Speculation: Ryzen 4000 series/Zen 3

[DrMrLordX]

@Shivansps

It is inaccurate to assume that continuing to use Vega would be a decision borne of "no demand for powerful iGPUs". AMD is going to continue to update Vega as a part of CDNA. The big question is whether or not updated CDNA will find it in any display-capable hardware. RDNA is allegeldy not compute-friendly, so I would not expect to find it in workstation cards and other professional dGPUs intended for display duty (which do exist, and have for some time). We already know Renoir has Vega and AMD has decided on Vega for mobile Renoir, so logically desktop Renoir - if/when it appears - will also be a Vega product. Whatever happens on the desktop will be an extension of whatever it is AMD decides is important for mobile - not for desktop. Desktop APUs are also-rans.

 

[soresu]

2xCU is ~4.1mm2.

Good grief, even at 40 CU's that is only 82mm2 - just goes to show how much die space could be saved by a more area efficient cache memory tech.

Likewise by having the memory directly stacked on die they might be able to reduce that hulk of a memory PHY and controller, just so much die space going on anything but logic.

 

[soresu]

The big question is whether or not updated CDNA will find it in any display-capable hardware. RDNA is allegeldy not compute-friendly, so I would not expect to find it in workstation cards and other professional dGPUs intended for display duty (which do exist, and have for some time).

From what I've seen so far RDNA is certainly good enough for compute - it just isn't good enough to match nVidia's ML dominance with those tensor units, hence CDNA.

I can't see any point in having display capabilities in CDNA, Radeon Instinct MIxx cards already lack display ports don't they?

I think we might expect to see increased INT64 performance in CDNA too, if not in Arcturus then later - AMD seems to be weak in this area vs nVidia.

As for pro dGPU's used for display/video/rendering - RDNA2 onwards will be perfect for this given the intersection/traversal acceleration should work out as a boon for pro work in offline rendering, as well as accelerating games asset creation that relies on RT capabilities (ie light/normal/displacement map baking), these have already been exploited by DCC apps using Turing RT cores.

Going forward I would expect Firepro/Radeon Pro to use RDNA uArch's and Radeon Instinct cards to use CDNA uArch's - I would definitely expect a future Radeon Pro SSG series with 4 TB+ of NAND using PCIE 4 or better controllers.

 

[Shivansps]

@Shivansps

It is inaccurate to assume that continuing to use Vega would be a decision borne of "no demand for powerful iGPUs". AMD is going to continue to update Vega as a part of CDNA. The big question is whether or not updated CDNA will find it in any display-capable hardware. RDNA is allegeldy not compute-friendly, so I would not expect to find it in workstation cards and other professional dGPUs intended for display duty (which do exist, and have for some time). We already know Renoir has Vega and AMD has decided on Vega for mobile Renoir, so logically desktop Renoir - if/when it appears - will also be a Vega product. Whatever happens on the desktop will be an extension of whatever it is AMD decides is important for mobile - not for desktop. Desktop APUs are also-rans.

I dont see anything wrong with that, EVERY APU ha been designed this way. In fact, the mobile market demands way more than desktop for an APU, AMD could have used Vega 3 for the entire Renoir line on Desktop and it will still beat the crap out of Intel IGPs, but on mobile things are diferent. Think about that, if it was only about desktop, AMD could have settled with just Vega 3, maybe 5 at best, for Raven and Picasso, because that was way more than enoght to beat Intel, it is due to mobile that we get bigger IGPs.

 

[DrMrLordX]

I can't see any point in having display capabilities in CDNA, Radeon Instinct MIxx cards already lack display ports don't they?

Not sure? Vega20 is in workstation cards, though:

https://www.amd.com/en/graphics/workstations-radeon-pro-series

I dont see anything wrong with that, EVERY APU ha been designed this way.

Precisely. Which is one of the reasons why availability of some desktop APU (3400G or otherwise) doesn't really tell us much. It is about as far down on the AMD totem pole as you can get. AMD isn't making major purchasing or design decisions based around their desktop APU lineup. Desktop APUs appear to get whatever dice are left over from mobile.

Desktop APUs could have had a 3CU configuration, but AMD isn't going to spend the extra money producing a separate die for that purpose. Everything will come from the standard 8CU Renoir die.

 

[soresu]

Not sure? Vega20 is in workstation cards, though:

Yes and Vega20 is the last gasp of the old unified uArch paradigm, not surprising that they are recouping the investment as much as possible.

I'd expect any workstation use case wanting double precision, heavy compute or ML/AI will fall to CDNA in the future.

They may have some basic display stuff in there - I'm not sure how many workstation boards have integrated display, but I would chance a guess at not many.

 

[mopardude87]

Haven't even gotten my 3900x YET and already i am hyped as all can be for this 4000 series. Hopefully its a bit more then a 15% performance increase. I would be ok perhaps with that if the efficiency is top notch, would love to see 32 threads easily air cooled. That would be DOUBLE the damn threads of a 9900k and on air.

