• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

5090 competitor?

supposedly, but already being talked about in all the other 9000 series threads...
 
AMD already confirmed it would not be pursing the high end segment this generation. This isn't a choice on their part and such claims are BS. Essentially, AMD has fallen too far behind to compete with NVIDIA. So price/performance in the midrange is all they can do. It's like Toyota building the 2007 Tundra to compete against Ford's F250 and falling well short of that and classifying it as a 1/2 ton truck. If it had built a competitive card at the top it could scale that down for other price points.

At best I think AMD can land somewhere in the territory of the 5070Ti if they are lucky.
 
Current speculation is a 9070 XT with 32 GB of RAM, which is really only going to be useful for the AI folks, I imagine.
 
Like any business you can only do what you can do.
Lamborghini
BMW
GM
Are you talking native, dlls/rsr, 4K, 2k, 1080, what game/work?
How many can afford what.
AMD needs to take the mid high as Intel is after the low mid today.
 
1739479708070.png

Ok rumor is false lol
 
cjcox said:
If you mean performance. No.

If you mean features. No.

If you mean pricing. No.

If you mean power consumption. No.

If you mean form factor. No.

If you mean availability. No.

Apart from those things though, maybe.
Click to expand...
If you mean fire hazard. No
If you mean Bricked cards. No
If you mean issues with the drivers, BIOS, or PCIe. No.
If you mean fake MSRP. No
If you mean fakeframes with a lot of latency. No
 
SamiiRoss said:
If you mean fire hazard. No
If you mean Bricked cards. No
If you mean issues with the drivers, BIOS, or PCIe. Probably.
If you mean fake MSRP. Yes
If you mean fakeframes with a lot of latency. Yes
Click to expand...

I fixed that for you as AMD cards do indeed support frame generation. It just isn't multi-frame generation FSR has a ton of latency with it as well. AMD has had its share of BIOS and driver issues in the past. Far more than NVIDIA has historically. Not to mention AMD even had its card boxes stripped of the PCI-Express compatible logo because AMD cards were out of spec pulling way more power than the spec allowed for. As for the MSRP, its not a fake MSRP. If you can find the FE cards in stock you can get them for MSRP. AIB's are all over the map charging whatever they like and scalpers are even selling through Newegg and Amazon so they can charge what they like.

There are people even trying to take advantage of the video card fever scalping the 7900XTX's for $1,500 right now. It's not purely an "NVIDIA" thing.
 
When you look the original post for yourself
https://www.chiphell.com/thread-2671520-1-1.html

At no point, Zhangzhonghao seem to mention anything (if the google translate is accurate) pointing toward a 5090 competitor that I can see ?

It even talk about it being on the 9070 series according to his second post, like a clampshell version a la 4060TI 16GB and that different rumours than the article is talking about was said to be false.

the 4060Ti 16GB or the 7800xt were not competitor to the 4080 super-5080 just because they shared the same vram amount.....
 
Last edited:
How many times over the years has AMD said "We arent competing on the high end range for GPU's"
 
We can hope. I always lean toward team green, but we desperately need someone to give them some form of competition on the higher end.
 
They just need to bring something to the table in the upper mid-range (at least 4070Ti tier or above) with at least 16Gb of VRAM that comes in at $650 or less. They would clean house given how over-inflated Nvidia’s pricing has become, not to mention the total lack of availability.
 
Last edited:
Dan_D said:
I fixed that for you as AMD cards do indeed support frame generation. It just isn't multi-frame generation FSR has a ton of latency with it as well. AMD has had its share of BIOS and driver issues in the past. Far more than NVIDIA has historically. Not to mention AMD even had its card boxes stripped of the PCI-Express compatible logo because AMD cards were out of spec pulling way more power than the spec allowed for. As for the MSRP, its not a fake MSRP. If you can find the FE cards in stock you can get them for MSRP. AIB's are all over the map charging whatever they like and scalpers are even selling through Newegg and Amazon so they can charge what they like.

