• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

It looks like the 'real' /affordable RDNA3 + next gen NV desktop launch won't launch until September. Thoughts?

I think it's to do with the memory bus specs.

Looking at these 8 and 12gb models:

they basically have to use a higher bandwidth bus for the card with higher amounts of VRAM. They'd rather not as these were considered mid range GPU dies (same core config for both), these days well suited to gaming at 1080p, but struggling a bit at 1440p.

The bit that is most worthy of criticism, is the 128 bit bus on the RTX 3060 with 8GB VRAM. Combined with 15gbps GDDR6, it's likely limiting the card's performance at 1440p.

A 128 bit bus design was quite common for older cards, and would use fewer components. There was a shortage in DRAM and semiconductor components in 2021 and 2022, I guess they thought it would be better to produce a higher number of graphics cards with 8GB VRAM + 128 bit bus designs, than 192 bit.

This year, there is an oversupply of DRAM components, according to this:
https://www.theregister.com/2023/02/04/semiconductor_market_bubble/

So I don't expect we will see the same kind of economising measures that we saw in 2022, for DRAM. For one thing, Nvidia knows quite a lot of PC gamers will only buy graphics cards with high VRAM amounts, it has become part of the marketing for graphics cards.

In January, TSMC reported an 'imbalance' of chips, stating that 'it will take until the first half of 2023 to rebalance inventories to 'healthier levels''.
https://kr-asia.com/chip-glut-to-last-most-of-2023-while-automotive-crunch-persists
 
Last edited:
What's wrong with 8GB for the RTX 4060, if the card is priced well?

Hardly going to be a 4K card is it?

Not really a stretch that two versions coming out, one with 12gb later on.
 
Last edited:
I wonder what went wrong with Navi31 production, as there still aren't that many 7900 XTXs available for sale... Just 12 on the website I'm looking at.

There's more 7900 XTs available, but not a lot more.
 
Last edited:
It looks like there might be some Navi31 chips with integrated vcache coming at some point.

I guess maybe I'm a bit out of date on this rumour, but it seems like a smart move if it's true.

It does make me wonder if that extra cache might be of use for ray tracing.

I think it would be a good idea to keep the 'X3D' branding for these GPUs...
 
Last edited:
There's one website I'm looking at (not a well known website) that is expecting RTX 4070 TIs for ~£780 in about a week. So the prices are slowly coming down.

The good news is that they seem to be out of stock for most of the last gen cards, like the RTX 3050, RX 6600, RX 6800, RX 6800 XT, RX 6900 XT, RTX 3080s and RTX 3090s.

There's still some stock for RTX 3060s, RTX 3060 TIs, RTX 3070s, RTX 3070 TIs and RX 6700 XTs though, and some RX 6500 XTs.
 
Last edited:
I think most people will just play at 1080p (as this is still a very common monitor resolution). As such, I don't think that will hold back game development.

An exception might be if people want to enable ray tracing at 1080p.
 
Last edited:
Do you think we will get Navi33 graphics cards in April?

No sign of Navi32 yet, not sure these are production ready.

But Navi33 is definitely shipping out in mobile devices. Can't say that's exciting for PC gaming, but it might be all we get.
 
Last edited:
I think there must be a production issue with mid/high end RDNA3 GPUs.

AMD wouldn't want to miss out on that sweet mobile GPU profit.
 
Last edited:
RTX 3000 series FE cards have finally been removed from Nvidia's shop websites:

Seems like good news.

Does anyone know what date Nvidia stopped selling RTX 2000 series FE cards on their website?
 
Last edited:
There's a possibility of a further cut down Navi31 card being released, as a successor to the 6800XT.

It doesn't seem like the yields for Navi31 have been that good (so far), but for a cut down chip, they may be improved.

It'd guess the MSRP would be about the same as the RX 6800 XT (UK MSRP was ~£600), probably with a similar amount of Compute Units.

Even with fewer CUs than the RX 7900 XT, it seems likely this hypothetical card would be competitive with the RTX 4070 TI.
 
Last edited:
March was quite an optimistic estimate, the production of RDNA3 GPUs (mobile or desktop) hasn't been sufficient so far. Prices coming down though for Navi31, so maybe in 2-3 months for AMD.

