Zen 2 APUs/"Renoir" discussion thread

[Insert_Nickname]

Even the 3000G has 16 rops

That must be a reporting error, Vega3 has a 192:12:4 setup.

 

[juergbi]

Instead of going through all the effort to practically reinvent the wheel to put a Zen2 design on GloFo 12nm, instead, do a Raven2/Dali style 7nm die with a single 4 core CCX and 4-6CU of Vega iGPU with only 8+4+4 PCIe lanes.

That's what I expect AMD to launch beginning of next year as successor to Dali, although I'm not expecting more than 4 CUs. I don't know whether they will use it for low end Ryzen-branded laptop chips or only for Athlon and Ryzen Embedded. However, I'll be surprised if they won't launch a Zen 2-based low power/budget APU (6-15W) at all. A year after Renoir, N7 should be cheaper per mm² and the whole SoC would be almost as small as a Zen 2 CPU chiplet.

 

[Asterox]

Pretty sure Renoir does HDMI 2.1 - media and display engine were brought over from RDNA. AV1 decode - not with RDNA2 I don't think. Maybe with RDNA3. Whether or not Rembrandt will get the same treatment as Renoir when it comes to media/display engine I have no clue.

Yes, it has suport for 8K/60hz PC monitor or TV.

Page 13 - Zen 2 APUs/"Renoir" discussion thread

Page 13 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[amd6502]

Even the 3000G has 16 rops

That must be a reporting error, Vega3 has a 192:12:4 setup.

If they indeed halved the ROPs in Renoir then maybe they are double pumping it (or maybe this is also a reporting error) on demand. 8k@60Hz is a gargantuan upgrade from RR/Picasso.

A 12nm Zen 2 would have better power efficiency than 12nm Picasso and it will still be highly competitive against Intels 14nm SKUs (Core i3 10100 etc).

I would think with the widened core and transistor growth in Zen2 that efficiency would take a nosedive vs Zen+ if both are on the same 12nm node. Especially at the relevant higher end desktop frequencies, 4ghz+, and if going with the huge desktop oriented L3.

Imho Zen+ is a very good match for 12nm and mainstream and budget oriented parts. I agree with Naukkis; there's little reason to invest and get an efficiency and possibly frequency decrease vs something that is already well tuned and matched for 12nm.

There is a huge market for Desktop APUs, the problem is that AMD doesnt have the resources to create more dies and invest money

12nm projects are relatively cheap vs 7nm projects. The Matisse chiplet and its IO hub exists (with IOX on 12nm), as does Vega 11. Merging 12nm Vega and 12nm IOX, plus adding a few CU would not be a massive ground up project.

On second thought, maybe the MCM packaging (with a double sized IOX) would need a redesign and that would cost a good bit of time and money.

 

[YAYgee]

Yes, it has suport for 8K/60hz PC monitor or TV.

Page 13 - Zen 2 APUs/"Renoir" discussion thread

Page 13 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

HDMI 2.0 according to Asus.
Specifically, PN50 will pump 8K visuals at up to 30Hz via a single DisplayPort and at up to 60Hz through DisplayPort Dual-Mode (DP++)"

No 48Gb/s HDMI 2.1, 4K60 max.

 

[IntelUser2000]

GPU-Z reports 8 ROPS for all Renoir Vegas, both mobile and desktop.

Even the 3000G has 16 rops

The Renoir Vegas are not doing bad on TMUs though. So it could be thought of as being rebalanced.

 

[Asterox]

HDMI 2.0 according to Asus.
Specifically, PN50 will pump 8K visuals at up to 30Hz via a single DisplayPort and at up to 60Hz through DisplayPort Dual-Mode (DP++)"

No 48Gb/s HDMI 2.1, 4K60 max.

Asus hardver is just a custom mini PC or motherboard.Asus can do what they wont, or shoot in the fog and hit in nothing.

The point is simple, Renoir APU suport 8K/60hz resolution. But it is optimal that the motherboard has one HDMI 2.1 port.

For example first link, this is my motherboard with HDMI 2.0+Athlon 3000G=4K/60hz.

B450M S2H (rev. 1.x) Specification | Motherboard - GIGABYTE Global

Lasting Quality from GIGABYTE.GIGABYTE Ultra Durable™ motherboards bring together a unique blend of features and technologies that offer users the absolute ...

www.gigabyte.com

Microsoft OneDrive

1drv.ms

On the other side, Asrock motherboard has only HDMI 1.4. So only up to 4K/30hz, but this is bad because 4K/30hz looks like s..t.