There are people even trying to take advantage of the video card fever scalping the 7900XTX's for $1,500 right now. It's not purely an "NVIDIA" thing.
Click to expand...
Why are people so desperate for 7900 XTXs all of the sudden? Is it really just deepseek?
 
Decko87 said:
Why are people so desperate for 7900 XTXs all of the sudden? Is it really just deepseek?
Click to expand...
Well considering they are on par basically with 5080 performance, I can see why people want them.

Nvidia's launch of the 5090/5080 has not been great, and people don't want to pay $1500 for a $1000 card that has the same performance of a 7900xtx you could of gotten a long time ago.
 
Decko87 said:
Why are people so desperate for 7900 XTXs all of the sudden? Is it really just deepseek?
Click to expand...
Not sure desperate for a particular GPU, but for any GPU in that range, in some market there is just absolutely nothing.

Looking at the 7900xtx-7900xt, 4070ti super or up current offer on newegg:
https://www.newegg.com/p/pl?N=10000...32394 601408874 601408875 601432392 601341679
https://www.newegg.com/p/pl?N=10000... 601408874 601408875 601432392 601341679 8000 (sold by newegg)

Not a single nvidia card, seem to be available (you can pay a forturne for some coming from hong kong...)

Nothing on best buy:
https://www.bestbuy.com/site/search...yid$abcat0507002&type=page&usc=All Categories

A single sku for the 7900xtx, 7900xt, 4070ti super, 4080, 4080s, 4090, 5080, 5090 is available, 7900xt xfx for $750. They are desperate for an high end gpu in general.
 
Dan_D said:
I fixed that for you as AMD cards do indeed support frame generation. It just isn't multi-frame generation FSR has a ton of latency with it as well. AMD has had its share of BIOS and driver issues in the past. Far more than NVIDIA has historically. Not to mention AMD even had its card boxes stripped of the PCI-Express compatible logo because AMD cards were out of spec pulling way more power than the spec allowed for. As for the MSRP, its not a fake MSRP. If you can find the FE cards in stock you can get them for MSRP. AIB's are all over the map charging whatever they like and scalpers are even selling through Newegg and Amazon so they can charge what they like.

There are people even trying to take advantage of the video card fever scalping the 7900XTX's for $1,500 right now. It's not purely an "NVIDIA" thing.
Click to expand...
True. I just checked Newegg and the 7900XTX is selling at $1499-1599 apiece. This isn't even a NEW generation GPU.

Scalpers are taking advantage of video card shortages by buying up the small amount of stock and reselling at even higher prices.

This has made the hobby for enthusiasts a very expensive proposition.

I still don't understand why there's no mechanism to allow real customers to buy video cards at MSRP from retailers.

Instead tit feels like 90% of the initial batch is swooped up by scalper bots and end up on secondary market for hugely inflated prices.

It makes the whole gpu industry feel very predatory and unfriendly for hobbyists.
 
Last edited:
Brackle said:
Well considering they are on par basically with 5080 performance, I can see why people want them.

Nvidia's launch of the 5090/5080 has not been great, and people don't want to pay $1500 for a $1000 card that has the same performance of a 7900xtx you could of gotten a long time ago.
Click to expand...
It's generous to call them "on par with a 5080". Considering a 7900XTX is only really equivalent to a 4080 Super and the 5080 is clearly faster than that, though not by huge margins. That said, what really puts the 7900XTX in the back of the bus is poor ray tracing performance and FSR being inferior to DLSS etc. In my opinion, despite its raw rasterization performance the 7900XTX is functionally the 6th best option for a video card today behind the 5090, 5080, 5070Ti, 4090 (if you can find one) and the 4080 Super. It lands more like 5th in terms of raw performance but the caveats are pretty significant.
 