For Nvidia, looks like mid/end of April - not sure if there will be 1 or 2 new cards at this point.

On the Nvidia side, it's possible to buy high end chips like this in laptops:

Which is an AD104 GPU.

For the last gen, we saw quite powerful desktop and mobile GPUs based on GA104 (the RTX 3070 TI and RTX 3080 mobile both maxed out the shader count), so it will be a similar story for this generation.
 
Last edited:
It's interesting to look at what different TSMC fabrication processes achieved when comparing Intel, AMD and Nvidia graphics cards:
https://www.techpowerup.com/gpu-specs/arc-a750.c3929
https://www.techpowerup.com/gpu-specs/geforce-rtx-4090.c3889
https://www.techpowerup.com/gpu-specs/radeon-rx-7900-xtx.c3941

Intel's Alchemist GPUs have transistor densities of 53.4 million transistors per mm², using TSMC's 6nm fab. process. This is similar to what we saw from last generation RDNA2 and Ampere GPUs.

'Ampere Next' GPUs have transistor densities upto 125.5 million transistors per mm² using TSMC's 4nm fab process.

RDNA3 achieved the highest transistor densities on the GCD itself, with transistor densities of 150.2 million transistors per mm² using TSMC's 5nm fab process.

If Intel's Arc Battlemage GPUs use either the 'Intel 4' or 'Intel 20A' fabrication processes, there's likely to be a huge improvement in transistor density, vs Arc gen 1 GPUs. So, Battlemage could end up being a hugely scaled up version of it's predecessor, potentially at a similar TDP of around 225w, or more if they opt to use GDDR6X. I'd guess that the 'Intel 4' process would be the more likely choice for Battlemage, as it's more likely to be ready by Q3-Q4 2024.

It's a fair bet that intel opted for TSMC's 6nm process for Alchemist GPUs, because the manufacturing cost is significantly lower, than more advanced processes like 5 and 4nm. EDIT - Intel said the decision to use an external fab. was also down to their available manufacturing capacity.
 
Last edited:
Here we go, a bit later than I would have liked:

RTX 4070 available to buy on 13th April - This info appears to be genuine.

No sign of Navi32, I'd guess they want to launch these alongside FSR3 (which there could be more info on this month at GDC).

No word on price. Personally, I would find between £550-£575 acceptable, considering this is intended to be a successor to the RTX 3070 (and taking inflation into account).

My impression is that the rumoured 5,888 shaders sounds rather low. Nvidia could only make up for that by clocking these GPUs higher.

To exceed 30 TFlops, a clock rate of 2600Mhz would be needed. 2800Mhz would get them to 32.9TFlops.

It looks like there is going to be a RTX 4070 FE, which is always something to watch out for.
 
Last edited:
Makes sense to launch cheap cards with FSR3 because then you can have deceiving PowerPoint slides showing super high performance compared to previous generation because of fsr3 frame generation
Deceiving or realistic? I personally think AMD should aim for a universal driver option that can enable FSR3 (or the frame generation part) in all DX11/12 games. Probably not going to happen though.

Any marketing material should really point out that achieving 4K (natively) at smooth framerates is still very difficult in the most demanding games.
 
Last edited:
It's notable that you don't see really high clockrates (at stock) on AD102 and AD103 GPUs... They can manage about 2600Mhz.

It's kind of what I was saying before in another thread - that the best value cards for consumers are often the ones with the highest clockrates (relative to other cards in the same generation). Also, they are generally cheaper to produce compared to the highest end models.
 
Last edited:
It's not a bad increase from the last gen is it?

EDIT - Well I was wrong about the RTX 4070 maybe having more shaders than 5888... Hopefully AMD will take advantage of that.
 
Last edited:
So, I thought this was worth mentioning:

Psjjh7qXPpY1uUtJ.jpg


A new professional Navi31 based card was released recently and it looks powerful and quite power efficient.

I hope there's a consumer version with less VRAM - Ideally at some point in Q2 2023.

Looking back at Navi21, all of the workstation cards had N21 consumer equivalents, all with 16GB of VRAM:
https://www.techpowerup.com/gpu-specs/amd-navi-21.g923
 
Last edited:
Back
Top Bottom