AMD Radeon RX 7900 XTX

[MxPhenom 216]

Faster than a 4080 for less, and the raytracing is plenty good enough in the games where it doesn't murder framerate. It's about where we expected it, given that AMD's reveal last month was always going to be cherry-picked games. Performance may increase as drivers get updated for the new compiler but I don't think that matters, because if the $999 price is real, the cheapest 4080 is almost $1400 and the XTX is already faster than it today.

For the games where the 7900XTX struggles with raytracing, so do the 4080 and 3090Ti. I'm with @Wizzard here, Raytracing is the future, but for now we're all stuck in the present.

I think 4080 suggested pricing is soft from Nvidia, as they wait to see where chips fall once AMD releases theirs. So, I am not putting a whole lot of weight in the idea that the 7900XTX is cheaper, maybe for its first week, assuming any of the rumors that Nvidia is planning price drops later this month is real.

 

[Opex]

See ya in 5-7 years GPU companies. Maybe by then you'll come up with something that's actually worth getting for the mid/higher-mid tier users.

So much hype and shitting on NVIDIA and in the end there's no mic drop. Just a "standard" mid/higher-mid tier card with bloated prices/margins.

 

[TheinsanegamerN]

Nvidia just 2 years ago released the 1030 with DRR4 ram under the same name and everything as the 1030 GDDR5! They just did the same just like 3 weeks ago with the RTX 3060 8GB, its a cut down card from the actual 3060, but you don't have that info anywhere! They are literally selling a 3050 with the same name as a 3060!

Should I mention the dirty tactics they play on reviewers and how they try and bully them into positive reviews? Maybe we should talk about all of their naming schemes and even undercutting their AIB partners for more profits!

How about right now and their issues with the 16pin connector melting?

How about when their driver bricked people GPU's? Or the driver that had so many people unable to boot into windows?

Again ALL of the Nvidia crap and BS and incompetence or greed or scams and whatnot never gets brought up! Nvidia is literally 10x times worse than AMD in terms of bad drivers, bad products, misleading products, scams, false narratives, false advertisement, anti consumer fraud, monopolistic tendencies, etc...

And there it is, right on que! Pure whataboutism to justify AMD releasing a upper mid range part for halo tier pricing.

Pro tip: Copium does not a competitive GPU brand make, AMD needs to be competitive to make a profit, you cannot charge premium prices for second tier products.

And UR utter BS about $200 7900XTX is actually mentally insane, even if you measure JUST the production cost of the 7900xtx or RTX4090 just those are over $300, just the production cost! Do you forget about R&D, testing, marketing, transportation, drivers and features development, packaging, paying royalties to various groups and consortiums, profit margin leeway for AIB partners, etc...

You can NOT claim that AMD or Nvidia can sell their cards at $250, the price for 4nm and 5nm process node is extremely expensive, the nodes are new and therefore more error prone so that means less GPU's comes from a single wafer compared to a more mature process, the GPU's are generally bigger to make, etc...

Bruh, BRUH



Did you just take that $250 joke seriously LMFAO

Literally everything you said in your post is complete and utter FUD!

Legend says if you say FUD five times the crypto bros will visit you in your sleep and give you a free JPEG.

 

[wolf]

'Unacceptable RT performance' that's quite something. First of all, I guess you consider all 3000-series cards 'unacceptable' too in RT since they offer similar performance there.

Well that bar was set 2 years ago, we expected more 2 years later, strange.

Nvidia didn't improve on RT performance

Well when you dig a little deeper, it's plainly evident they did, like in @birdie 's subsequent post.

Wait, arent you the same "birdie" that is always trashing AMD and kissing nvidias behind at Phoronix?

Wait, aren't you the same "Neo Morpheus" that is always trashing Nvidia and kissing AMD's behind at Techspot?

Welcome to Techpowerup!

 

[AusWolf]

Well that bar was set 2 years ago, we expected more 2 years later, strange.

Well when you dig a little deeper, it's plainly evident they did, like in @birdie 's subsequent post.

If the previous gen was let's say, 40% slower with RT enabled vs disabled, and the current gen is also 40% slower, then there was no improvement, was there? From this point of view, I see no improvement on the RT front with either AMD or Nvidia. Increasing the number of RT cores together with the number of more traditional shader cores is not an improvement.

 

[Dimitriman]

I wish AMD would put more effort into the RT and creator side of things. This is where Nvidia commands it's premium, and without competition there, prices for both companies will slot on top of each other rather than compete directly.

It would have been awesome if one of the chiplets was a dedicated silicon just for RT and multimedia, call it something cool like "the reality engine" and make it so good that it actually beat Nvidia's offering.