Dan_D said:
It's generous to call them "on par with a 5080". Considering a 7900XTX is only really equivalent to a 4080 Super and the 5080 is clearly faster than that, though not by huge margins. That said, what really puts the 7900XTX in the back of the bus is poor ray tracing performance and FSR being inferior to DLSS etc. In my opinion, despite its raw rasterization performance the 7900XTX is functionally the 6th best option for a video card today behind the 5090, 5080, 5070Ti, 4090 (if you can find one) and the 4080 Super. It lands more like 5th in terms of raw performance but the caveats are pretty significant.
Click to expand...
I agree with this statement, 5070 ti is a bit arguable because there's enough of a raster difference there to matter, but overall i'd say it's a better card, especially if you can get it for 750-850. At the same price it's a toss up IMO. The 9070 XT should be the only real go to for AMD post march 6th. Unless you're running deepseek locally, then you might want the vram. I still think the 7900 XTX is a great card, but it likely wont age well.
 
Dan_D said:
It's generous to call them "on par with a 5080". Considering a 7900XTX is only really equivalent to a 4080 Super and the 5080 is clearly faster than that, though not by huge margins. That said, what really puts the 7900XTX in the back of the bus is poor ray tracing performance and FSR being inferior to DLSS etc. In my opinion, despite its raw rasterization performance the 7900XTX is functionally the 6th best option for a video card today behind the 5090, 5080, 5070Ti, 4090 (if you can find one) and the 4080 Super. It lands more like 5th in terms of raw performance but the caveats are pretty significant.
Click to expand...
I think people are (were?) starting to freak out about the 16 GB VRAM "issue" and that drove a lot of sales. They saw that in like, one or two games at ultra-high 4K settings you *might* start to run out of VRAM at 16 GB.

Let's not discuss that you're not running those ultra pathtracing settings with a 7900 XTX in the first place...
 
No amd is not competing with 5090 or even the 4090 with this gen. The are targeting the 5070/5070Ti performance tier.

Wait for RDNA5 maybe it can compete with 5090 because the 5000 series performance gains were pretty lackluster.
 
I mean technically I do believe they are capable of such. Keeping in mind 9070XT is on par with a 4080/5070ti in raster, and being around 390mm in die size compared to the 5090's 744 is at least something of merit. They avoided the high end this gen as the plan was to work on a MCM GPU for the high end like they did last gen, but apparently had problems with it and scrapped the idea. Also RDNA 3 high end didn't hit the expectations performance wise in their metrics. If they scaled up RDNA 4's monolithic die, I think they could do pretty well against Nvidia's high end this gen, but the money is the mainstream segment...So targeting it is not an unwise position to take. I'm curious what RDNA 5's approach will be, if they've solved the shortcoming of MCM for a GPU, or if they go the monolithic route.
 
Ikasu said:
I'm curious what RDNA 5's approach will be, if they've solved the shortcoming of MCM for a GPU, or if they go the monolithic route.
Click to expand...
Nvidia has been king of monolithic gpus. I expect Amd’s only chance at a performance leap to match or overtake Nvidia will be a multi gpu chiplet solution. It will be tricky to balance the load and feed them data but they have done it well on cpu side for years. Can they translate that know how to the gpu division which has been losing for years? Stay tuned for RDNA5.
 
vjhawk said:
Nvidia has been king of monolithic gpus. I expect Amd’s only chance at a performance leap to match or overtake Nvidia will be a multi gpu chiplet solution. It will be tricky to balance the load and feed them data but they have done it well on cpu side for years. Can they translate that know how to the gpu division which has been losing for years? Stay tuned for RDNA5.
Click to expand...
Let's not forget the 6900/6950XT gave Nvidia a run for their money. The 6950XT beat the 3090/3090ti, and was also monolithic. Nvidia is beatable, the only reason Nvidia had such a demanding lead for the 4000 series with the 4090, is due to the fact the MCM approach ended up not hitting the mark AMD was hoping for. It didn't hit the performance estimation of what they expected on paper. It would be nice if they solved the shortcomings they ran into with the 7000 series...but MCM is definitely more complex on the gpu front compared to their CPU line...
 
