AMD Radeon 9070 XT Rumored to Outpace RTX 5070 Ti by Almost 15%

[Rabit]

if price right 9070xt 32 Gb will be mine personal pick for local LLM
Also we have similar situation in past RX 480 4Gb and 8GB not much difference when released by after some time we seen it when game star using above 4GB.

 

[OneSorcerer]

if price right 9070xt 32 Gb will be mine personal pick for local LLM
Also we have similar situation in past RX 480 4Gb and 8GB not much difference when released by after some time we seen it when game star using above 4GB.

Do you really think 16gb vs 32GB is comparable to 4 vs 8GB? Like, you expect a 16GB jump in memory usage in < 5 years?

 

[Sound_Card]

Do you really think 16gb vs 32GB is comparable to 4 vs 8GB? Like, you expect a 16GB jump in memory usage in < 5 years?

He's talking about running a local LLM.

My last guess was conservative, a more hopeful outlook:

9070xt ~ 7900xt raster / 4080 ray - $550 and up
9070 ~ 7900gre raster / 4070 super ray - $450 and up
9060xt ~ 7800xt raster / 4070 ray - $400

 

[Rabit]

Currently consoles have 16GB and we see games use even 16GB Vram, if next gen consoles in 2028 will have 32GB is sure that new games will star utilise it.
This cycle always repeat new gen consoles arrive VRAM requirement increase.
Today 32GB will be nice for LLM or other work application.

 

[mkppo]

This is ridiculous and stupid. rdna3 register files are 196k, while RDNA2 has 128kB, and rtx 30/40/50 has 64kB. The one that really needs improvement here is NV) That's why we see 0% IPC SM improvements for RTX40->50
As for L2, NV went to such a waste of crystal area for one (may be not only) function - SER, which reloads all register files via L2. But they were wrong about the scale of this feature. One game in two years is nothing.

Your claims about no changes are not true. The recommendations for configuring the last level cache and memory are also incompetent.

A calm(er) discussion would be appreciated and I do appreciate your response except the, erm, aggressive part. The reason I said I expect changes to L2/registers isn't necessarily to do with capacity. I was thinking more along flexibility - two operands cannot read from the same register bank and the destination registers also can't be both even or odd. As of now, that massive increase in FP32 throughput isn't really realized in games and such because dual-issue capability is limited and reliant on hand-compiling the code.

As for L2, Ada only does a two tier cache subsystem so their L2 also works to prevent memory bandwidth bottlenecks as compute capacity gets scaled up. RDNA2/3's does 4 way though, and L2 plays a smaller role and L3/Infinity cache rather deals with preventing bandwidth bottlenecks. I can see them rebalancing L2 size along with infinity cache, it's a given because it'll depend on memory configuration and compute throughput among other things. I didn't claim anything about no changes though, nor did i recommend anything specific about L3/IC or memory so i'm not sure what you meant?

 

[medi01]

´´´´´´


9070xt ~ 7900xt for $600 or less
9070 ~ 7900gre for $500
9060xt ~ 7800xt for $400

Sounds right.

Also, HUBs comment, to go lower price upfront, instead of opening with higher MSRP and dropping it later, makes a lot of sense
Reviews rate cards based on MSRP.

Right now RX 7900 XT costs 635€ and RX 7900 GRE 546€. Similar improvement to blackwell

Why are you comparing MSRP to MSRP for NV cards and discounted last gen price to MSRP for AMD cards?

Also, 680 Euro is the cheapest I see here, with most being 700:

https://geizhals.de/?fs=7900xt&hloc=at&hloc=de

 

[Contra]

As of now, that massive increase in FP32 throughput isn't really realized in games and such because dual-issue capability is limited and reliant on hand-compiling the code.

In fact, it is not only manual compilation. Any code launched in wawe64 mode will have double output, and only for wawe32 this will not always be the case, and here the compiler or developer will add as much as possible.

In your comment you confirmed the statement that RDNA3 has no architectural improvements, there is a clock speed deficit compared to RDNA2 and the cache/memory setup should be like NV. Sorry, this is shocking

 

[mkppo]

In fact, it is not only manual compilation. Any code launched in wawe64 mode will have double output, and only for wawe32 this will not always be the case, and here the compiler or developer will add as much as possible.

In your comment you confirmed the statement that RDNA3 has no architectural improvements, there is a clock speed deficit compared to RDNA2 and the cache/memory setup should be like NV. Sorry, this is shocking

I mentioned the wave64 part in my previous post and i'm aware of it. But thanks for their rx7600 article, it seems to have an update on the dual-issue capabilities.

