RDNA 3 - AMD's Path To Victory 

MLID is garbage, he'll make 40 different claims on something and eventually claim he was right months later ignoring all the wrong stuff

Feel that. I missed Jim and now I can enjoy both Moore's Law is Dead and AdoredTV. I can now die happy.

@Boinz He does do that but he also gets amazing interviews with people who reveal things about the tech industry that you never see coverage anywhere else.

Agreed mostly, but personally I prefer Jim’s brand of humility, deep dives, and analysis. As OP says, there’s no substitute

@Boinz yeah that’s my biggest problem - the massive Ego and CONSTANT denigration of the ever-nebulous “some people” that are never identified, along with the suggestion that I, as a viewer, am retarded if I didn’t already know something because “OMFG dude I totally talked about this 1.25 yrs ago what the fuck is wrong with you don’t you watch every minute of everything I say???” @pspublic - yup lol this is true as well no doubt - although again he frequently let his ego invade his interviews… I don’t think anybody being interviewed gives a shit what he called and when he called it lol, but that’s the paradoxical nature of the channel for me… good content but jfc I just wish he’d jerk his chain OFF-camera a tiny bit more lol

And I can't understand why! Because this is good competition at a different price point, will offer alternatives to overpriced green team. It's like people wanted so much that AMD would "beat" Nvidia that they're disappointed and blame AMD for their self hyping. As a sff owner, I could fit a xtx in my rig, not have to change PSU, case, anything, and benefit immediately. I can't do that with Nvidia. Huge value for me. It's a plug and play replacement instead of having to rebuild everything around the new GPU. Really can't understand the fuss.

@@liberteus Totally agree, I will mostlikely change my RTX3090 for the RX7900XTX. It is a great value GPU!

@@liberteus I think leakers are partially to blame and peoples inability to take any unofficial claims on the internet with a grain of salt. I see people in the comments in videos quoting leaker statements as if they are official because a particular leaker got something right in the past and many things wrong. I see what looks to be a pretty good generational improvement from AMD but people bought into the two and half times faster narrative and now they are upset.

@@liberteus Same here, got a small form factor case. I could fit the XTX in it and I think my 15mm fans would just and just fit over it still as well. I did not expect that.

How is this divisive? This is better than what Nvidia is offering all day and night.

A card that's about 85ish% the performance for $600 dollars less, it's a bargain for that.

That's why I don't understand what the fuss is about. 600USD buys a motherboard and cpu, along with ram. Instead of buying higher end Nvidia I'm going to buy parts, retire my current GPU and build a new rig for my kids. Or buy an IPS ultrawide screen. Something to make my whole setup better, not just fps that anyways are already way above 100 in any game in 4k with such monsters.

If AMD wants to make a 450-500 watt GPU than it would kill the 4090, but thats not the direction they want to take

@@liberteus The tiny pee-pee NVidia fanboys are making any kind excuse they can to justify paying 60% more on top of the price of 7900XTX to get that monstrosity that is the 4090.

@@ssaini5028 Yeah, AMD would've made a such GPU monster with big power draw, if they would bring it to against 4090

or 7990XTX, they could then name the 7900XT refresh 7950XT

7970 3 ghz edition

That would be EPYC! :)

Ditch the numbers and call it the " Radeon BFG XT" Just to troll Nvidia even harder

7970 XTX would be such a nostalgic throwback. I'd ride that hype train.

Adored shows his working 😉

besides gamers nexus and hardware unboxed, every other tech youtubers doesnt know what they are talking about, its just entertainment, same with most audio/keyboard or anything neighboring computers

I seem to recall that faceless Kopite7kimi clown and his Twitter post in May: "I am disappointed with RDNA3. That’s all." Everyone actually took that way too seriously and expected RDNA3 to be disappointing....

@@jimtekkit they just reported on whatever information arrived, the clowns are the people which took it as fact

there are a few others for sure. I'm a Mac for example, great channel

@@LunaWaves But I wouldn't waste my time with Moore's brain is dead

Moores Law is Dead

How is this a GPU channel? Jim talks about a whole range of stuff.

@@tuckerhiggins4336 Good joke

I want an FPGA in these cards just in case encoders can be "upgraded" in the future.. that'd kick so much ass.

me too :D i see jims vid - grabing popcorn - enjoying it!

Hehe

Especially after Cortex's crazy freakout lol.

its not a matter of personal belief, amd has already stated xtx is a 4080 competitor

What is this madness of xtx being 4090 competitor @$1000 when amd itself said its a 4080 competitor

There's a 7950xt.