Last edited:
Ikasu said:
Let's not forget the 6900/6950XT gave Nvidia a run for their money. The 6950XT beat the 3090/3090ti, and was also monolithic. Nvidia is beatable, the only reason Nvidia had such a demanding lead for the 4000 series with the 4090, is due to the fact the MCM approach ended up not hitting the mark AMD was hoping for. It didn't hit the performance estimation of what the expected. It would be nice if they solved the shortcomings they ran into with the 6000 series...but MCM is definitely more complex on the gpu front compared to their CPU line...
Click to expand...
the 6000 series had a node advantage on the nvidia 3000 series, remember those cards were made on mediocre samsung process. Considering the 9070 xt is more dense than a 5080 and loses to it by about 15-17 percent in raster and even more in RT means they're behind by a generation at least. Don't get me wrong, I expect the 9070 xt to be a great card, but they have a bit further to go. Fortunately (for amd), Blackwell offers almost no improvements so they could catch up come UDNA time if they really put the resources forward. Or if they don't catch up, be significantly less behind.
 
Last edited:
Decko87 said:
the 6000 series had a node advantage on the nvidia 3000 series, remember those cards were made on mediocre samsung process. Considering the 9070 xt is more dense than a 5080 and loses to it by about 15-17 percent in raster and even more in RT means they're behind by a generation at least. Don't get me wrong, I expect the 9070 xt to be a great card, but they have a bit further to go. Fortunately, Blackwell offers almost no improvements so they could catch up come UDNA time if they really put the resources forward. Or if they don't catch up, be significantly less behind.
Click to expand...
Yep this is absolutely a good point to make. One wonders what a proper 384-bit card in the RDNA2 era might have looked like vs the 3090 / 3090 Ti. The problem RDNA2 had was that the performance scaling dropped off at 4k. As for RDNA3, I think they absolutely thought it would compete with the 4090, but the MCM approach really did not work to the levels they thought it would.

As for the process differences, it's rather interesting.

GeForce 30-series (Samsung 8nm) vs RX 6000-series (TSMC 7nm)
GeForce 40-series (TSMC 4N) vs RX 7000-series (TSMC N5 / N6)
GeForce 50-series (TSMC 4N) vs RX 9000-series (TSMC N4P)
 
Decko87 said:
the 6000 series had a node advantage on the nvidia 3000 series, remember those cards were made on mediocre samsung process. Considering the 9070 xt is more dense than a 5080 and loses to it by about 15-17 percent in raster and even more in RT means they're behind by a generation at least. Don't get me wrong, I expect the 9070 xt to be a great card, but they have a bit further to go. Fortunately, Blackwell offers almost no improvements so they could catch up come UDNA time if they really put the resources forward. Or if they don't catch up, be significantly less behind.
Click to expand...
True. The samsung process left more to be desired. But considering the 7900XTX outpaces a 4080/s (in raster) on a first time stab at MCM on a GPU approach isn't too shabby. I agree blackwell's lackluster performance gains are a benefit to AMD...But it's really hard to gauge how far behind AMD really is. Don't get me wrong, they are definitely behind on their AI software stack, DLSS, other gaming related software enhancements, etc....Pretty much playing catch up. But when you really reflect on the entirety of the situation, they weren't really that far behind on the hardware front. They had the better raster card for the 6000 series, took a bold approach for 7000 with MCM, which didn't pan to expectations, beating everything but the 4090. Here comes RDNA 4, roughly same die size as the 4080/5080, and offering 4080 tier performance (raster) in a monolithic chip (and not much slower than 5080), catching up quite a lot in regards to ray tracing as well. If they didn't take the MCM approach for the 7000 series, who knows, they may have had a 4090 class competitor on their hands (in raster).

If they can nail MCM for GPU's, that would be impressive, even Nvidia hasn't been able to solve it... There approach isn't really a true MCM design, but more of a side by side glue of dies based on my limited understanding.

