger you as necessary). Navi 31 is out, it's successful, it's selling well, and it's competitive. How it's flawed or not isn't really the issue and more importantly, it's not going to change past, present, or future Navi31 cards. The only relevance Navi31's design success has is on the speculation of whether AMD found something they could fix in the last ten months or not. At worst, Navi31 is as good as RDNA3 gets, and we're going to get a lukewarm, unexciting (but welcome) Navi32. On the other hand, if there were flaws that AMD has managed to correct in the last 10 months, Navi32 could turn out to be an absolute belter of a GPU.
Since this is a discussion based on a MLID source, here's a 'wonderful' MLID take on Navi31:
ALL RX cards can't be clocked over 2150/2400/whatever VRAM base speed just because AMD decided not to let you push it this hard.
RX 6700 XT can't be clocked to match RX 6750 XT just because the 6750 variant would lose its entire point.
I don't see any reason why they won't limit the cards. The only question is how. 2500 MHz on-die looks too silly to be true, I'd rather expect same clocks as in RX 7600, perhaps with slight differences (no more than 200 MHz).Even if their only competition was Ada their pricing is yet to be justified. RX 7600 is absolutely destroyed by RTX 4060 in RT performance, doesn't have DLSS and is obliterated in terms of power efficiency. And it's not faster than 4060 in pure raster, they're almost the same pace. So this 10 percent discount compared to 4060 is a little bit too much of a rip-off. How does RX 7700 make 4060 Ti a funny purchase, I don't really know. To REALLY compete, 7700 has to be as affordable as $270 but I guess Lisa has a 4 hunnit number in her mind. 7800, OTOH, is perfectly dead because aftermarket 3080s and BNIB 4070s sell really well compared to AMD cards of this calibre. Oh by the way, $300 6800 XT on the aftermarket. And its price is dropping.
I always hated nVidia but AMD is literally the king of the disappointment nowadays. If it's not for conspiracy I don't get it why they try SO EFFIN HARD to lose in every segment. Only the RX 7900 XTX makes sense as 7900 XT is priced virtually the same and is slower by too much to justify this pricing. RX 7600 is effectively a 6600 XT for the price of 6700 XT, thus utter garbage. Other RXes, I don't know... And not sure if I want to know.I'm entirely sure they had. The question is did they use this chance. And the answer is 95 percent negative.
Design changes > tape out > wafer lithography is a 3-4 month process at minimum, last time I read about process nodes - but that was back in the TSMC 7FF and Intel 10>7 rebranding days. I have no such info on TSMC 5nm so I'm making the assumption that it's as long or longer.
As for leakes, there are to be taken with a grain of salt.
Clearly you cannot back up what you said, because there is no way to confirm your opinion. All actual evidence points to the exact opposite of your opinion.
Only 9 should be $200 at most. Unless RGB is very good and makes me look kewl in the eyes of 25y old kids.
If they keep this up, a lot of casual gamers are going to move on to other things like programming, tinkering and outdoor activities. Who knows.. Maybe even get a girlfriend.
So... Husbond as opposed to GF. Nice one