I didn't state RDNA3 has no architectural improvements though, in fact i was referring to the same C&C article you linked and am well aware of the improvements even without factoring in the dual-issue throughput increases. I merely stated that the uplift was less than they expected partly due to them not hitting their clockspeed targets. I can't remember exactly what it was, but I vaguely remember it being around 9.5% vs a ~18% target. A 6950XT reference has an average clockspeed of around 2.4ghz during gaming workloads. 7900xtx reference does around 2.63ghz. 18% would've been ~2.84ghz, which would've changed things up a bit. It's also apparent when looking at the overclocking results how much it would impact performance because the arch itself responds very well to clockspeed increase.

The RX7600 article you linked actually made me realise something I probably should have thought of earlier - we can extrapolate data from there to figure out what a monolithic die would do for L2/VRAM latency. Looks like it's not so insignificant even factoring in smaller gpu's inherent latency advantage - so my initial post about a monolithic ~450mm 7900xtx with 10% higher average clocks landing not too far off a 4090 in raster was plausible.

 

[JAEMzWOLF]

It's around ~20% faster than the RTX 4070 Ti @749$ if RX 9070 XT manages to produce ~15% more performance for 13-20% less money it's obviously a win, win situation. But that's still a very big if.

You really have to be too oinline to say this with a straight face or without an S-tag. Barely better for barely la cost saving is going to do nothing to nvidia's 80-90% market share. NOTHING.

Look, I will sing AMD's praised if I am wrong, but really, they need to blow Nvidia out on either price or performance or both, not sort of with both, bigly with at least one or both. 5070 Ti for 5070 money, is what they need, but we both know they are too "in bed" with nvidia to do that. Every single time nvidia does crazy with prices, AMD just slots in underneath them and then acts like all of us will sing their praise.

It never happens and it never will. I mean, its very VERY clear what to do - Intel released affordable trash and got praised a ton. So. AMD better make them prices great or they get kiss my butt and they can also kiss any gain in market share goodbye - that is, if they ever care.

 

[Contra]

The RX7600 article you linked actually made me realise something I probably should have thought of earlier - we can extrapolate data from there to figure out what a monolithic die would do for L2/VRAM latency

There is another article, right at the beginning of the previous one)) now the difference in latency with a monolithic structure is less than 10%. Compared to the same large monolithic GPU like 6900xt, MCM shows improvements in latency

 

[mkppo]

There is another article, right at the beginning of the previous one)) now the difference in latency with a monolithic structure is less than 10%. Compared to the same large monolithic GPU like 6900xt, MCM shows improvements in latency

Thanks, I must have missed that one as well. I was calculating ~20% based on the 7600 review but turns out subsequent testing by forcing higher power state gives around ~10% which is exactly what AMD claimed. Pretty impressive

 

[Bronan]

The Insanity rumour mills are running as usual AMD must sell it for free and the performance is Blah blech this that and more nonsense i for sure will buy the RX 9070 XT no matter what i never EVER buy the garbage from ngreedia ever again.
But here so many people do not care one bit just whining that AMD is a bad company who does make bad hardware, the same goes for Intel products they are bad and not worth your money
Apparently people lost their brains in the real world they all make good cards and no i do not care if it has whatever logic to improve the game with things i find not worth my money
But it does not matter what i think most here are babling like ngreedia is the only vendor on the planet which makes good cards
And you pay a nice high price for your love for this company, a very smart choice .......
So now you all can react with your negative biased comments i will not read them anyway
I am sick of the insane love for a company who is selling cards which should not exist and keeps selling them for insane high prices yes the ti / super or whatever they cook up nonsense.
It is amazing that almosy 80% of the world think that the competition has bad products or are underwhelming which is again clear nonsense

 

[mkppo]

It is amazing that almosy 80% of the world think that the competition has bad products or are underwhelming which is again clear nonsense

Make that >85% and a quarter of them don't even know about the competition. All they know is they need a RTX GPU to play RT games.

All the clickbaity sheep in the video world don't make it any better either by not even mentioning the competition but rather compare nvidia's product stack.

There were times when 7800xt was priced close to 4060Ti's. You know, the card that's faster in both raster and RT. Sales comparison? Don't even mention it.

It is what it is but it will change at some point.

 

[Krit]

There were times when 7800xt was priced close to 4060Ti's. You know, the card that's faster in both raster and RT. Sales comparison? Don't even mention it.

That's the main reason why RTX 5080 is so slow in comparison to RTX 4080 Super. Consumers are allowing it to happen i mean typical nvidia users who are not even thinking just buying it like a milk in supermarket.

In Latvia RX 7800 XT costs the same as RTX 4060 Ti 16GB and that says a lot.

