AMD roadmaps reveal Zen 4 and RDNA 3 details, promise significant performance increases

midian182

Posts: 9,763   +121
Staff member
Highly anticipated: AMD's Financial Analyst Day 2022 has just taken place, and with it came a slew of information about its upcoming and future processors, including Zen 4 and the next-generation RDNA 3 graphics cards. In the case of its CPUs, AMD promises a whopping 35% performance increase over Zen 3, while the GPUs offer a 50% performance per watt uplift.

AMD's desktop CPU roadmap confirms a few facts about Zen 4. Firstly, the company's 3D V-cache technology used in the Ryzen 7 5800X3D will be part of the new processor series, though it never revealed how many chips will utilize it and whether they will launch alongside the new processors. As a reminder, the Ryzen 7 5800X3D stacks a 41mm2 die containing only L3 cache over the Zen 3 core complex die, increasing the amount of L3 cache from 32MB to 96MB.

When AMD discussed Zen 4 at Computex last month, it promised a greater than 15% single-thread performance uplift over Zen 3. It now says overall performance improvement is 35% greater than the current generation. Whether that also refers to single-threaded performance or multi-threaded (or a mix of both) is unclear.

We know increased clock speeds, an area where AMD has long stood in the shadow of Intel, will be a big part of Zen 4's performance boost. The processor it showed off at Computex, believed to be a prototype of the Ryzen 9 7950X, ran at 5.5 GHz on "most" threads when running a Ghostwire Tokyo demo, and the company later confirmed the CPU was not overclocked.

We also see Zen 4c (Bergamo) on the roadmap, which increases the standard maximum Zen 4 core count from 96 to 128, and mention of the 4nm process—an enhanced 5nm node that will be used in the mobile Phoenix Point line that mixes Zen 4 cores, RDNA 3 graphics, and Xilinx's Artificial Intelligence Engine (AIE).

Elsewhere, the roadmaps reveal that the Threadripper family will be getting the Zen 4 treatment. There have been fears that AMD was letting the brand die off, but we don't know if, like Zen 3, only the Threadripper Pro series will be part of the next-gen chips.

As for Zen 5, codenamed Granite Ridge and likely called the Ryzen 8000 series, it will be built on the 4nm/3nm processes, use the same AM5 socket as Zen 4, and is set to arrive in 2024.

Jumping to AMD's graphics cards and RDNA 3, the highlight statistic is that the architecture will offer a 50% performance per watt uplift over RDNA 2, which sounds enticing but should be taken with some salt as it will likely be a best-case scenario. We can still expect the flagships to be power-hungry. Whether they will rival the rumored RTX 4090/4090 Ti's 600W to 800W remains to be seen.

Like Zen 4, RDNA 3 will be built on 5nm process technology from TSMC. It boasts "advanced multimedia capabilities" such as AV1 encode/decode support and comes with DisplayPort 2.0 support to deliver uncompressed 4K gaming at 240Hz thanks to the 80Gbps of bandwidth.

Other new features include a reworked compute unit (CU) and next-generation Infinity Cache. We also have confirmation that RDNA 3 GPUs will return to a chiplet design, something that has been rumored for a while. It will be the first time AMD has used the design in consumer cards since the Vega era.

AMD also confirmed the existence of RDNA 4. It's based on the 3nm process and should launch in the same year as Zen 5: 2024.

The Zen 4 CPUs and RDNA 3 graphics cards are set to launch later this year.

Permalink to story.

 
Why not RDNA 1 to RDNA 2 was 50%+ why can't they do it again.
Oh yeah, I'm all for it. Let's do this. :)

I just hear a lot of people saying that AMD and Intel are both guilty of making claims of higher performance improvements than they can deliver. I guess AMD has entered a magical zone when they are actually backing up most of their claims lately it seems. Good times!
 
Just hoping they don't go in the "we don't care about power" like the rest.
they'll probably go that route also.

at this point, devs should honestly just optimize their applications and games better. we know they can do it because the switch, deck and mobile platforms exists (also look at what the ps4 and xbox are still pulling off) and they'll magically make a game scale down and work on that lowbrow tech quite nicely.

BUT, theres no point to do that when the hardware can just brute force past any jankiness I guess.
 
Why not RDNA 1 to RDNA 2 was 50%+ why can't they do it again.

Because it is not linear and easy.

Anyway it all comes down to reality, FSR & Raytracing performance.