Right now the GPU industry feels just Meh..
- monster cards that won't fit you PC and will break your mobo
- no level playing field in RT and creator arena
- 600w power cables with incinerate function
- min $1k after tax for anything new
- focusing resources for upscaling and fake frames rather than RT and raw power

In the end, I feel like Nvidia and AMD are having a private party but I was not invited. That kinda summarizes my vibe with these launches.

 

[DemonicRyzen666]

If the previous gen was let's say, 40% slower with RT enabled vs disabled, and the current gen is also 40% slower, then there was no improvement, was there? From this point of view, I see no improvement on the RT front with either AMD or Nvidia. Increasing the number of RT cores together with the number of more traditional shader cores is not an improvement.

It varies on this thing on somes times on games it's 3-5% fasters on others it's 1-3% slower on games.

Doubling the instruction rate to two instructions pre-clock to, double what RNDA2 was, is pointless. If they doubled the shaders too. They left a single lone RT core with the need for a lot more instructions with the new BHV trasveral stuff. they just left it lone with barely any more instructions. The shaders sucked up all that extra instructions pre-clock, seems like wasted instructions.

 

[AusWolf]

I wish AMD would put more effort into the RT and creator side of things. This is where Nvidia commands it's premium, and without competition there, prices for both companies will slot on top of each other rather than compete directly.

It would have been awesome if one of the chiplets was a dedicated silicon just for RT and multimedia, call it something cool like "the reality engine" and make it so good that it actually beat Nvidia's offering.

Right now the GPU industry feels just Meh..
- monster cards that won't fit you PC and will break your mobo
- no level playing field in RT and creator arena
- 600w power cables with incinerate function
- min $1k after tax for anything new
- focusing resources for upscaling and fake frames rather than RT and raw power

In the end, I feel like Nvidia and AMD are having a private party but I was not invited. That kinda summarizes my vibe with these launches.

I completely agree. The only thing you left out is Nvidia's complete disregard of the entry level, and AMD's "okay, fine, here's something, just leave me alone" edition 6400.

It varies on this thing on somes times on games it's 3-5% fasters on others it's 1-3% slower on games.

Doubling the instruction rate to two instructions pre-clock to, double what RNDA2 was, is pointless. If they doubled the shaders too. They left a single lone RT core with the need for a lot more instructions with the new BHV trasveral stuff. they just left it lone with barely any more instructions. The shaders sucked up all that extra instructions pre-clock, seems like wasted instructions.

Exactly. Meanwhile, Nvidia has the same-looking shader cores with the same-looking RT cores as before. One step forward and one step back from AMD, no change from Nvidia.

It seems I made a good choice buying the 6750 XT a couple weeks ago. This generation is pretty lacklustre from both companies. Okay, we have a bit more performance, but so what?

 

[wolf]

If the previous gen was let's say, 40% slower with RT enabled vs disabled, and the current gen is also 40% slower, then there was no improvement, was there?

If that were what was happening, sure there would be no improvement. And in some games, that's what we see, but in others, especially newer ones and certainly Portal RTX, the difference in RT perf outstrips the increase in Raster Perf, Ada absolutely has stronger RT than Ampere when and if the extra capability can be/is leveraged. W1zards testing suite doesn't show much of this, but examples are out there.

 

[AusWolf]

If that were what was happening, sure there would be no improvement. And in some games, that's what we see, but in others, especially newer ones and certainly Portal RTX, the difference in RT perf outstrips the increase in Raster Perf, Ada absolutely has stronger RT than Ampere when and if the extra capability can be/is leveraged. W1zards testing suite doesn't show much of this, but examples are out there.

Except that Portal RT isn't a "game" in the traditional sense. It's a tech demo made by Nvidia for Nvidia cards.

The other examples in the post are purely RT benchmarks measured against Ampere. That's not how you measure the improvement in RT.

1. You need a game.
2. You measure performance with RT on, then with RT off.
3. You take the difference in percentage.
4. You compare that percentage with the last generation.

 

[Icon Charlie]

@W1zzard. First off I want to say thank you so very much on revue on both video cards. It confirms that THIS is the web site for tech. Not Youtube and their questionable vbloggers (You all know where I stand on almost all of the so called techies so I won't go there right now). As stated before. Silicon Valley is in my back yard. I see these people who do and create things.

The information that I gathered before your review is close on what you have reported. You really work hard to give us reliable information. Heh. I should know. I was a technology reporter doing similar stuff 20+ years ago. So I kind of know how important and at times stressful to make sure your data is correct and helpful to us... The Masses.

In this day and age of "talking heads", questionable reporting on just about everything... Gimmicks to get people to buy garbage, it is so damned good that I can come to a site to get RELIABLE information on something.

I thank you for your hard work and I hope others take the time to say thanks as well.

