AMD Confirms Zen 5 will Get Ryzen 8000 Series Branding, "Navi 3.5" Graphics in 2024

[btarunr]

AMD in one of its Meet the Experts presentations to the retail channel vendors, confirmed that the next-generation "Zen 5" architecture will see its desktop part branded under the Ryzen 8000 series. The company has known to skip a thousand-number sequence each generation for its mainstream-desktop series, the way it skipped Ryzen 4000 series nomenclature between the "Zen 2" based Ryzen 3000 "Vermeer" and "Zen 3" based Ryzen 5000 Vermeer; and more recently, between "Vermeer" and the "Zen 4" based Ryzen 7000 "Raphael," which makes this an interesting development. AMD's next-generation mainstream-desktop processor is expected to be codenamed "Granite Ridge," it will feature up to 16 "Zen 5" CPU cores across up to two CCDs. The processor I/O (and its 6 nm cIOD) is expected to be largely carried over, except that it could be upgraded with support for higher DDR5 memory speeds.

Another major disclosure is the very first mention of "Navi 3.5" This implies an incremental to the "Navi 3.0" generation (Radeon RX 7000 series, RDNA3 graphics architecture), which could even be a series-wide die-shrink to a new foundry node such as TSMC 4 nm, or even 3 nm; which scoops up headroom to dial up clock speeds. AMD probably finds its current GPU product stack in a bit of a mess. While the "Navi 31" is able to compete with NVIDIA's high-end SKUs such as the RTX 4080, and the the company expected to release slightly faster RX 7950 series to have a shot at the RTX 4090; the company's performance-segment, and mid-range GPUs may have wildly missed their performance targets to prove competitive against NVIDIA's AD104-based RTX 4070 series, and AD106-based RTX 4060 series; with its recently announced RX 7600 being based on older 6 nm foundry tech, and performing a segment lower than the RTX 4060 Ti.



View at TechPowerUp Main Site | Source

 

[wolf]

Very interesting development, I wonder if this process will enable them to ratchet up the clock speeds a bit and fix the rumored problems with the 7900 series which had it punch below expectations. Very interesting indeed.

 

[katzi]

yea and I bet they ramp up the pricing 20-30% alongside the supposed IPC gains.

 

[john_]

Maybe Navi 3.5 will be a... working Navi 3.0?

 

[Snoop05]

Navi 3.5 is mentioned in the context of AM5 so might be iGPUs only, also could be just Display / Media engine update.

 

[TumbleGeorge]

Navi 3.5 is mentioned in the context of AM5 so might be iGPUs only, also could be just Display / Media engine update.

Yes Ryzen 8000 series was mentioned with rDNA3.5 for iGPU.

 

[Tom Yum]

"AMD probably finds its current GPU product stack in a bit of a mess. While the "Navi 31" is able to compete with NVIDIA's high-end SKUs such as the RTX 4080, and the the company expected to release slightly faster RX 7950 series to have a shot at the RTX 4090; the company's performance-segment, and mid-range GPUs may have wildly missed their performance targets to prove competitive against NVIDIA's AD104-based RTX 4070 series, and AD106-based RTX 4060 series; with its recently announced RX 7600 being based on older 6 nm foundry tech, and performing a segment lower than the RTX 4060 Ti."

This is a big call to make without supporting evidence. Navi 33 (7600 family) is 204mm2, 13.3b transistors on 6nm (which is really an optimised 7nm process). AD106 (4060Ti) is 190mm2, 22.9b transistors on the much denser and more expensive 5nm process. AD104 (4070) is 295mm2, 33.8b transistors at the same 5nm. A 5nm wafer is estimated to cost about 70% more than a 7nm wafer (https://www.tomshardware.com/news/t...aled-300mm-wafer-at-5nm-is-nearly-dollar17000) so a AD106 die is probably costing NVIDIA ~50% more than a Navi 33 die cost AMD.

