AMD Radeon RX 7600

[toyo]

Thank you for retiring the Recommended badge.
There's not much of a difference between this card and the 4060ti lunacy. We used to get 970 cards at around the price between the 4060ti and the 7600.
Neither AMD or Nvidia (or Intel) should receive any good press. Continued criticism, as harsh as possible without falling into being toxic, until they stop trying to swindle the customers, sell *50 cards as *60 and 60 as 70.
"The market has changed" is not an acceptable explanation. Sell your product at lower profit, or (hopefully, if smart) consumers won't buy it and leave it bad reviews.

 

[oxrufiioxo]

6nm is supposed to be the efficiency node, but it's not even any better in terms of perf/Watt.
Is RDNA3 so bad that it can negatively offset the advantages of TSMC 6nm over 7nm?!

Hard to say. Last gen AMD had a decent node advantage and that only meant slightly better efficiency at similar performance. This generation Nvidia seems to have a node advantage and it seems their products are decently more efficient.

Ryzen 6000 laptop used the 6nm node and there seemed to be decent gains so this must just be an issue with RDNA3 scaling down.

 

[Chrispy_]

Well from W1zz first page it does offer several new and or improved features. Whether or not they are used or not does not negate the fact they are there.
“
Since it's based on the latest RDNA3 graphics architecture, the RX 7600 comes with advanced Dual Issue-rate Compute Units with over 17% IPC improvement over the previous RDNA2 CUs, second generation Ray Accelerators with a claimed 50% increase in ray intersection performance; and for the first time on an AMD GPU, hardware acceleration for AI in the form of two AI Accelerator units per CU. RDNA3 also introduces hardware-accelerated AV1 video encoding, and the new Radiance Display Engine, with support for the latest DisplayPort 2.1 and HDMI 2.1a ports, and advanced 12-bpc color formats.
“

I'm with Dr. Dro on this; You are indeed quoting "marketing talk" improvements that don't appear to exist in the real world.

That 17% IPC is 2%
That 50% increase in ray intersection performance achieves zero.
No games give a toss about AI accelerators
AV1 is useful to a few people, but also not relevant to the overwhelming majority of gamers.

Perhaps RDNA3 will age like fine wine, but right now it's a turd that has achieved less then any achitecture before it.

 

[Feyd]

@W1zzard Sentence in conclusion "the RX 7600 non-XT in this review is already the full Navi 31 GPU" that's a typo. You probably meant Navi 33.

 

[Charcharo]

@W1zzard Wizzard, please update the Relative Performance charts in the GPU Database for the 7600 and the 4060 Ti

AMD Radeon RX 7600 Specs

AMD Navi 33, 2655 MHz, 2048 Cores, 128 TMUs, 64 ROPs, 8192 MB GDDR6, 2250 MHz, 128 bit

www.techpowerup.com

 

[Paranoir andando]

It's a restyling of 6600XT, as they do in cars, a 6600XT with new headlights.

 

[evolucion8]

You can't just increase the number of CUs like it's nothing. Execution units account for very little area compared to control/cache/registers, this is why everyone making something like a GPU would much rather increase the number of FPUs and minimize the size of the control logic. If they were to increase the number of CUs to match the same number of executions the chip would probably be so large it would be impossible to even manufacture.

Please stop pretending like you know better than the people who design these things.

You have no idea what are you talking about, what I meant linear, means building a bigger RDNA 2 side of things instead of partition it just to benefit some specific instructions ala Ampere. How about proving me wrong instead of ad-hominem childish attacks?

 

[Mahboi]

Well I think I remember the 6600 MSRP was $330. So this has a new media engine, more performance 2 more years of inflation and MSRP is $270. I don’t see anything to complain about. That’s about $365 CAD which is what some 6600 are going for here. Seems like a win to consumers to me.

The 7600 is a fine card. The problem is that it's budget of the budget (smaller die than rx 6600, TSMC N6 when everyone's on 5...). And right now, at launch, it's not that budget. The price will come down, it'll nicely replace the rx 6600, and it may even go lower than the current $200 6600 price. Should AMD have made it a $250? It would have helped their reviews a lot. I'm not sure what the current $270 will really bring them, especially when they could just release at $250, have a tiny batch, and let prices inflate, just like Ngreedia's been doing for years.

