• 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.

AMD RDNA3 unveiling event

Good to hear. Will be interesting to see if any power fixes improve performance by freeing up power for the desired tasks. The reference seems to choke from lack of power at times. Wishful thinking that perhaps the power is there but horrifically tied up.
 
AMD: rdna3 has +50% perf/watt over rdna2

Reality: say what???

Yes, it's nowhere near what they claimed and no idea what is going on when frame-limited to 144Hz, but the relevant chart was the one before, so unsure why the twitter used that one:
DODPBXz.png
so +17% to +18% but -4%.

Still, chiplet was always going to be negative for perf/watt.

If AMD had passed the cost savings from going chiplet onto us consumers that would probably have acceptable to most (but that doesn't let AMD PR off the hook), but instead AMD went up in price and kept the increased* margins from going chiplet to themselves.

Except, with the way they have over-priced Navi31 (not just absolutely, but it has actually done the impossible: made 4080 like less of a rip-off) it will only be higher margins not overall profit as they seem to be aiming for a 5% marketshare at this rate.

* The only caveat about margins is that we don't know if there are any bottlenecks going chiplet at binning and packaging. I'm inclined to think they are minimal since Zen2/3/4 have done very similar.
 
so +17% to +18% but -4%.

Still, chiplet was always going to be negative for perf/watt.

If AMD had passed the cost savings from going chiplet onto us consumers that would probably have acceptable to most (but that doesn't let AMD PR off the hook), but instead AMD went up in price and kept the increased* margins from going chiplet to themselves.

Except, with the way they have over-priced Navi31 (not just absolutely, but it has actually done the impossible: made 4080 like less of a rip-off) it will only be higher margins not overall profit as they seem to be aiming for a 5% marketshare at this rate.

* The only caveat about margins is that we don't know if there are any bottlenecks going chiplet at binning and packaging. I'm inclined to think they are minimal since Zen2/3/4 have done very similar.
Will be interesting to see how navi 33 performs as that one is supposed to be monolithic.
 
Good to hear. Will be interesting to see if any power fixes improve performance by freeing up power for the desired tasks. The reference seems to choke from lack of power at times. Wishful thinking that perhaps the power is there but horrifically tied up.

The pre-launch rumours were that the 7900XTX reference would be limited to allow the AMD AIB's to add more value. This was in opposition to Nvidia's policy.

I think this rumour looks like one that was true.

mtVQFx8l.png.jpg

The TUF is a solid Mid-tier card, the Tier 1 cards have not shown up yet. The 7900XTX Strix OC LC could be a bit of a monster. In this type of configuration the 7900XTX is about twice the speed of a 6900XT. I would not be surprising to see (un-announced but-we-know-its-coming) 7950XTX Strix OC LC to be in the 70-80fps range, at stock.
 
Last edited:
The pre-launch rumours were that the 7900XTX reference would be limited to allow the AMD AIB's to add more value. This was in opposition to Nvidia's policy.

I think this rumour looks like one that was true.

mtVQFx8l.png.jpg

The TUF is a solid Mid-tier card, the Tier 1 cards have not shown up yet. The 7900XTX Strix OC LC could be a bit of a monster.
isnt that oc manually applied tho? stock aib card is green, oc yourself and you can get purple for who knows what power draw!
 
So, from my light reading, I get the impression the XTX is fantastic for rasterization but for RT it is lacking somewhat.

They really haven't made upgrading both cpu and gpu this gen easy.

8K res seems pointless. Maybe if VR becomes more widespread?
 
Last edited:
Will be interesting to see how navi 33 performs as that one is supposed to be monolithic.
Might be but the earlier rumours said TSMC 6nm so see how that works out.

I suspect it will have much better perf/watt but not as much as it could have had.

That is, at a rumoured 200mm² Navi33 seems to be attempt to make a GPU as cheap as possible to manufacture (and remember kids: cost to make has little do with how much they'll attempt to charge us!), I wouldn't be surprised that it comes running outside of the process' clock/watt sweetspot, might be memory bandwidth starved etc. all to save 10-15mm² to increase margins.

Well, that's my fear anyhow. Designed to be extra small and cheap to make but ends up loosing in benchmarks by a few %, so starts off overpriced for what it is, AMD get next to no sales and 6/9/12/18 months later they sell it really cheap.