IMO Coreteks deserved the roasting he got over his highly critical video of RDNA 3. It was poorly considered and completely off the mark in more than one way. He's not considering that the XTX is not a 4090 competitor or considering other issues such as pricing, energy use, and market placement, and has doubled down on his highly critical view by posting a very long pinned response to all the criticism and dislikes he's been getting from his own fans. Edit: the other thing Coreteks seems to have missed, is that unlike Nvidia, AMD continues to rely on 3rd party card makers (AIBs) to sell the majority of their GPUs. AMD had released reference cards that is not intended to compete with AIBs, but Coreteks seems to have forgtten this. For certain, we'll see 3rd party cards with higher clock speeds that push performance further up.

@@geekinasuit8333 Lol, yes AMD rely on AIB to make faster card, at much higher prices, so the 1000usd MSRP is just as fake as when rx6000 launched, more likely AIB cards will be selling at 1200usd Nvidia forced the AIBs to sell lots of SKUs at MSRP since their FE is so good.

you could always sell it and put that money towards the new thing. New hardware is incredibly enticing, I have an RTX 3080 and am incredibly excited to see how the 7000 series looks once we get reviews in.

Sadly? That's a great card, enjoy every moment, many including myself would love to have one of those badboys.

I'll happily take your sadness. My vega 56 is tragic in comparison.

@@makavelismith hey now, the Vega 56 was my last Radeon card and it was super fun to under volt and over clock. I had a good few years with that card.

@@SebastianE18 Ya, it's a decent card but it is a bit under-powered for my 1440 144hz monitor for a lot of games. I am thinking I may grab a 6800XT soon and keep this as a backup, in case something goes wrong.

You are a great friend! :)

i recommend waiting a bit with the xtx partner cards, & reviews so you can make a well informed deccission.

If the GCD is NOT bandwidth starved, they need the MLiD source tipped larger GCD to face 4090 Ti full AD102 die. As is winning raster by a little won't cover for RT losses

@SoundwaveSinus9 They won't do a mid-gen refresh and design a new gigantic die to only serve as the very high-end option at the $500 price segment. It's ridiculous. This prediction is just more empty hopes and dreams for poor AMD fans who just can't get enough of embarrasment on review day.

I meant 1500 dollar segment.

I agree with you, that AMD will probably not introduce a larger Chip in this Generation. What I could imagine is a experimental 7950/7970 XTX with two Navi 32 dies and maybe 3DCache. It could be a pipe cleaner for the technology which is currently only used to have the memory subsystem offloaded to the chiplets. AMD seems to be doing this a lot: making a single product as experimental pipe cleaner and to get some experience in the field before they go out and base a whole generation on the new tech. The 5800X3D comes to mind, also the Multi-Die steps from Epyc/Threadripper with Zen 1(+) and then going all in with chiplets with Zen2. The MCM Packages of first gen Ryzen gave them experience in the field, to determine specific issues, which makes it easier to sort them out for further, more elaborate designs. Zen 3 only got one consumer product for 3DVCache and a server lineup to test the field and they are now expected, and probably will bring it, to release a wider portfolio of VCache products with Zen4. Therefore, it would fit the overall strategy to bring a dual-compute-chiplet design with the smaller Navi32 die to test the waters in the field and sort out issues for future generations which will probably emphasize on the multi compute chip design. Navi 32 is expected at around 200mm² with 64 CUs. A 2x package would give them 128 CUs with 8192 Shaders. N31 has 96CU/6144 Shaders. This would give them ~30% more shaders to work with on 2 smaller, maybe higher clocking dies. If the connections between the chips and scheduling of workloads goes well, we could very well be getting a 20-25% or even more (depending on clock speeds) increase in performance over the 7900XTX. That's IF the design works relatively well from the start. I wouldn't rule it out but I definitely don't expect to see a 500mm² compute-chip in this generation.

@@VPGTube Hmm, interesting thoughts. It would be interesting to see if AMD could finally be able to solve the multi-GPU problem for good. The only architecture that achieved close to perfect scaling (well, at least in some situations, while in most others it was anywhere between 40 and 70%) was 3dfx VSA-100 (Voodoo5 series) back at the dawn of the millennium. :)

Incredible that Zen 4/Raptor Lake appear to have been GPU bottlenecked when they launched, and graphics cards immediately go back to being CPU bottlenecked. Everything's advancing rapidly. Great time to be in the space.

