AMD announces Zen 3 APUs with "the world's fastest graphics in a desktop processor", OEM only for now

[Random_Person1234]

1 minute ago, EduarDoc said:

It was not clear to me, are these available to the public or they will be just available through pre-build?

OEM only, which means it will only be in pre-builds. But, they will likely end up on Aliexpress and eBay.

[DeScruff]

Not surprising OEM only atm.
They likely have contracts with OEMs to supply X amount of CPUs per quarter, or OEMs are readying to refresh their lineup. Maintaining contracts is higher priority.

Just sucks with the whole GPU shortage, because otherwise this really wouldn't be that big of a deal.

[Rauten]

 

PCIE 3.0? DAFUQ?

 

Also, it'd be nice if they confirmed whether these APUs are still using Vega or have made the jump to Navi - instead, all the site says is "Radeon Graphics" - very helpful, AMD! </s> 

 

-- EDIT - Just saw in Guru3D that they're still using f*{#ing Vega still.. guess they didn't feel quite threatened by the new Intel Xe-based iGPUs. And this also means that I don't trust their "enjoy 1080p gaming!" marketing bullshit.

 

[leadeater]

1 hour ago, Rauten said:

PCIE 3.0? DAFUQ?

PCIe 4.0 uses too much power and isn't a necessity for this class product, PCIe 3.0 x4 NVMe is still wayyy faster than what you need.

 

1 hour ago, Rauten said:

-- EDIT - Just saw in Guru3D that they're still using f*{#ing Vega still.. guess they didn't feel quite threatened by the new Intel Xe-based iGPUs. And this also means that I don't trust their "enjoy 1080p gaming!" marketing bullshit.

Vega 8 is not the same as the Vega desktop architecture, many of the improvements in RDNA are in Vega 8. 

[Rauten]

24 minutes ago, leadeater said:

PCIe 4.0 uses too much power and isn't a necessity for this class product, PCIe 3.0 x4 NVMe is still wayyy faster than what you need.

I understand not needing it for a GPU - the whole point of an APU is that don't need one.
But are there really no use cases in a corporate environment for PCIe 4.0 other than GPUs? If these are OEM parts I expect most of them to end in desktop computers for offices which, sure, most are going to be used for Excel and Word and the like, but not all of them...
I don't know, it feels weird to leave it out, specially when so much marketing seems to be based on making bullet lists with as much stuff as you want.

24 minutes ago, leadeater said:

Vega 8 is not the same as the Vega desktop architecture, many of the improvements in RDNA are in Vega 8. 

Hrm; guess I will withhold judgement until I see reviews.

[leadeater]

10 minutes ago, Rauten said:

But are there really no use cases in a corporate environment for PCIe 4.0 other than GPUs?

Not for a low end/mid range desktop PC, higher end workstations won't be using these and even then PCIe 3.0 is enough for the vast majority of things. Unless you want a dual port 100Gb NIC you won't encounter bandwidth limitations sticking with PCIe 3.0.

 

If they put PCIe 4.0 in then the CPU and/or GPU clocks would be lower which would actually impact performance, unlike PCIe 4.0.

 

10 minutes ago, Rauten said:

Hrm; guess I will withhold judgement until I see reviews.

It's still only a maximum of 8 CUs, it's not going to be very fast. A lowly RX 5300 has 22 CUs and is still slow so you're going to get less than like 30% of that. Edit: Ok so my guess was wrong, it's about 50% of that, still crap just less crap

[Quackers101]

Still not able to run crysis!

[Doobeedoo]

All very good, but it will be far more interesting to see navi iGPU in newer series.

[RejZoR]

So, still VEGA graphic cores? It's weird they haven't stuffed NAVI into their APU's yet. I'd say it's much needed shift. VEGA worked well for lower clocked iGPU's, but it's starting to show its age.

[leadeater]

6 hours ago, RejZoR said:

I'd say it's much needed shift. VEGA worked well for lower clocked iGPU's, but it's starting to show its age.

You do know Vega 8 7nm clocks to 2.1GHz? Vega 8 7nm != Vega 8/11 12nm as it's more than just a straight process shrink. It's 700MHz higher than Vega 12nm counterparts along with architectural changes.

 

 

 

 

Also figures based on first generation Vega 8 7nm only clocked at 1.75GHz not 2.1GHz.

 

[RejZoR]

Then why call it VEGA?!

[WolframaticAlpha]

On 4/14/2021 at 6:26 AM, TetraSky said:

"The fastest graphics in a desktop processor"
 

Is this really an achievement, when they've held that crown for years already?

shIntel dethroned gAyy-MD for a while with Xe I guess. Now maybe amd is back on top.

[WolframaticAlpha]

On 4/14/2021 at 10:15 AM, Random_Person1234 said:

Gamers Nexus video for anyone interested.

He ain't using freedom units. He is a sinner who uses metric/s

[WolframaticAlpha]

9 minutes ago, RejZoR said:

Then why call it VEGA?!

Maybe they are paying homage to Raja Koduri(Vega was his last project at AMD iirc)

[RejZoR]

4 hours ago, WolframaticAlpha said:

He ain't using freedom units. He is a sinner who uses metric/s

23 millionth of an inch. American style.