ASRock > B450M Pro4-F

Supports AMD AM4 Socket Ryzen™ 2000, 3000, 4000 G-Series, 5000 and 5000 G-Series Desktop Processors; Supports DDR4 3200+ (OC); 1 PCIe 3.0 x16, 1 PCIe 2.0 x16, 1 PCIe 2.0 x1; AMD Quad CrossFireX™; Graphics Output: HDMI, DVI-D, D-Sub; 7.1 CH HD Audio (Realtek ALC892 Audio Codec), ELNA Audio Caps...

www.asrock.com

The vast majority of cheaper B450 motherboards(launched in March 2018) have only old HDMI 1.4.This is stupid or absurdly no doubt.

HDMI 1.4b, released in October 2011

HDMI 2.0, realased in September 2013

HDMI 2.1, realased in November 2017

 

[Gideon]

5000Mhz DDR4 with 1:1 FCLK on Renoir :O

While I'm quite sure this is a very unstable overclock (just enough to run AIDA64), if Renoir can do this, I'm a bit hopeful that Vermeer can do at least 4400-4600 Mhz with custom timings.


EDIT:

Yeah, according to the info on reddit this setting pretty much only booted and ran AIDA. He also managed to get memtest to pass with 1:2 FCLK DDR4 5700 C20-23-23-43 CR1(1T).
I still hope Vermeer can do at least 4200 MHz minimum, 4400-4600 would be a sweetspot.

 

[Shivansps]

3200G stock with DDR4-3200 vs 4300G stock DDR4-4133

We needed Renoir to realise how good Picasso was

Now, there is something wrong with Renoir, not sure if the ROP reduction is the issue or what, but the 4300G has a massive clock/bandwidth advantage i cant belive the results are so close... even the lighter Rocket League was way too close.

This is like the perf jump from the 2200G to the 3200G.

 

[coercitiv]

We needed Renoir to realise how good Picasso was

Yeah, Renoir only brought AMD to ~20% market share in the mobile market.

 

[coercitiv]

Oh look, large Ryzen Mobile 4000 APU shortage due to serious uptick in demand from larger brands.

But how can this be? Renoir is surely too expensive.

We are seeing unprecedented demand for our AMD Ryzen 4000 mobile series processors based on leadership performance and energy efficiency. AMD Ryzen 4000 processor sales ramped faster than any mobile processor in AMD history. We are increasing production to address the incremental demand requests from our customers for AMD Ryzen 4000 series processors and are focused on growing our footprint in the notebook market.

 

[Shivansps]

Oh look, large Ryzen Mobile 4000 APU shortage due to serious uptick in demand from larger brands.

But how can this be? Renoir is surely too expensive.

cheap trolling.

 

[Shivansps]

Yeah, Renoir only brought AMD to ~20% market share in the mobile market.

Yeah, no one said anything about mobile.

BTW, AMD decided to limit Picasso to DDR4-2400 in mobile.

 

[coercitiv]

Yeah, no one said anything about mobile.

Desktop APUs pricing and stock is directly linked to mobile APUs pricing and stock. Faking ignorance does not become you.

 

[Shivansps]

Desktop APUs pricing and stock is directly linked to mobile APUs pricing and stock. Faking ignorance does not become you.

Stock yes, price no. No one said anything about mobile Renoir being too expensive, starting from the point we have NO WAY to know the exact price and it does varies OEM to OEM.

 

[itsmydamnation]

I would think with the widened core and transistor growth in Zen2 that efficiency would take a nosedive vs Zen+ if both are on the same 12nm node. Especially at the relevant higher end desktop frequencies, 4ghz+, and if going with the huge desktop oriented L3.

Imho Zen+ is a very good match for 12nm and mainstream and budget oriented parts. I agree with Naukkis; there's little reason to invest and get an efficiency and possibly frequency decrease vs something that is already well tuned and matched for 12nm.

Nope AMD detailed that process normalised Zen2 is about ~15% switching capacitance improvement over Zen.
So Zen2 on 12nm should perform better then Zen on 12nm.

 

[Thunder 57]

3200G stock with DDR4-3200 vs 4300G stock DDR4-4133

We needed Renoir to realise how good Picasso was

Now, there is something wrong with Renoir, not sure if the ROP reduction is the issue or what, but the 4300G has a massive clock/bandwidth advantage i cant belive the results are so close... even the lighter Rocket League was way too close.