I think i got bit by the hardware bug, haven't been bitten since the C2D days.

 

[Topweasel]

The point was about AMD could potencially keep using Vega after Renoir because there is potencially no demand for more powerfull IGPs. This is the exact quote i was repling to.

They answered why you just don't believe them. Yes they prioritized Cores over CU's but that is because they had a target (8c) and a certain size (150mm considering 7nm). They went vega because they knew how with RR and Picasso experience, on top of work with Vega20, to optimize that as an APU versus Navi. Also well, this is how APU's have always gotten their GPU's they are of cadence and launched with the older GPU.

So limited experience with Navi as iGPU, only just launched, competitive graphics might have limited Core count, on what is easily one of the most important launches ever. Mobile accounts for a ridiculous amount of sales, if they can make a dent, it would do wonders for them. Vega was the safe choice. That doesn't mean they have given up or that we won't ever see Navi on an APU. It means an assessment was made and for this release a tuned Vega seemed like the safer bet.

 

[AtenRa]

Next APU will not have VEGA.

Also, there is strong demand for desktop APUs, the vast majority of desktop sales are coming with iGPU only, just look at the GPU market share of Intel vs AMD/NVIDIA.
Currently the problem is that AMD doesnt have the resources to invest in to OEM//ODMs to create desktop APU systems, they focus on Mobile first and later you will see desktop systems with Renoir.

 

[DrMrLordX]

just look at the GPU market share of Intel vs AMD/NVIDIA.

Intel includes iGPUs whether or not you want them.

 

[AtenRa]

Intel includes iGPUs whether or not you want them.

True but iGPU only Desktop systems are the majority, 55-65%

edit: total PC GPU attach rate is only 30-35%

 

[soresu]

edit: total PC GPU attach rate is only 30-35%

Does this mean % of desktop users that buy dGPU's?

 

[soresu]

I have to admit to being disappointed by the slow progression of APU performance.

Kaveri was supposed to be the first 1 TFLOP APU way back on 28nm, and underwhelmed somewhat in an entirely different way from the Steamroller cores.

Renoir seems very decent for raster engines, but still not much more than 2.2-2.5 TFLOPS of equivalent performance vs Kaveri in raster engines, and even less impressive in general compute, for which it loses to the PS4 chip from 2013, albeit while drawing far less power.

With any luck Cezanne and Rembrandt will take advantage of those jumps in perf/watt and increase the CU count substantially - just 16 for Cezanne would be a good start, and not step too hard on the toes of the 5500 XT, then perhaps with Rembrandt we can finally get an AMD brand APU matching the CU count of the Vega M using package from Intel. Or 12 for Cezanne perhaps considering the greater uplift going to N5P vs N7+.

 

[mopardude87]

I wonder how many people are like me, who just use both? I use my HD630 as primary while my 1080ti folds. Get tons of lag and stuff using the 1080ti as a driver as well as folder. The hd630 makes things almost to good to be true. I forget i am folding usually cause things are that smooth.

I got so close to opting for a 9900k over the 3900x just cause of my use case but i rather have a dedicated GT1030 as driver with the 3900x then the 9900k with HD630 ANY DAY. If the GT1030 wants to play nice with the 1080ti will be up for debate. Between 2 gpus and a soundcard, yeah it should be interesting to see how my Aorus Elite x570 handles all this.

 

[moinmoin]

I have to admit to being disappointed by the slow progression of APU performance.

Kaveri was supposed to be the first 1 TFLOP APU way back on 28nm, and underwhelmed somewhat in an entirely different way from the Steamroller cores.

Renoir seems very decent for raster engines, but still not much more than 2.2-2.5 TFLOPS of equivalent performance vs Kaveri in raster engines, and even less impressive in general compute, for which it loses to the PS4 chip from 2013, albeit while drawing far less power.

With any luck Cezanne and Rembrandt will take advantage of those jumps in perf/watt and increase the CU count substantially - just 16 for Cezanne would be a good start, and not step too hard on the toes of the 5500 XT, then perhaps with Rembrandt we can finally get an AMD brand APU matching the CU count of the Vega M using package from Intel. Or 12 for Cezanne perhaps considering the greater uplift going to N5P vs N7+.

I think you are jumping to conclusion. The natural progression of mobile APUs is to deliver more in a smaller power envelope, and in that regard Renoir absolutely delivers. Regarding raw graphic performance the iGPU seems to be limited still as seen with the iGPU in the U line performing equally both at a cTDP of 15W and 25W (haven't seen reviews looking at the behavior of the iGPU in H/HS parts at different cTDPs yet). So if you want to know the true performance of Renoir we'll have to wait for the desktop APU where the iGPU is bound to be unlocked again.