3 years ago you wouldn't expect a joke BFGPU at $1,500 to sell well to gamers. To me the Nvidia high end is ridiculous due to rebuild requirement, I'm happy with 1440p. It seems perfectly plausible that the expected 7800xt became a 7900xt with the xtx intended against a 4080Ti priced at $1,200.

@@RobBCactive Considering how long I've been tormenting my 2070 with my 4k screen (gpu performance does not meet the demand), I'll probably get one of these new AMD cards and be good for the next 5 years :D

@SoundwaveSinus9 lol, the 7900XTX will cost 1400 euros. People with that money to spend on hardware do not care about the slightly increased electricity bill caused by a powerful GPU.

@SoundwaveSinus9 I imagine reference cards will be 1250 and then AIB cards around the 1400€ mark.

I think so, partner could have 3x 8pins and push the limits to best the 4080 in performance and price. No disrespect, lol

@@Magjee In the end AMD is a "boring execution machine" in that they want to sell chips, not boards as chips are where the margins are. AMD I think is clearly willing to bin the best chips for partners to make margin because by selling the lower bin chips themselves they are still making money and it's not about the ego of "best flagship" in referfence design like NVIDIA. Winning a war is not about victory all the time, it's about choosing and winning the right vicories. Time and time again the best tech companies have lost to rivals with all around better execution. Drifting from what customers want to empower new markets has helped NVIDIA massively but... there is a break point there. Ray tracing and higher costs are features abliet negative ones for average markets not the professional high end ones, add design flaws and now is an aweful time for NV to srew up.

This is what I think might be happening here, that AMD had a chance to not push their chips in their reference cards giving AIBs a chance to offer something that gives a good bump in power. Still it in impressive that this 7900XTX will be getting near 4090 performance as a starting point with some 3GHz AIB cards as stated using 3 8 pin connectors pushing power up to over 400 watts.

EVGA should go to AMD

Partners have off the record have stated that it is easier to make margin off AMD cards than nvidia cards.

Sell the 6900 xt then buy the next one with the proceeds :D

the 6900xt is gonna eat up every single rasterized game on 1440p for the next 5 years

@@spitfire1789 that sounds reasonable. my vega64 is still going pretty strong. I wont to swap more for vanity than actual *_need_*

Wasnt this discussed in one of the PCWorld videos from the event and confirmed? I might remember wrong.

@@CorporalCookie yeah someone spoke a bit too much in a PCWorld Q&A

heavily doubt it. Iirc the last hardware that you could daily overclock by a ghz was the 2700k. With modern boosting algorithms there are hardly any gains left on the table. Maybe you can just remove the powerlimit with a hack and see how the boost algorithm goes BRRRRR but you better be prepared to supply 700w to that GPU.

​@@invalid8774 I recommend you watch the live hangout video on Level1Techs at 2850 seconds. Jarred spoke a bit too much there, the face and eye movement of Wendell says it all. In the video Jarred speaks about the fact that the cards are designed up to 3Ghz. So there is the possibility for board partners to wring out more performance of it. However everything ties in to the power draw and that will definitely be higher. Anyways exciting times are ahead that is for sure :)

Lovelace can also hit 3GHz. The 4090 can hit 3GHz at under 600 watts. It literally needs more voltage to clock higher and hit the 600 watt 133% power limit.

"What's next, a 5-slot GPU requiring integrated high-end liquid cooling" While it's fun to poke at Nvidia, let's at least be realistic; any card with water cooling is not going to be absurdly massive. They already have 1-slot water cooled 4090s, it's the radiator that would possibly need to increase in size but that's less of an issue (for standard size ATX cases anyway).

Next you stick the motherboard into the GPU.

Nvidia jumped from a budget Samsung node to the same 5nm node AMD is using only it's 'optimized' for their power requirements and given a suspicious name which keeps making people forget its not TSMC's 4nm node. The 4090 is this powerful because it's a monster die with a lot more cache crammed in and given effectively unlimited power headroom (which is very out of character for them).

@@JJFX- I made a acomment based around exactly this. The 4N is pretty much just a name saying it is a cuntom Nvidia node.

@@AlucardArkaine Yeah and it's not unusual for silicon fabs to optimize a manufacturing process for their big clients but it's not a new node. It's unclear exactly why 4N is called that because there's little public information but I think it's safe to say the marketing dept. has something to do with it. I do know Techpowerup changed all the Ada GPU listings to "4 nm" from 5 after the keynote and they don't seem interested in correcting it.