A 4060Ti is approx 25% faster than a 7600, but takes a die that costs 50% more to make to do so. Which basically suggests the 7600 performs exactly where AMD targeted it, it is just that they deliberately weren't targeting the 4060Ti/4070. Which makes sense, they are still trying to clear the channel of excess 6700's/6800's, and don't want to take further margin hit by adding another competitor to the 6700/6800 before the channel has been cleared.

 

[TheinsanegamerN]

"AMD probably finds its current GPU product stack in a bit of a mess. While the "Navi 31" is able to compete with NVIDIA's high-end SKUs such as the RTX 4080, and the the company expected to release slightly faster RX 7950 series to have a shot at the RTX 4090; the company's performance-segment, and mid-range GPUs may have wildly missed their performance targets to prove competitive against NVIDIA's AD104-based RTX 4070 series, and AD106-based RTX 4060 series; with its recently announced RX 7600 being based on older 6 nm foundry tech, and performing a segment lower than the RTX 4060 Ti."

This is a big call to make without supporting evidence. Navi 33 (7600 family) is 204mm2, 13.3b transistors on 6nm (which is really an optimised 7nm process). AD106 (4060Ti) is 190mm2, 22.9b transistors on the much denser and more expensive 5nm process. AD104 (4070) is 295mm2, 33.8b transistors at the same 5nm. A 5nm wafer is estimated to cost about 70% more than a 7nm wafer (https://www.tomshardware.com/news/t...aled-300mm-wafer-at-5nm-is-nearly-dollar17000) so a AD106 die is probably costing NVIDIA ~50% more than a Navi 33 die cost AMD.

A 4060Ti is approx 25% faster than a 7600, but takes a die that costs 50% more to make to do so. Which basically suggests the 7600 performs exactly where AMD targeted it, it is just that they deliberately weren't targeting the 4060Ti/4070. Which makes sense, they are still trying to clear the channel of excess 6700's/6800's, and don't want to take further margin hit by adding another competitor to the 6700/6800 before the channel has been cleared.

AMD doesnt know what it's targeting, given its own 6600xt undermines the 7600 at every turn. The 4060ti was a wet fart that showed how gimped nvidia's GPUs are, and the 7600 was an embarrassment that really showed off how little rDNA3 (or its compiler) brings to the table.

 

[john_]

"AMD probably finds its current GPU product stack in a bit of a mess. While the "Navi 31" is able to compete with NVIDIA's high-end SKUs such as the RTX 4080, and the the company expected to release slightly faster RX 7950 series to have a shot at the RTX 4090; the company's performance-segment, and mid-range GPUs may have wildly missed their performance targets to prove competitive against NVIDIA's AD104-based RTX 4070 series, and AD106-based RTX 4060 series; with its recently announced RX 7600 being based on older 6 nm foundry tech, and performing a segment lower than the RTX 4060 Ti."

This is a big call to make without supporting evidence. Navi 33 (7600 family) is 204mm2, 13.3b transistors on 6nm (which is really an optimised 7nm process). AD106 (4060Ti) is 190mm2, 22.9b transistors on the much denser and more expensive 5nm process. AD104 (4070) is 295mm2, 33.8b transistors at the same 5nm. A 5nm wafer is estimated to cost about 70% more than a 7nm wafer (https://www.tomshardware.com/news/t...aled-300mm-wafer-at-5nm-is-nearly-dollar17000) so a AD106 die is probably costing NVIDIA ~50% more than a Navi 33 die cost AMD.

A 4060Ti is approx 25% faster than a 7600, but takes a die that costs 50% more to make to do so. Which basically suggests the 7600 performs exactly where AMD targeted it, it is just that they deliberately weren't targeting the 4060Ti/4070. Which makes sense, they are still trying to clear the channel of excess 6700's/6800's, and don't want to take further margin hit by adding another competitor to the 6700/6800 before the channel has been cleared.

You should compare Navi 23 and Navi 33 to see if Navi 33 is really performing as expected.
Comparing with Nvidia you should take into consideration all the tensor cores and stuff that Nvidia GPUs include, remove their die area and then make assumptions about gaming performance. Maybe even remove the die area or RT cores and compare those two chips just for raster performance per die area.