 

[soresu]

I think you are jumping to conclusion. The natural progression of mobile APUs is to deliver more in a smaller power envelope, and in that regard Renoir absolutely delivers. Regarding raw graphic performance the iGPU seems to be limited still as seen with the iGPU in the U line performing equally both at a cTDP of 15W and 25W (haven't seen reviews looking at the behavior of the iGPU in H/HS parts at different cTDPs yet). So if you want to know the true performance of Renoir we'll have to wait for the desktop APU where the iGPU is bound to be unlocked again.

I know its meant to do that, but it doesn't stop me wanting to ditch the dGPU and just have a nice compact SFF PC that doesn't bite the bucket when I try to use it for some significant compute (aka Blender or other DCC OpenCL).

 

[chrisjames61]

Despite having a 1600, 1700, 2700 and 3600 I am a big fan of AMD apu's back from FM2 days. I also have a couple Ryzen apu's. 2200GE and a 2400G and would really like to see more coming down the pike for the AM4 socket. Not really interested in a 3200 or 3400 because they are just speed bumps from the earlier cpu's. Correct me if I am wrong.

 

[darkswordsman17]

True but iGPU only Desktop systems are the majority, 55-65%

edit: total PC GPU attach rate is only 30-35%

That's because most of those others don't care about GPU and don't know its even there. Its simply their display controller. Those types of systems to me won't last much longer, as there's going to be ample ways to handle alternatives, like simple terminals, maybe ARM based ones with most of the work being done on servers. Which ARM itself could potentially even start to unseat that market (by that, I mean ones using computers for basic computer tasks, not server market) if Microsoft keeps pushing that direction, and with it looking increasingly like Apple is moving there, and Android being pretty much the Windows version of Linux for ARM. If a company were to find a way to put a PC spec GUI overlay while being able to keep updates with Android, I think they could do reasonably well.

 

[AtenRa]

Does this mean % of desktop users that buy dGPU's?

No, Desktop attach rate (% of Desktop systems with discrete GPUs) is between 40-45% and can reach up to 55% depending on the Quarter.
On average is closer to 45-47% now days were Desktop volumes are lower.

Total PC attach rate includes all Desktop, Workstation, Servers, rendering farms etc. The attach rate (% of all those systems that use a discrete GPU) is between 30% and 35% depending on the Quarter.

 

[soresu]

No, Desktop attach rate (% of Desktop systems with discrete GPUs) is between 40-45% and can reach up to 55% depending on the Quarter.
On average is closer to 45-47% now days were Desktop volumes are lower.

Total PC attach rate includes all Desktop, Workstation, Servers, rendering farms etc. The attach rate (% of all those systems that use a discrete GPU) is between 30% and 35% depending on the Quarter.

Oh right, I kinda just include all those under the desktop umbrella in my head anyways - still got the old CRT box on top of another steel box picture in my head instead of the tower cases I've actually been using since the late 90's / early 00's.

I can't for the life of me actually remember when I got my first tower case now that I think about it, pretty sure it was a server throwaway because it was huge.

 

[amrnuke]

So what is this news coming on Thursday supposed to be about? That's May 7, two months before the infamouse 7/7 date everyone was previously talking about.

 

[DrMrLordX]

It's the annual stockholder's meeting, fwiw.

 

[soresu]

So what is this news coming on Thursday supposed to be about? That's May 7, two months before the infamouse 7/7 date everyone was previously talking about.

It's certainly time enough to have some significant information dump on Zen3 - assuming it's coming out in September/October timeframe we are already closer to it than the Next Horizon event that first outlined Zen2 prior to its launch.

 

[uzzi38]

Just a little something I squeezed out of AdoredTV. Had a strong feeling it was the case with how late B550 is, but good to get confirmation.

This should also probably throw out any rumours that Zen 3 will not support AM4 or bresk backwards compatibility.

https://twitter.com/x/status/1257414380377059330

This is likely one of the things that will be discussed on Thursday.

EDIT: All I know of is that there is an embargo on Thursday to discuss something that could have easily been covered in the 3300X/3100 launch. I can't say I know specifics apart from one other thing - desktop Renoir is rumoured to launch alongside B550. Take that how you will.

 

[YAYgee]

In April Forbes said that AMD "announced an embargo of today for part of the news in a briefing I listened to this morning,", i.e. the Ryzen 3/B550 press release, "with the rest under wraps for a couple of weeks", this Thursday that part of the briefing will be made public. Regarding B550, Linus and others are also under NDA. It's made out to be a big deal. We'll see if that's indeed the case and if not I agree with @uzzi38 that could have easily been covered in the 3300X/3100 launch". Sadly, it sounds like you already know it isn't a big deal.

I doubt we'll get bad news on Thursday. Still, throwing this quote in here as I can't tell if there's any truth to it:

AMD has screwed up a bit and there's a hardware bug affecting all currently made B550 boards. It's not something that prevents regular operation, but it's a silly screwup that would be annoying, if those boards ship as they are. I can't share what it's is here though, as it might reveal where I got the info from. - from TPU's forum