But well, I believe in AMD's engineers. Not in their marketing or salespeople.

 

[Dr. Dro]

Ehh 7600 will be more than enough for CS2.

Yes that's what I said... but even with all of the improvements from Source 2 and the likely impeccable optimization since it's the biggest e-sport around (plus Valve spending an inordinate amount of resources on perfecting it), it's still a more sophisticated engine and programmers aren't magicians, I fully expect the system requirements to rise and have exactly this segment feel it - don't think anyone with a 3090 or 7900 XT is gonna be feeling it regardless.

Perhaps RDNA3 will age like fine wine, but right now it's a turd that has achieved less then any achitecture before it.

The fine wine talk has always struck me as more of a fan pitch than an actual reality. It's true that AMD's early GCN cards aged significantly better to the point that they went a full generation/tier up over time, but AMD stuck to the same foundation from late 2011 (HD 7970 launch) to early 2019 (Radeon VII launch) and kept building there. Indeed, each and every one of these improvements carried over to some extent to all previous generation GCN hardware because they were highly iterative instead of complete replacements or redesigns. But it's very easy to look at this iterative work after the many years it's been in the making and call that fine wine, ignoring that many times you would find yourself with a suboptimal configuration because of driver bugs or inefficiencies that would only be worked out X weeks/months/years into the future.

RDNA doesn't carry that trait, each generation having significant differences between each other - the first generation being significantly downlevel hardware incompatible with DirectX 12 Ultimate (and without supporting the optional software fallback as Nvidia offers with Pascal), or the third generation's chiplet architecture, dual-issue workgroups, overhauled instruction set, etc.

You see what bothers me, I genuinely think that RDNA 3 is exceptionally well architected. I just have no idea why it doesn't deliver. I had a faint hope that this would deliver an extra 20% over the 6650 XT, alleviating my concerns that the 7900 series' relatively lukewarm performance was down to the first-generation chiplet architecture and perhaps some internal bottlenecks through its multiple interconnects and buses... but that didn't occur.

I'll need some time to process all of this, read more, understand it better. I'm really confused right now.

 

[Vya Domus]

means building a bigger RDNA 2 side of things instead of partition it just to benefit some specific instructions ala Ampere. How about proving me wrong instead of ad-hominem childish attacks?

I literally just explained why they designed it the way they did.

My God when will AMD learn and start hiring forum dwellers to design their chips.

 

[oxrufiioxo]

The 7600 is a fine card. The problem is that it's budget of the budget (smaller die than rx 6600, TSMC N6 when everyone's on 5...). And right now, at launch, it's not that budget. The price will come down, it'll nicely replace the rx 6600, and it may even go lower than the current $180. Should AMD have made it a $250? It would have helped their reviews a lot. I'm not sure what the current $270 will really bring them, especially when they could just release at $250, have a tiny batch, and let prices inflate, just like Ngreedia's been doing for years.

But well, I believe in AMD's engineers. Not in their marketing or salespeople.

Given their much smaller size and their much smaller R&D budgets I do give them credit for at least staying somewhat competitive with Nvidia my issue is Nvidia really isn't even trying to be competitive so them releasing somewhat equally bad price to performance alternatives isn't really all that impressive.

 

[Ayhamb99]

What the hell is going on with this generation? Like at this point what even?

Nvidia releasing a reskined 3060 Ti that's basically DOA and in some cases performs worse than the last generation

Now AMD Botches their RX 7600 launch with a last minute panicky MSRP change and the card is barely 4-5% faster than a 6650 XT that costs less, and the 6700/ 6700 XT is selling for like $30-$40 more but has a around a 15/20% performance difference.

This generation from both Nvidia and AMD looks to be more of a dud everytime a new card comes out

 

[RDNAuser]

What the hell is going on with this generation? Like at this point what even?

Nvidia releasing a reskined 3060 Ti that's basically DOA and in some cases performs worse than the last generation

Now AMD Botches their RX 7600 launch with a last minute panicky MSRP change and the card is barely 4-5% faster than a 6650 XT

Next Motorola will join the desktop GPU war for all we know and how weird things are getting.

 

[cellar door]

Great review as always W1zzard, thank you!