That's like the "4090 Ti Super Ti" I came up with at work. It's the size of a NZXT H510 case and has its own PSU on board, so you can plug it directly into the wall for its dedicated 1000 watt requirement.

That can't be right because it's clearly not enough X's.

XRXX 9990 XXXTZTXX Plus Pro MaXXX Xtreme Edition

That would be 7990XT

@@alaeriia01 nVidia® GeForce™ RTX 4090Ti Super TITAN DragonForce Ultimum XR

AMD fell farther behind this generation. If you can't afford a 4090, then but a 3090. AMD is getting crushed

Same. My last Radeon was a 4870. They just can't put together a decent product, it seems.

@@mattgreenfield8038 what a pathetic baseless comment from an obvious fanboy. None of what you wrote makes any sense

@@mattgreenfield8038 With 7900xtx around corner,why would anyone get a useless 3090? 7900xtx crushes it and the 3090ti and is cheaper.3000 series cards have not really dropped in price unless you want used mining one off ebay or something.Anyone wasting money on 3090 or 3080 with 7900xtx around the corner is a fool.

@@johnnyboogalo4897 This AMD thing should be called a 7800XTX. Since it only competes with 4080. Last gen AMD sold 1000$ 6900XT that was actually toe to toe with Nvidias best 1500$ 3090. This gen, AMD still wants their 1000$, but cant compete with 4090. Only an AMD fanboy would think this was good. Nvidia is curb stomping this chiplet disaster AMD made this gen. What a disappointment for AMD. Coreteks called it best.

I'm hoping for a 7990 XTX like the old 7990 dual gpu card.

Them were the good days they should bring back ruby

hell ya

Somehow I doubt that. Do you have a link?

@@invalid8774 I would if I could find it again

>upgrading $1000 video card to new $1000 video card >worried about budget for a new PSU

@@randommcranderson5155 Not warried, just looking at better options, so I either stick with what I have or see what's coming from AMD, the 4090 would also cover my 2nd PCIe slot which I will use for more NVME storage since its a x8 PCIE 4.0 slot. So its more than the 750watt PSU, I can also sell my 3080Ti to a friend if I want to. I'm not desperate just tempted by a similarly priced, sized and lower power consumption GPU which may potentially give me 40% more rasterized performance and I'm already on a 5800x3D (B550) so SAM would get a decent boost as well. I may have been worried if I didn't have the below: 3080Ti Time alternative options a 5800x3D on a B550 board Already maxed out my platform with 32GB RAM I'm just tempted by the 7900XTX more than a 4090 and if I buy, it would probably not be on launch day but closer to a year afterwards. Which is normally my buying schedule since I wait for drivers to mature and for me to feel like I actually NEED it since for now its just a want.

@@dawienel1142 thats a lot of words to say you don't really have a budget problem and can afford a new power supply for your new $1000 video card replacing your last $1000 video card.

@@randommcranderson5155 But Nvidia is at least 60% pricer at the start. Add to it PSU and probably Nvidia costs you 80% more than 7900xtx. I don't see why you are confused about it

@@randommcranderson5155 Having the budget and feeling the need to do something is very different, I just don't feel like spending another $200+ on a 1000watt PSU.

Im still cautiously optimistic the 3ghz might only be possible on the lower monolthic models dont know yet if the clocks was limited by this much below 3ghz due to the chiplets or if its just purely for power savings

@@sbrader97 IF LTT are right about the extra 8 pin PCIE connector then there's possibly a 40% boost in available power. In PCworld vid with Frank Azor, Frank flat out stated "The reference 7900XTX cards will be the lowest perfoming cards in the RDNA 3 lineup". For reference Nvidia's 4090 founders ed is 1fps slower than the AIB models and 3-5 fps slower with a 33% power increase in MSI afterburner, HUB did some tests showing this and also that the power (actual usage) went from ~430W to just 508W and ACTUAL increase of just 16% NOT the requested 33%. So as I mentioned above and the leaked slide showed IF RDNA3 WAS designed for 3GHz+ then a 40% increase in available power could WELL see the 7900XTX hit 3GHz or more.. Bear in mind as well the 4090 CAN do just above 3GHz and it's got 16K cuda cores not 13K... Just sayin....