Not sure what to think of the unified gpu architecture though. Don't get me wrong, it's a good move for decreasing R&D costs. But I believe it kinda ends up hurting gamers doing so. Looking at the current situation, you have Nvidia with a design shared between AI and gaming, which they blatantly favor AI and have paper launched the Geforce counterparts with no stock in sight. I believe it would make more sense to have AI on the bleeding edge node, and a separate design for gamers one node behind, in order to avoid cannibalizing each other. I mean overall, yes, you could have a unified design and still take this approach...I'm not an engineer or chip architect, but I believe having a chip designed specifically ground up for AI would probably yield better performance for those tasks. Compared to having a hybrid approach as a gpu built to do both. I'd expect the BIG companies out their...meta, alphabet, and any other big honcho developing their own chips to have an advantage in performance by building tailored chips. Which in the long run, will end up hurting Nvidia's bottom line....Which probably explains why they are spearheading AI development so quickly, to milk as much money as they can before that bubble pops, and the market stabilizes with more competition, less demand, and big companies developing their own chips.
 
  • Like
Reactions: Decko
like this
vjhawk said:
Nvidia has been king of monolithic gpus. I expect Amd’s only chance at a performance leap to match or overtake Nvidia will be a multi gpu chiplet solution. It will be tricky to balance the load and feed them data but they have done it well on cpu side for years. Can they translate that know how to the gpu division which has been losing for years? Stay tuned for RDNA5.
Click to expand...
If chiplet are the future of GPUs, I imagine Nvidia getting there first in once they can get performance to be at least on par with monolithic. Let's be honest here, Nvidia's R&D alone totally dwarfs AMD's whole gaming revenue. If someone is going to make it work right, it won't be Radeon first.
 
Ikasu said:
I mean technically I do believe they are capable of such
Click to expand...
The way GPU work and it could be naive I do not see why AMD could not just make it 5090 big and match NVIDIA performance.

Copy paste CU, 50% more memory controller, maybe the GPU become too long and not square enough with their current design.

13.56 x 26.3 mm is getting far from a square and maybe that shape was important in some way ? make 256 bits controller possible on a smaller die (the less edge you have by surface the closer you get to a perfect square) ?

They seem to have 3 32bits controller on the long side, 2 32bits at the top:
Gk4uxCmW4AAB8gP?format=png&name=900x900.png
AMD-RDNA4-NAVI48-1200x624.jpg

Maybe it start to get strange yield wise with that shape if you get big.

I always imagined a big part of the issue was a lack of secondary usage for that big die, nvidia do x40-RTX 6000, a low volume 5090 would not be a big deal as long gb202 overall is a giant multi-millions units sold at really good price, hard to compete head to head with that if your pro line do not sales.

I do not know how much money goes into doing different version at different size, but maybe it would not be a lot of unit to amortize it for AMD,


AD102 was milked higher and down the line a lot:
https://www.techpowerup.com/gpu-specs/nvidia-ad102.g1005

Near perfect die L40-RTX 6000 down to 8448 core 4070ti super made on it and L20, if it did cost 40 millions to develop, if you sales 3 millions units a lot of them more than $5000, easy to amortize, if it is possible you would sell less than 100,000.... that $400 by gpu sold, before spending a cent making and selling them. Nvidia R&D reach 1 billion a month I think lately.

Obviously developing a new gpu generation cost a giant fortune (more than a billion maybe nowaday ?), just no idea how much it is to develop an extra sku and how much you add to the tag if your biggest one you made was a bigger one. My intuition, the bigger the biggest die of your family generation, the costlier it is to make that generation. Planning all that signals and sync over that complex of a system being a bigger deal. More power needed, more complicated to manage and distribute and so on.

Hard to think AMD couldn't do it technically, could they sell enough of those larger die to be worth all that trouble and added cost is maybe more what blocking it. Doing what Apple do, that maybe more something out of reach of others just on a technical level.