 

[oxrufiioxo]

You see what bothers me, I genuinely think that RDNA 3 is exceptionally well architected. I just have no idea why it doesn't deliver. I had a faint hope that this would deliver an extra 20% over the 6650 XT, alleviating my concerns that the 7900 series' relatively lukewarm performance was down to the first-generation chiplet architecture and perhaps some internal bottlenecks through its multiple interconnects and buses... but that didn't occur.

If all the gains in every game where similar to what they are getting in the COD engine all these RDNA3 cards would be amazing.



The problem is this game is the only one this arch seems to scale well in over the 6000 series.

 

[evolucion8]

I literally just explained why they designed it the way they did.

My God when will AMD learn and start hiring forum dwellers to design their chips.

Not really, all that you said was all about control logic and stuff, if you ever bothered to read the whitepaper of RDNA 1 and RDNA 2, the biggest difference was just that RDNA 2 could issue 16 waves per SIMD instead of 20 like RDNA1 and that the registers available for occupancy were upped from 51 to 64, yet everything else including shaders, TMUs and execution resources doubled and the performance also doubled up and more. Stop pretending you know what you are talking about as your lazy effort defending your argument is just more ad-hominem posts.

 

[Pepamami]

AMD Radeon RX 4060 Ti

 

[WannaBeOCer]

Inference hardware will accelerate locomotion, physics, RT denoising, audio to facial animation. We’ve seen multiple game engines already demo this. It’s just a matter of time when a developer utilizes them.

Fine-wine came from AMD having crappy drivers for GCN. Seems like AMD has a gripe on their driver optimizations so I don’t see that being the case. Intel Arc will age like fine wine.

I do see all GPUs with inference hardware starting with the RTX 20, Intel Arc, RDNA3 series improving with new features.

 

[rv8000]

What this proves for the first time since RDNA3 launched, is that RDNA3 is worthless.

We have a near perfect comparison with the 6650XT - the only significant difference being the architecture - and it achieves precisely nothing.

What’s useless about RDNA3? Better raster? Better RT vs last gen? AV1? Small but present efficiency gains?

With your reasoning you can say the same exact thing about ADA. What we have here is another example of hardware at a price where it’s basically not worth buying vs last gen. It just happens that AMD didn’t screw the pooch and end up offering a replacement card at worse performance per dollar; 3060ti to 4060ti is a regression in value like every other ADA card.

AMD also seems to have missed matching efficiency whether that be because internal goals were not met or using 6nm/process node for RDNA3 limited efficiency in the end.

 

[ARF]

What’s useless about RDNA3?

It is in fact useless because it doesn't offer better performance numbers.
Radeon RX 7600 could have been called Radeon RX 6655 XT and nobody would have ever noticed.

Better raster? Better RT vs last gen? AV1? Small but present efficiency gains?

Only AV1 and neither of the others.

 

[Dr. Dro]

If all the gains in every game where similar to what they are getting in the COD engine all these RDNA3 cards would be amazing.

View attachment 297399

The problem is this game is the only one this arch seems to scale well in over the 6000 series.

CoD has been an extreme outlier for RDNA 2/3 for some time now. There must be something with the Nvidia drivers that really doesn't tick with that game. You can see both Ampere and Ada underdeliver running it.

 

[vlad.coolish]

Good card for $200

 

[oxrufiioxo]

CoD has been an extreme outlier for RDNA 2/3 for some time now. There must be something with the Nvidia drivers that really doesn't tick with that game. You can see both Ampere and Ada underdeliver running it.

I agree, but even just looking at the 7600 it goes from barely beating the 6650XT in most games to decently outperforming the 6700XT.

 

[Nkd]

Of course not. It’s getting close to 2 years old. It’s rotted away. But it is a benchmark in time which is still useful for tracking and comparison purposes for release vs release. Obviously time changes everything. I mean you can get a new GT1030 for $100 if you wanted to.

You probably wont be able to buy RDNA2 in few months once they have the entire line up. That is one of the only reason it looks bad. I think 6600 probably would have been discontinued if it wasn't for bad economics and low demand. Likely why the pushed the 7600 later but still first and have no 7700 or 7800 series.

 

[W1zzard]

@W1zzard Sentence in conclusion "the RX 7600 non-XT in this review is already the full Navi 31 GPU" that's a typo. You probably meant Navi 33.

fixed, thank you!