 

[wolf]

Except that Portal RT isn't a "game" in the traditional sense. It's a tech demo made by Nvidia for Nvidia cards.

The other examples in the post are purely RT benchmarks measured against Ampere. That's not how you measure the improvement in RT.

From where I'm sitting, it's precisely a way to measure improvement to RT performance. And I disagree with dismissing those results which frankly are valid in context.

For arguments sake, lets try your way, there are games where it's a wash and broadly equal to Ampere, and there are ones that show a reasonable difference, even without alterations to game code to leverage Ada's improvements like Portal RTX does, at least they haven't said so iirc.

AMD Radeon RX 7900 XTX Review

The Radeon RX 7900 XTX is a pretty good GPU, at least relative to its GeForce competitor, but whether or not it's worth $1,000 will depend on...

www.techspot.com

Marvels guardians of the galaxy
3090 has 48.4% of RT off perf with RT on
4090 has 52.4% of RT off perf with RT on ~8% better result than Ampere

Cyberpunk 2077
3090 has 45.5% of RT off perf with RT on
4090 has 54.2% of RT off perf with RT on ~19% better result than Ampere

 

[systemBuilder]

I wish AMD would put more effort into the RT and creator side of things. This is where Nvidia commands it's premium, and without competition there, prices for both companies will slot on top of each other rather than compete directly.

It would have been awesome if one of the chiplets was a dedicated silicon just for RT and multimedia, call it something cool like "the reality engine" and make it so good that it actually beat Nvidia's offering.

Right now the GPU industry feels just Meh..
- monster cards that won't fit you PC and will break your mobo
- no level playing field in RT and creator arena
- 600w power cables with incinerate function
- min $1k after tax for anything new
- focusing resources for upscaling and fake frames rather than RT and raw power

In the end, I feel like Nvidia and AMD are having a private party but I was not invited. That kinda summarizes my vibe with these launches.

I think your reality has been distorted by the "news-hype" cycle.
It's true that the only high-end cards with reasonable power consumption and size are 3080, 7900xt, 7900xtx, and maybe the 6900xt.
But if you freeze frame rates to 100 fps then a 4080 only uses 170 watts of power, which is extremely good, and only 6800xt can compete on "Quiet".
It may be a gigantic oversized 3.5-slot lump in your computer, but joules per frame rendered on 4080 is second to NONE.
And RT really doesn't matter. Its just a FUD-hype technique by Nvidia.
Ever hear of ANYONE making a kill shot due to a puddle reflection? No? Of course not!
NVENC and OpenCL matter, and AMD is finally matching NVidia on OpenCL with the 7900 XTX.

I have bought 2 high-performance video cards this year (3080, 3070 TI), and both times it was hell and exasperation.
One was from the newegg shuffle during card rationing, and I won a much-hated 3070 TI for ONLY 50% over MSRP.
One was 30d before xmas when I won a NEW 3080 on amazon for ONLY 15% over MSRP.
These companies have to stop lying with lowball/fake MSRP prices, I think the FTC should step in.

Listen, both NVidia and AMD are now using the same semiconductor foundries. Since NVidia has the engineers from SGI and 3dfx that INVENTED the 3D gaming industry, did you expect AMD to beat them on price AND performance? Really? Now that EVERYONE is using the same chip factories, the designs are probably not that different any more! Going forward, you get what you pay for. A 30TFlops 3080 for $800 is a screaming deal compared to a 5TFlops rx580 costing $300 from just 3-4 years ago. I know, because I have bought both of them. The 30TFlops card is 6x faster, for 2.67x the $$$. It may not following moore's law but hopefully some competition from Intel will help (Although I feel like Raj Koduri is an idiot and AMD is better off without him, it was a brilliant move to dump him on Intel graphics division ...)

 

[Arkz]

Nice to see improvements but they've been working on this for a few years and it comes out of the door with 100w idle on dual monitors... The hell? Not as big of an improvement on the 6900XT as I was hoping, and RT is still a fair bit behind the current NV, but at least they're closing the gap. At the moment though all prices are just daft. I thought my £660 3080 was expensive when I got it but compared to todays price for performance 2 years later, maybe it wasn't that bad. Not like I plan on upgrading any time soon anyway, and the way prices are going I'll probably still be using this 3080 in 5 years.

 

[AusWolf]

From where I'm sitting, it's precisely a way to measure improvement to RT performance. And I disagree with dismissing those results which frankly are valid in context.

Those results measure performance in general, not the improvement in RT core performance specifically. Of course they're valid, just not in this context. You can't cram 60% more RT cores onto a GPU die and brag about your RT cores being improved, because they're not - there's just more of them.

For arguments sake, lets try your way, there are games where it's a wash and broadly equal to Ampere, and there are ones that show a reasonable difference, even without alterations to game code to leverage Ada's improvements like Portal RTX does, at least they haven't said so iirc.