As for the chip costs. A percentage doesn't give any information here. 50% more over $5 is totally different than 50% more over $50.

As for 7700/7800, if they where performing as expected, they would have been already in the market. But probably they are slower that 69x0/6800 equivalent models, maybe even compared to 67x0 (I mean 7700 might be slower than 67x0, 7800 than 6800/69x0).

What the article says is more or less what many, if not the majority, suspects.

 

[AusWolf]

Wait, we don't even have a full Navi 3 lineup, yet.

 

[rv8000]

Interesting conclusion to say AMD has wildly missed midrange segments when they haven’t even released a full stack of products. The quality and integrity of reporting from TPU is really lacking as of late.

AMD and Nvidia have both released astonishingly bad product line-ups offering almost nothing of value this generation. Releasing 7x50 cards is going to be more bad news for AMD if they don’t even bother releasing 7700 and 7800 cards.

Its disappointing to see the complete opposite between AMDs CPU and GPU branches right now.

 

[john_]

Wait, we don't even have a full Navi 3 lineup, yet.

GHave you seen any 7700/7800 models on the desktop?

 

[AusWolf]

GHave you seen any 7700/7800 models on the desktop?

That's what I mean. Where are they?

 

[Squared]

Vega and RDNA3 haven't been a part of CPU-oriented AMD processors (Ryzen 7000 is the only CPU-oriented line with integrated graphics, RDNA2), so the slide is referring to desktop APUs. These usually use the same chip as the mobile APUs, so they follow what's released on mobile (although sometimes, like with Ryzen 6000, no mobile variant comes to desktop). The slide calls for a yearly cadence, which is cool because that's a faster cadence than AMD has been maintaining. It also means that Zen 5 is due out this time next year in a mobile APU, and customers will want updated graphics too, but last I heard RDNA4 is still due in late 2024 following the typical 2-year cadence for graphics. So RDNA3 or 3.5 are the only options.

So RDNA3.5 might just be RDNA3 with a small clock speed boost, or on a new node like TSMC N3. In the past AMD has used names like "Zen+" and "Zen 3+" for that kind of update to the CPU, so maybe "3.5" is something more. Maybe the rumors are true and RDNA3 does have a flaw holding it back from AMD's aspirations, and version 3.5 corrects this flaw and thereby achieves a nice boost in performance.

Going back to idea of an annual cadence, it sounds like Zen 5 is on track for release just one year after Zen 4, but it's possible the cadence described in the slide is actually a staggered cadence, with GPU and CPU upgrades taking turns each year. But it probably doesn't matter; when processor companies promise a cadence they usually keep it up for like 18 months then fall behind.

 

[AnotherReader]

As for 7700/7800, if they where performing as expected, they would have been already in the market. But probably they are slower that 69x0/6800 equivalent models, maybe even compared to 67x0 (I mean 7700 might be slower than 67x0, 7800 than 6800/69x0).

Given the purported specifications of the 7800 XT and the performance of the 7600 compared to the 6650 XT, I expect The 7800 XT to be about 5-10% faster than the 6800 XT. That explains why they haven't released it yet. They have to wait for the 6800 and the 6900 to go out of stock before they can sell something that has essentially the same performance for most games.

 

[AusWolf]

Given the purported specifications of the 7800 XT and the performance of the 7600 compared to the 6650 XT, I expect The 7800 XT to be about 5-10% faster than the 6800 XT. That explains why they haven't released it yet. They have to wait for the 6800 and the 6900 to go out of stock before they can sell something that has essentially the same performance for most games.

That would make sense, although they didn't wait for the 6600 series to go out of stock to make way for the 7600, so I don't know...

 

[Dave65]

Maybe Navi 3.5 will be a... working Navi 3.0?