[Moonzy]

On 4/14/2021 at 6:49 PM, leadeater said:

Not for a low end/mid range desktop PC, higher end workstations won't be using these and even then PCIe 3.0 is enough for the vast majority of things. Unless you want a dual port 100Gb NIC you won't encounter bandwidth limitations sticking with PCIe 3.0.

doesn't APU only have x8 lanes for GPU though?

unless these are different

 

pcie 3.0 x8 is like a double whammy, and it's silly on a modern 8c16t cpu that's likely to be paired with a 3080 or something

[OldFord76]

I'm still using an Athlon 3000g by itself lol I need to look in to this 

[thewelshbrummie]

On 4/15/2021 at 8:26 AM, Moonzy said:

pcie 3.0 x8 is like a double whammy, and it's silly on a modern 8c16t cpu that's likely to be paired with a 3080 or something

Except in bandwidth terms that's still plenty for VEGA 8. Take a look at this page from Techpowerup - they tested a lot of games using a 3080 but were testing different PCI-E standards, with some tests using x8 rather than x16 going right back to v1.1. Even on PCI-E 2.0 x16 it was 95%+ the performance of running it on a v4.0 slot x16 (which would also be similar, if not the same to using v3.0 x8).

 

For graphics, PCI-E 4.0, right now, isn't relevant. Also APUs aren't designed to be used with a dGPU - it's a nice backup, especially in the current climate of GPU pricing but in most cases builds with APUs won't be paired with a 3080.

 

I remember that when the 1st GPUs with PCI-E 3.0 support launched they were just saturating PCI-E 1.1 and that "only" having a PCI-E 2.0 was fine, with online comments saying that the new version wasn't needed at the time (which was true then, and the Techpowerup article confirms that this is true now for PCI-E 4.0). Essentially for GPUs, new PCI-E version support is pretty much a marketing gimmick that, unfortunately, people are buying into.

 

This time round it's changed somewhat with storage now using PCI-E - booting games/Windows doesn't need it but there are workloads that justify PCI-E 4.0, which could be run on an AMD APU. That's the main problem with these APUs not supporting Gen 4.

[porina]

On 4/15/2021 at 9:30 AM, WolframaticAlpha said:

shIntel dethroned gAyy-MD for a while with Xe I guess. Now maybe amd is back on top.

I haven't looked how the Tiger Lake (mobile) Xe implementation compares to these new APUs, but that may be why AMD had to say this was the fastest DESKTOP iGPU. The Xe implementation in Rocket Lake is only 1/3 the execution units of Tiger Lake.

 

19 hours ago, Moonzy said:

pcie 3.0 x8 is like a double whammy, and it's silly on a modern 8c16t cpu that's likely to be paired with a 3080 or something

The bigger performance concern to me is the APUs have half the L3 cache compared to the main desktop parts. If you are looking to build a system with a fast GPU it makes more sense to avoid the APUs and go for the desktop parts.

[Moonzy]

2 minutes ago, porina said:

The bigger performance concern to me is the APUs have half the L3 cache compared to the main desktop parts. If you are looking to build a system with a fast GPU it makes more sense to avoid the APUs and go for the desktop parts.

Damn, if they're equivalent I would've bought them probably, I like having iGPU because it's useful to me

[porina]

9 minutes ago, Moonzy said:

Damn, if they're equivalent I would've bought them probably, I like having iGPU because it's useful to me

AMD APUs are still monolithic. They want to keep the die size down for cost and yields, and cutting features is a way to do that.

 

Suppose we should keep it in context. Desktop Zen 3 in full CCX configurations is 4MB/core L3 cache with 0.5MB/core L2. Intel's best (currently mobile only) Tiger Lake is 3MB/core L3 and 1.25MB/core L2. Rocket Lake is 2MB/core L3 and 0.5MB/core L2. So Intel desktop currently has a similar problem to the APUs in that they probably wanted to have bigger caches for more performance but it would increase die sizes with the trade offs that come with it.

 

The L3 cache will contribute not insignificantly to Zen 3's performance. It would be interesting in due course to see people test the 16 vs 32 MB caches and see how much difference it makes in different workloads. Probably over thinking it here as my gut feeling is the differences, like many things, will be measurable but not that big in absolute terms. Still, the AMD desktop parts just make more sense if paired with a high end GPU.

[leadeater]

23 hours ago, Moonzy said:

doesn't APU only have x8 lanes for GPU though?

unless these are different

 

pcie 3.0 x8 is like a double whammy, and it's silly on a modern 8c16t cpu that's likely to be paired with a 3080 or something

Even at PCIe 3.0 x8 the impact even on very high end GPUs is marginal and no 3080's are unlikely to be paired with these CPUs as you'd be buying Ryzen Desktop not Ryzen Desktop APU. These APUs are legitimately slower than their full desktop counterparts due to multiple factors.

 

Having an integrated GPU is sort of nice but has it's own drawbacks too, primarily shifting power away from CPU cores even if not active which reduces achievable clocks. The situations where you would actually need another GPU to test a potentially faulty one is few and far between, and for many enthusiasts they have another old GPU anyway.

[Pixel5]

if these APU´s including the IGPU get solid support under Linux and support hardware accelerated transcoding in Plex these CPU´s will become the number one for anyone building a DIY NAS/Plex server 

 

Literally the only reason i bought an intel CPU for my NAS was the iGPU is supported in Plex.

[Moonzy]

21 minutes ago, leadeater said:

Even at PCIe 3.0 x8 the impact even on very high end GPUs is marginal and no 3080's are unlikely to be paired with these CPUs as you'd be buying Ryzen Desktop not Ryzen Desktop APU. These APUs are legitimately slower than their full desktop counterparts due to multiple factors.

 

Having an integrated GPU is sort of nice but has it's own drawbacks too, primarily shifting power away from CPU cores even if not active which reduces achievable clocks. The situations where you would actually need another GPU to test a potentially faulty one is few and far between, and for many enthusiasts they have another old GPU anyway.

I just want it because GPU nowadays only comes with 1 HDMI port while I need 2

 

And using adapters has been giving me headaches

[leadeater]

2 minutes ago, Moonzy said:

I just want it because GPU nowadays only comes with 1 HDMI port while I need 2

There are some but yea few and far between, and not always a GPU you would prefer to get over another.