Apple spent a billion just to do the tape out of the M3, if you do not move millions of chips that you reuse in many product like them, you cannot do that if you need to sales them cheap. (https://www.extremetech.com/computing/apple-spent-1-billion-on-the-m3-tape-out-says-analyst), when you look at apple chips design it look so much more complicated and work than everyone else.

In that field, the more units you move, the more the dev$ / per unit go down and the more you can sell of them by reducing their price, that the magic of AMD chiplet they reused in everything, the dev cost by chiplets sold get extremelly low and Nvidia giant skus stack for a new gen and how much use of each die they get, super expensive and nice volume on the biggest die pro line and exploding gaming Laptop product on the rest of the stack.
 
Last edited:
If AMD could simply beat Nvidia by going bigger they probably would have tried it already.

That means there's some limitation with their technology that doesn't allow them to just make a bigger die GPU to win.

As far as the multi-gpu chiplet design. Rumors are that AMD has been working on it for years but scrapped it for RDNA4 when they saw it wouldn't yield the performance they wanted.

As far as Nvidia mastering multi-gpu first. I don't see a reason for them to do it until AMD can put out a multi-gpu solution that beats their monolithic design.

AMD has a real "prove it" or remain relegated to second tier status problem. Until they can deliver a halo product that can beat Nvidia's top card, I don't see them getting out of the 10-20% market share area.

Right now Nvidia dominates with 90% of the GPU market share and doesn't feel threatened at all by AMD.

Yes the 9070XT might carve out a niche in the $599 gpu market segment but that's only because Nvidia doesn't choose to compete in that segment. They priced the 5070 Ti at $749 for a reason. It's still faster than the 9070XT, it's just unobtainable at MSRP because of supply.

Nvidia is content to push as much silicone to corporate sector at $25k per gpu rather than waste that silicone on consumer products that top out at $2000 for the 5090 which is already a 25% markup of their previous product the 4090. This to me explains the extreme paper launch. They don't feel the consumer market is profitable enough to waste their top silicone on it. AI is just far too profitable right now.

There still exists a world where AMD could take Nvidia by surprise and push RDNA5 out as faster than an Nvidia halo card because Nvidia was too fat and content to push their consumer GPU's more than 30% per generation as we see with the 5090. But we'll have to wait and see. 6090 or RDNA5 which one will be the fastest?
 
vjhawk said:
That means there's some limitation with their technology that doesn't allow them to just make a bigger die GPU to win.
Click to expand...
I am really unsure about that one, 6900xt was 520mm the next gen biggest chip was 300mm and I doubt it was because tech wise it could not have been 600 to match a 4090, I really do not know anything about any of this, but it could be financial and product line more than raw unable to do it technically or some design difference that will not scale.

If developing a large version make that generation of GPU cost a 50 million more and they think they would sell only 100,000 of that really big die product over its life, the first $500 per GPU is just paying your R&D that you are paying a year of interest on (if you do not have a pro-line that use it directly that relevant)

You can spend 700 millions making a new consoles SOC, if you sales 70 millions of them over their life time, that become cheap per unit to amortize.

Nvidia I do not know how much spend more on blackwell R&D for the biggest die to be that size and the rest of the biggest card to exist, Nvidia not only believe he will sell more GB202 products (5090-RTX6000/X40/etc...) to amortize but some of those are sold for more than $7,000 still today, able to sales a lot of 5090 even with an over $2,000 price tag, much easier for it to make sense financially and turn a profit.

That a bit the same reason AMD make less amount of dies in general, not just not the big one, it can start to be also true in the other direction.

RDNA 3 was down to navi 31-32-33, RDNA 2 had 21-22-23-24, this one seem will have only 2 of them. While Ampere-Lovelace tend to have like 5 different one (without counting the A tsmc variants, the jetson and others).

It is just that the bigger you go and more power the more complicated it is (so maybe they could do it, but cost more) and we are more aware of it.
 
I'd try it out. I want to do a side-by-side comparison of the graphics quality again to see if they have caught up yet...

1757827018621.jpeg
 
You must log in or register to reply here.
Back
Top