 

[Rabit]

That's the main reason why RTX 5080 is so slow in comparison to RTX 4080 Super. Consumers are allowing it to happen i mean typical nvidia users who are not even thinking just buying it like a milk in supermarket.

In Latvia RX 7800 XT costs the same as RTX 4060 Ti 16GB that says a lot.

In Poland RX 7800 xt cost less than RTX 4060 TI 16GB

 

[freeagent]

All they know is they need a RTX GPU to play RT games.

Err..

I guess you forgot that AMD doesn't think ray tracing sucks as much as their users do.

 

[AusWolf]

Err..

I guess you forgot that AMD doesn't think ray tracing sucks as much as their users do.

People think RT sucks because you can't run it on a midrange GPU. Not because it doesn't add to your game. If AMD changes that with the 9070 XT, the narrative will change, too.

 

[freeagent]

the narrative will change, too.

Just waiting for the tune to change to wow it isn't shiny like before lol.. this is great

 

[Visible Noise]

That's the main reason why RTX 5080 is so slow in comparison to RTX 4080 Super. Consumers are allowing it to happen i mean typical nvidia users who are not even thinking just buying it like a milk in supermarket.

In Latvia RX 7800 XT costs the same as RTX 4060 Ti 16GB and that says a lot.

So why is the 9070XT so slow? Are consumers allowing that to happen also?

 

[AusWolf]

So why is the 9070XT so slow? Are consumers allowing that to happen also?

Do you know how much the 9070 XT will cost?

 

[mkppo]

Err..

I guess you forgot that AMD doesn't think ray tracing sucks as much as their users do.

Ehhhh I don't know, I can call myself an nvidia user because i've been daily driving a 3090 in my main for the past few years, And I think RT sucks.

Mind you, i'm not talking about RT itself but rather their current implementation. In all these game engines where RT is tacked on, visuals either improve slightly or don't. Or they get worse. But the performance hit is absolutely immense. HUB did a nice little RT roundup which I think it was posted here a few weeks ago which mostly aligns with my views as well.

There's a time and place for everything. Lumen is a step in the right direction but there's a long way to go.

So why is the 9070XT so slow? Are consumers allowing that to happen also?

Can you send me the 9070XT benchmarks please? And the price? Not the first post where you talk about the 9070XT like it has already been released. Give it a break.

 

[freeagent]

i've been daily driving a 3090

Puny 3090

I kid I kid

 

[mkppo]

In Poland RX 7800 xt cost less than RTX 4060 TI 16GB

Yeah this one never really gelled well with me but it is what it is.

Puny 3090

I kid I kid

Haha you say that but next to the 5090 it's pretty 'puny' i'd say. I was cleaning the 3090 block the other day while thinking what my water temp would rise by if I do manage to get a 5090 and throw another ~250w in this loop. The 9950X isn't going to be happy

 

[Super XP]

Well, the 50-series so far is giving AMD lots of room to breathe if you ask me.

Besides, it's not about breathing room. Nvidia is after profits, not more market share. And as long as people buy their cards no matter what, they don't have to lower prices.

The 50-series is probably one of Nvidia's worst "Paper" launches in a while. Not to mention the major issues people are having with them.

Lots of NVIDIA GeForce RTX 5090 & 5090D GPUs Are Getting Bricked, Possibly Due To Driver, BIOS or PCIe Issues

Users of NVIDIA's GeForce RTX 5090 & 5090D GPUs face many issues, with some cards even getting bricked with the latest drivers.

wccftech.com

And still, the Jensen cock throaters cannot see through the myriad of Nvidia's ineptitude. Ban me for a month please before I really lose my shit.

AMD better not fuck this up. I am so sick of Nvidia doing nothing wrong for most of the population but those in the know see clearly that they are scamming assholes.

OPEN YOUR EYES YOU BLIND FUCKWITS!!!!!!!!!!!!!!!!!!!!!!

Couldn't Agree more lol

 

[AusWolf]

The 50-series is probably one of Nvidia's worst "Paper" launches in a while. Not to mention the major issues people are having with them.

Lots of NVIDIA GeForce RTX 5090 & 5090D GPUs Are Getting Bricked, Possibly Due To Driver, BIOS or PCIe Issues

Users of NVIDIA's GeForce RTX 5090 & 5090D GPUs face many issues, with some cards even getting bricked with the latest drivers.

wccftech.com

That article is a hoax. Nvidia cards are immaculate, and their drivers are perfect, they all get released without errors, doncha' know.

Edit: I shouldn't be laughing at this, but the sheer number of "AMD makes crappy drivers" posts I've seen pop up in the forum (by a select few users I might add) kind of makes me.