NAVI 3 works just fine but thanks for asking

 

[john_]

Interesting conclusion to say AMD has wildly missed midrange segments when they haven’t even released products in this segment. The quality and integrity of reporting from TPU is really lacking as of late.

AMD and Nvidia have both released astonishingly bad product line-ups offering almost nothing of value this generation. Releasing 7x50 cards is going to be more bad news for AMD if they don’t even bother releasing 7700 and 7800 cards.

Its disappointing to see the complete opposite between AMDs CPU and GPU branches right now.

First let me say that I am full AMD. What I post is not Nvidia BS, just disappointment.

Let's see.

RX 7600 2048 cores, RX 66x0 XT 2048 cores, more or less same performance. RDNA 3 was suppose to offer higher performance because of double... something. We don't see it unfortunately.

RX 6950 XT 5120 cores, RX 7900 XT 5376 cores, little performance improvement.

So it's like we get the same performance from the same number of cores. RDNA 3 seems to be working in.... RDNA 2 mode... Power consumption is not improve either, meaning those slides about 50% perf/watt that AMD was promising, never materialized.


Now, RX 6800 non XT, 3840 cores, RX 6800 XT 4608 cores. RX 7800 is expected to come with 3840 cores, meaning more or less performance same with RX 6800 non XT. So, a DOA chip if it is priced above $400.

But, RX 6700 non XT 2304 cores, RX 67x0 XT 2560 cores. RX 7700 3072, 7700 XT 3476 cores. RX 7700 seems to be the only line of cards that do have some chance in the market if they are priced based on current RX 67x0/XT retail prices. (wasn't remembering specs correctly when doing previous post)

Off course these are personal opinions.

That would make sense, although they didn't wait for the 6600 series to go out of stock to make way for the 7600, so I don't know...

They priced RX 7600 just above RX 6650 XT, so they do wait for 66x0/XT stock to go out. Then slowly they will drop the RX 7600's price towards $230.

 

[AnotherReader]

So it's like we get the same performance from the same number of cores. RDNA 3 seems to be working in.... RDNA 2 mode... Power consumption is not improve either, meaning those slides about 50% perf/watt that AMD was promising, never materialized.

I think AMD took the gains from some games that they optimized by hand and expected the compiler to match that. Look at the Callisto Protocol: the 7600 is 16% faster than the 6650 XT despite the 3% to 7% higher clock speed for the 6650 XT that TPU used. Unfortunately, the compiler is well known to miss obvious optimization opportuntities. On top of that, it's clear from their presentations that higher clock speeds were expected for RDNA3. Those didn't materialize either. I know Raja is much lampooned, but I think he was right about the cause of the shortcomings of Vega. AMD focused on RDNA2 and he didn't get the support from the physical engineers to get Vega's speeds up to mark. Perhaps, in time, we'll find out that RDNA3 was the victim of a successor.

 

[john_]

I think AMD took the gains from some games that they optimized by hand and expected the compiler to match that. Look at the Callisto Protocol: the 7600 is 16% faster than the 6650 XT despite the 3% to 7% higher clock speed for the 6650 XT that TPU used. Unfortunately, the compiler is well known to miss obvious optimization opportuntities. On top of that, it's clear from their presentations that higher clock speeds were expected for RDNA3. Those didn't materialize either. I know Raja is much lampooned, but I think he was right about the cause of the shortcomings of Vega. AMD focused on RDNA2 and he didn't get the support from the physical engineers to get Vega's speeds up to mark. Perhaps, in time, we'll find out that RDNA3 was the victim of a successor.

Some times I wonder how much faster AMD cards would have been if... Nvidia was programming the drivers.

 

[AnotherReader]

That would make sense, although they didn't wait for the 6600 series to go out of stock to make way for the 7600, so I don't know...

Yes, I can only speculate that the 6600 series is closer to going out of stock. Only AMD would know the answer. I have been thinking about alternate configurations and given the small size of Navi 33, I think there was room for an alternate Navi 33 with the same CU count as the 6700 XT. Give it a 192-bit bus to forestall the 8 GB objections and it would have been just around 250 mm^2 with a commensurate 48 MB L3. The tradeoff would have been the higher power draw. It would have been almost 200W at the same clock speeds as the real Navi 33.