AMD Radeon RX 7900 XTX Review

The Radeon RX 7900 XTX is a pretty good GPU, at least relative to its GeForce competitor, but whether or not it's worth $1,000 will depend on...

www.techspot.com

Marvels guardians of the galaxy
3090 has 48.4% of RT off perf with RT on
4090 has 52.4% of RT off perf with RT on ~8% better result than Ampere

Cyberpunk 2077
3090 has 45.5% of RT off perf with RT on
4090 has 54.2% of RT off perf with RT on ~19% better result than Ampere

A good point. I guess there is some improvement in some cases. Just not enough (improvement and cases), imo.

 

[Richards]

Disgusting how the 5800x bottle necks the 7900 xtx and 4080.. 5800x should never been used

 

[Fluffmeister]

Disgusting how the 5800x bottle necks the 7900 xtx and 4080.. 5800x should never been used

Yeah, bloody AMD!

 

[wolf]

Those results measure performance in general, not the improvement in RT core performance specifically. Of course they're valid, just not in this context.

I still disagree, it's measuring like for like products in like for like testing, when we know in rasterization the 4090 is on average about 60-70% faster than a 3090, but in RT testing it's ~2x ish faster. I'm happy to leave it there though, if you think those results lack merit in this context, that is your prerogative.

A good point. I guess there is some improvement in some cases. Just not enough (improvement and cases), imo.

I agree it's not as large as I would have hoped, but I get the impression lessening that hit is quite the uphill battle for both camps. Gotta inch forward every time.

 

[evernessince]

Looks like another generation I'll be keeping my 1080 Ti. Neither AMD nor Nvidia's cards are competing on price. At this point that's painfully clear. Instead AMD and Nvidia price around each other, ensuring that they both extract maximum profits while customers are left wondering how much they'll have to spend.

I didn't watch the video but that title is terrible clickbait.

 

[shovenose]

Alright, it's been decided. I put in an offer on a used Dell OEM 6900XT on eBay. I want that specific card because it's a true dual slot card so I can retain my PCI-E X1 front panel USB C adapter. If that offer is accepted by tomorrow morning when the 7900 series launches, great, I got a 6900XT. If not, I will retract my offer, and buy the 7900XTX, yay for credit cards. Too bad for my front panel USB C port at that point LOL. Like I commented on the XT review, I honestly really wanted a 7900XT but the price/performance is a joke compared to the XTX. If it was 799 or 849 I would have gone for it, but you lose over 15% performance for a 10% discount which is NOT cool with me.

 

[konga]

There's something weird going on with the 4090 results in this review. It is getting lower frame rates at 4K in a lot of games than it got with a 5800X in the original 4090 review. What's the explanation for this?

 

[gmn17]

I'm going to wait for better drivers and 3dv cache version of this GPU hopefully out by NH summer 2023

 

[TheinsanegamerN]

I completely agree. The only thing you left out is Nvidia's complete disregard of the entry level, and AMD's "okay, fine, here's something, just leave me alone" edition 6400.


Exactly. Meanwhile, Nvidia has the same-looking shader cores with the same-looking RT cores as before. One step forward and one step back from AMD, no change from Nvidia.

It seems I made a good choice buying the 6750 XT a couple weeks ago. This generation is pretty lacklustre from both companies. Okay, we have a bit more performance, but so what?

Oh God, don't get me started on the total lack of low end stuff. I went with a rx 560lp years ago instead of a 1050ti, and felt the 1650 wasn't enough of an upgrade.

What do I get? No cards from Nvidia that fit in the 75w tdp required for low profile cards, Intel's ARC a380 being a mess and requiring bios settings my old optiplex media machine doesn't have, and AMD throwing out the laughingstock rx 6400 when they have mobile gpus like the 6650m or 6700s that would have done the job FAR better as 75w desktop parts.

My only hope now is some Chinese brand takes aforementioned mobile GPUs and makes a low profile model out of them

 

[wolf]

What do I get? No cards from Nvidia that fit in the 75w tdp required for low profile cards

Well, there was one option for those that must have LP, but you'd have needed relatively deep pockets for one, the RTX A2000. Absolutely though they could have made a 3050LE or something LP with a 'sane' price tag, they just chose not to.

Interestingly this generation I feel like Laptop GPU's are going to be really potent from both camps given the levels of efficiency on display, I reallllly hope that means at least one new LP card from both camps that raises the bar significantly over the GTX1650/RX6400

 

[Space Lynx]

I'm going to wait for better drivers and 3dv cache version of this GPU hopefully out by NH summer 2023

I never thought of that before, is that actually going to be a thing? 3d cache like on 5800x3d will be on gpu's?