For me the most important development are the iGPUs, I hope along the line to see Steam Deck 2 & Surface Pro 10 with such AMD chips (and not only Intel Intel Intel)
 
Because it is not linear and easy.

Anyway it all comes down to reality, FSR & Raytracing performance.

For me the most important development are the iGPUs, I hope along the line to see Steam Deck 2 & Surface Pro 10 with such AMD chips (and not only Intel Intel Intel)
Never said it was linear and easy just that they did it before and they can probably do it again.

Its not like they are staying on the same node.
 
I won't be buying Zen 4 for a LONG time because my Zen 2 R5-3600X is an almost perfect match in FPS for my 6800 XT. When that becomes too slow, I'll just swap in an R7-5800X3D on the same motherboard for a dramatic gaming performance increase. So, my next upgrade won't need a motherboard or RAM. Thanks to AMD's belief in making platforms last, all us AM4 owners are sitting pretty when it comes time to upgrade.
 
A 50% increase in graphics performance? HOLY !@#$. Is that possible?
It's not like they haven't done even better than that before. RDNA2 (aka "Big Navi" aka RX 6000) had a >100% performance-per-watt uplift over RDNA1 (aka "Navi" aka RX 5000). AMD claimed a 100% generational increase which many said was impossible. So ATi said "hold my beer" and proceeded to over-deliver on the promise that AMD made.
 
f3e22584a3d4efbdf2ee45b7e47f16e54e082497803fc09dc714ef41630a3cd2.png
 
It's not like they haven't done even better than that before. RDNA2 (aka "Big Navi" aka RX 6000) had a >100% performance-per-watt uplift over RDNA1 (aka "Navi" aka RX 5000). AMD claimed a 100% generational increase which many said was impossible. So ATi said "hold my beer" and proceeded to over-deliver on the promise that AMD made.
bd5e07eaaa1a69eaad755c08221a3c546f42e7519c62029f2f81d68c6187d007.jpg
 
The question in my mind is: Will AMD increase ray-tracing performance? When the 2XXX series launched RT was a gimmick hardly any game used that tanked performance to a ridiculous degree. Now the consoles have it, it's part of DX, and games are playable with the effect enabled. With RT becoming less of a gimmick, I want to see some commitment to the technology from AMD.

Their raster performance is competitive, but RT is leagues behind, and that is starting to matter.
 
The up to 50% performance increase for GPUs wouldn't excite me too much if they were dumping more wattage into the cards and calling it day which on the surface is what it looks like Nvidia is doing for the 4000 series (a gross simplification but still). However, specifically an increase per watt hopefully means better efficiency and doing more with less, which does excite me. I can't wait to see what comes of RDNA3 when it drops.

My i7-6700 and GTX 1080 are getting long in the tooth and I want a more compelling reason to FINALLY upgrade ha ha.
 
The up to 50% performance increase for GPUs wouldn't excite me too much if they were dumping more wattage into the cards and calling it day which on the surface is what it looks like Nvidia is doing for the 4000 series (a gross simplification but still). However, specifically an increase per watt hopefully means better efficiency and doing more with less, which does excite me. I can't wait to see what comes of RDNA3 when it drops.

Both camps are heading to provide graphics card for 4K 120FPS or so. Or at least bring out a beast that is capable of taking the thrown of one and another. You cant build a graphics card with just a 250W TDP and claim a top position. The clocks and amount of shaders or clusters need to significantly go up for that.

The avg GPU clock is'nt 900 to 1500Mhz anymore; it's double of that, woud'nt suprise me if we start seeing GPU's ticking at 3GHz which was 10 years quite difficult to accomplish.

As for the CPU's: I am not putting lots of hope on overclocking anymore. They both have designed a technique that extracts the best possible speed or performance out of a given workload and such. Because of that true overclocking kind of is slowly changing to fields where only extreme cooling would be needed to even make a punch.

 
A 50% increase in graphics performance? HOLY !@#$. Is that possible?

No it will be around 100% uplift but with a 50% improvement in performance per watt, so it will still use a fair bit more power, but nowhere near as much as Lovelace for the same performance.

You could look at it in two ways. Assuming we have standardised way of comparing two different generations of cards, RDNA3 at the same fps as RDNA2 could use 33% less power, or it could use the same power at 50% more fps or it could use 33% more power at 100% more fps.
 
Hopefully they can deliver what they promised. I think this year and the next 2 years we will see a great competition between AMD, Intel and NVidia. That's great for us, the customers.
 
Back