History? Well sometime it can repeat.
 
Last edited:
Might be but the earlier rumours said TSMC 6nm so see how that works out.

I suspect it will have much better perf/watt but not as much as it could have had.

That is, at a rumoured 200mm² Navi33 seems to be attempt to make a GPU as cheap as possible to manufacture (and remember kids: cost to make has little do with how much they'll attempt to charge us!), I wouldn't be surprised that it comes running outside of the process' clock/watt sweetspot, might be memory bandwidth starved etc. all to save 10-15mm² to increase margins.

Well, that's my fear anyhow. Designed to be extra small and cheap to make but ends up loosing in benchmarks by a few %, so starts off overpriced for what it is, AMD get next to no sales and 6/9/12/18 months later they sell it really cheap.

History? Well sometime it can repeat.
I'm supprised they just didn't build them all on 6nm monolithic, would have probably got similar perf per watt and could have lowered the costs.
 
Last edited:
I'm supprised they just didn't build them all on 6nm monolithic, would have probably got similar perf per what and could have lowered the costs.
Well, I guess someone at the Radeon Group is proud of being first to chiplet.

However, like being proud about contributing a lot to HBM's development, well, it seems nobody did any risk assessment as Fury and Vega demonstrated, as like going DDR5 only with Zen4 going big on chiplet is a very risky move. In fact, HBM was even riskier than DDR5 as that will eventually become mainstream.

I think caution would have seen AMD only do one GPU as chiplet this gen.
 
Last edited:
Well, already missed the chance to grab the current cards (for now). Might as well see if there's a RX7950 XT(X) coming later on, and grab that to pair with the 7950X CPU and of course, as a rememberance of the venerable HD7950 3GB I had before it died from sag popping out some pins. The RX580 8GB I've got is working a treat with the CPU in BG3 thankfully at 1080p/1440p, so I can live with that for now. :) A noticeable improvement going from a 4930k 4.2Ghz to a 7950X at 4Ghz all cores (Eco 65W).
 
So, from my light reading, I get the impression the XTX is fantastic for rasterization but for RT it is lacking somewhat.

They really haven't made upgrading both cpu and gpu this gen easy.

8K res seems pointless. Maybe if VR becomes more widespread?

So the RTX 3090Ti is `somewhat lacking` at Ray Tracing as well? Literally it matches it
 
Well, I guess someone at the Radeon Group is proud of being first to chiplet.

However, like being proud about contributing a lot to HBM's development, well, it seems nobody did any risk assessment as Fury and Vega demonstrated, as like going DDR5 only with Zen4 going big on chiplet is a very risky move. In fact, HBM was even riskier than DDR5 as that will eventually become mainstream.

I think caution would have seen AMD only do one GPU as chiplet this gen.

I think they are doing ; RDNA3 is the test for all chiplets next time, just as 5800X3D was the test for stacked vcache, and w`ll see the killer chip with the 7800X3D
 
Yes, it's nowhere near what they claimed and no idea what is going on when frame-limited to 144Hz, but the relevant chart was the one before, so unsure why the twitter used that one:
DODPBXz.png
so +17% to +18% but -4%.

Still, chiplet was always going to be negative for perf/watt.

If AMD had passed the cost savings from going chiplet onto us consumers that would probably have acceptable to most (but that doesn't let AMD PR off the hook), but instead AMD went up in price and kept the increased* margins from going chiplet to themselves.

Except, with the way they have over-priced Navi31 (not just absolutely, but it has actually done the impossible: made 4080 like less of a rip-off) it will only be higher margins not overall profit as they seem to be aiming for a 5% marketshare at this rate.

* The only caveat about margins is that we don't know if there are any bottlenecks going chiplet at binning and packaging. I'm inclined to think they are minimal since Zen2/3/4 have done very similar.

Am sure projected doesn't mean guaranteed.
This is just another example of the Internet not reading a slide correctly and running away with it.
 
Yet there more people with VR headsets than RDNA 2 cards (Steam Survey)

It would seem to me that they can't afford to leave any potential market behind.

Those vr headsets are also the phone connected ones such as samsung or the free cardboard ones that bt gave away for the football....
 
Back
Top Bottom