@@SRS13Rastus The 4090 can draw upto 600w but with nvidias voltage limit it doesnt hit the tdp as the limiting factor so hopefully amd doesnt screw overclocking with limitations like that there is confirmed aib model with 3x 8 pin but at this moment its speculation if it will mean they can scale to 3ghz+ with that power or if theres further limitations to the clocks like the chiplets just unknown right now what the limit is. Hopefully they allow control of both the shader and front end clocks too

You're letting product monikers fool your sense of pricing. The 6900XT was the full die. The 7900XTX is the full die. BUT.. you get 8GB VRAM more, so the pricing has indeed improved beyond generational performance gain.

For Nvidia to match a theoretical 500mm2 GCD from AMD with the same lead they have now, they would need something like a 1000mm2 die, which they cannot make and would be ridiculusly expensive if they could. If AMD really does make massive GCDs, then Nvidia cannot compete either in performence or price. AMD will just brute force stuff with more silicone then Nvidia can

It comes down to shared resources, and AMD website designers either not understanding the architecture or AMD wanting to win with a smaller number. It has 12288 shaders based on the definition of shaders since unified shaders were a thing. Half of the shaders are fed with a Thread Level Parallelism (TLP) approach. This means spamming threads at the shaders until they are well utilized. Some threads are complex and could use more resources. With these threads an Instruction Level Parallelism (ILP) approach is taken to fill the other half of shaders. ILP breaks down the thread into the instructions generating “mini-threads” (instructions hence the name) to fill the execution block. Fermi below the flagships GPUs and Kepler from Nvidia used an ILP approach to feed 1/3 of their shaders. These GPUs were well utilized. AMD with VLIW5 (HD 3000-5000 and some 6000/7000) would use ILP across 5 shaders getting an average utilization of 3.4. AMD likely saw Nvidia getting a lot of crap for doubling shaders in Ampere and getting nowhere near that performance increase. This could explain why the engineers cut the shaders in half if the website is intentionally mislabeled.

@@dex6316 I get what you're saying, but AMD's slides even showed 64 "Dual Issue SIMD" per CU - which again, would add up to 6144. AMD's website still has 6144 Sp with 384 TMU listed, same with Techpowerup. IF AMD had doubled the shader count equal to the TMU count, it would have been 768 TMU. AMD mentions "Dual Issue SIMD" many times - all of this adds up to 6144 being the actual number, but they seem capable of 2x throughput per clock in some manner that has yet to be understood. Also, to go from 5120 Sp to 12,288 Sp with ONLY a ~1.6x improvement overall and ~1.5x in RT, well then that means RT perf is not even equal to the number of (legacy) CUs the 7900 XTX has increased from the 6950 XT. (But the perf per watt increase is still insane regardless) At 96 full "legacy" CU the RT uplift is much more impressive. Regardless, lots of very interesting details for any true tech nerd. Could be you are right, and I am wrong. But either way, I want to know how they got those gains in FP32 and more. I still NEED to know what AMD means by "Dual Issue SIMD" but it really seems to point to 6144 shaders at 2x clock frequency.

​@blue-lu3izwell there are rumors that there won't be high end RDNA4, Jim should not be making BS video like this anymore LOL, make him look extremely stupid

In what way?

@@adoredtv aged like fine milk, RDNA4 will compete with 4090Ti

I literally said 20% slower than the 4090, it's around 25% slower...

@@adoredtv when choosing between 100,000 car vs 110,000 car price shouldn't be a factor . Competitive price only matters at 200-500$

@@rx58000 So you agree the 7900 XTX is better than the 4080?

@@adoredtv if 7900 xtx was ~54% faster than 6950 xt then I would have said yes . But the XTX is Whitney plastic card with worse power efficiency . This card is making 4080 look good

@@rx58000 Sure, so long as you're content with the slower card.

If you want a GPU that's not for gaming, they want you to get one of their CDNA cards instead.

Yeah agree, It's always next time, never now.

How does AMD's RT help in workstation tasks? Nvidia has optix rendering with their RT cores, but AMD's are useless, so why emphasise them?

For 99% of consumers gpus are only about gaming genius and be ready to cry when it annihilates 4080 😅

​@@Soutar3DG That seems to be the case, yes. AMD's competitors (especially NVIDIA) put more effort into supporting general-purpose computing on their consumer-grade graphics cards. Even if individual users like me wanted to get CDNA cards for our own desktops or home labs, they don't seem to be sold for that purpose at any retailer I'm aware of.

Devour the market share and performance crown will become irrelevant.