 

[Daven]

Some times I wonder how much faster AMD cards would have been if... Nvidia was programming the drivers.

Since Nvidia is between -5 to 20% faster for the equivalent SKU, I would say not much faster. But like everyone has said, neither company has finished the whole product stack yet. Besides, are you comparing the 4090 to the 7900xtx (20% faster) or the 4080 to 7900xtx (5% slower)? I guess I really have no idea what you are saying.

 

[rv8000]

First let me say that I am full AMD. What I post is not Nvidia BS, just disappointment.

Let's see.

RX 7600 2048 cores, RX 66x0 XT 2048 cores, more or less same performance. RDNA 3 was suppose to offer higher performance because of double... something. We don't see it unfortunately.

RX 6950 XT 5120 cores, RX 7900 XT 5376 cores, little performance improvement.

So it's like we get the same performance from the same number of cores. RDNA 3 seems to be working in.... RDNA 2 mode... Power consumption is not improve either, meaning those slides about 50% perf/watt that AMD was promising, never materialized.


Now, RX 6800 non XT, 3840 cores, RX 6800 XT 4608 cores. RX 7800 is expected to come with 3840 cores, meaning more or less performance same with RX 6800 non XT. So, a DOA chip if it is priced above $400.

But, RX 6700 non XT 2304 cores, RX 67x0 XT 2560 cores. RX 7700 3072, 7700 XT 3476 cores. RX 7700 seems to be the only line of cards that do have some chance in the market if they are priced based on current RX 67x0/XT retail prices. (wasn't remembering specs correctly when doing previous post)

Off course these are personal opinions.


They priced RX 7600 just above RX 6650 XT, so they do wait for 66x0/XT stock to go out. Then slowly they will drop the RX 7600's price towards $230.

There is an efficiency improvement though. People also need to remember AMD and Nvidia tend to explicitly say “up to”, and that should be emphasized because saying well X company promised Y and yada yada they didn’t deliver is wrong.

Take CP 2077 @4k 6900XT vs 7900 XT (TPU 7900XT review data)

6900XT - 38 fps @ typical power consumption of 310w
7900XT - 52 fps @ typical power consumption of 322w

In this case the 7900 XT is 28% more efficient not accounting for the minor difference in shader count. AMD being on a less efficient node than the 4000 series, that’s still not a dismiss-able efficiency or performance improvement.

With the entire mid range of the line up being non existent, it’s difficult to say how RDNA3 scales based on what’s being lopped off the lower you go in the stack. I don’t find the 7600 to be a compelling product, but at the very least from a value standpoint if you needed to buy a new card within that price range it’s not total insanity.

 

[Nordic]

AMD PLEASE! 7970 or bust!!!

 

[AusWolf]

There is an efficiency improvement though. People also need to remember AMD and Nvidia tend to explicitly say “up to”, and that should be emphasized because saying well X company promised Y and yada yada they didn’t deliver is wrong.

Take CP 2077 @4k 6900XT vs 7900 XT (TPU 7900XT review data)

6900XT - 38 fps @ typical power consumption of 310w
7900XT - 52 fps @ typical power consumption of 322w

In this case the 7900 XT is 28% more efficient not accounting for the minor difference in shader count. AMD being on a less efficient node than the 4000 series, that’s still not a dismiss-able efficiency or performance improvement.

With the entire mid range of the line up being non existent, it’s difficult to say how RDNA3 scales based on what’s being lopped off the lower you go in the stack. I don’t find the 7600 to be a compelling product, but at the very least from a value standpoint if you needed to buy a new card within that price range it’s not total insanity.

The 7600 is not a bad card... just not much better than the 6600/6650 series. But not much more expensive, either, so there's that.

It's wasted development more than anything, but at least AMD is not trying to milk customers with it, which is unfortunately a plus these days.