This is like the perf jump from the 2200G to the 3200G.

Tomshardware has a review out, but does not include the 4300G. I like the TLDR graphs.

If you're after the utmost in integrated graphics performance, the Renoir chips are undoubtedly the new king of the hill. At stock settings, the 4750G's Vega graphics engine performs roughly in line with the overclocked previous-gen Ryzen 5 3400G. After overclocking, the tuned 4750G beats the OC'd 3400G by 23% at the FHD resolution, and by 20% at 1280x720.

Seems to disagree with the narrative we've been hearing lately. Also makes me question that video's accuracy.

 

[Shivansps]

Tomshardware has a review out, but does not include the 4300G. I like the TLDR graphs.



Seems to disagree with the narrative we've been hearing lately. Also makes me question that video's accuracy.

Thats actually in line with what we saw so far, on the 3400G vs 4750Gs IGP... the 4750G it is always 5 to 15 fps faster depending on the game.
And they seem to confirm the really bad CPU perf in games, with the 4750G being below the 3300X in every test but one.
Good thing it does a lot better in productivity benchmarks, but i would want to see real world tests.

 

[Thunder 57]

Thats actually in line with what we saw so far, on the 3400G vs 4750Gs IGP... the 4750G it is always 5 to 15 fps faster depending on the game.
And they seem to confirm the really bad CPU perf in games, with the 4750G being below the 3300X in every test but one.
Good thing it does a lot better in productivity benchmarks, but i would want to see real world tests.

Really bad seems a bit hard. It's ~14% lower than the 3700X. That's not bad when you consider you get a functional iGPU. Would be be nice if spent some of that cut CU space on cache though. That doesn't seem to stop them from flying off the shelves as coercitiv mentioned.

 

[jpiniero]

Really bad seems a bit hard. It's ~14% lower than the 3700X.

IMO it should be pretty close to the 3700X. Something funky is going on.

 

[Thunder 57]

IMO it should be pretty close to the 3700X. Something funky is going on.

It's got 1/4 of that "GameCache" thingy. 8MB vs 32MB.

Actually, Tomshardware mentioned that in the conclusion.

AMD's "GameCache" marketing may seem hokey, but L3 cache capacity clearly has a big impact on Zen 2's performance in latency-sensitive workloads, like gaming.

 

[cortexa99]

Well we know the L3 is important for gaming now, are there any tests that other than gaming show 'less L3 + high clock memory(fclk)' vs 'more L3 + less clock memory(fclk)' difference?

 

[amd6502]

Nope AMD detailed that process normalised Zen2 is about ~15% switching capacitance improvement over Zen.
So Zen2 on 12nm should perform better then Zen on 12nm.

Is it a similar number too vs Zen+?

IMO it should be pretty close to the 3700X. Something funky is going on.

Probably simply the lack of a huge L3.

How would it compare to 2700, which has 16MB?

 

[naukkis]

Nope AMD detailed that process normalised Zen2 is about ~15% switching capacitance improvement over Zen.
So Zen2 on 12nm should perform better then Zen on 12nm.

That's all about backporting, thing that just appeared and everybody seems to think that it can be done to every silicon design. It can't, specially for Zen2 AMD stated that they have to do compromises to even 7nm design, they halve the L1i cache to be able to physically fit bigger uOP cache in design. Porting that design to 12nm would result to make many more much bigger compromises and resulting core design would be much slower than 14nm optimized Zen1 design without whole core redesign. And even with whole core redesign they can't come close to 7nm Zen2-design as there isn't as many transistors to be used in 12nm design as with 7nm.

 

[Arkaign]

Dang. I mean, it's not *bad*, but it is compromised enough that for a desktop I'd recommend getting a normal Zen2 and nearly any dGPU vs one of these if possible. I had hoped that it would be something where you got roughly equal performance, but with the bonus of a workable IGP, something to get you by during RMAs, or upgrade (eg; sell an RTX 20xx now, buy a new one next month or so).

Also missing an IGP made selling 6/8 core Zen2s in corporate desktop a little more difficult vs Intel where you had the unexciting yet competent Intel HD for business desktop use. This meant savings in complexity, cost, and reliability over dGPU setups when there is no need for one. This technically fixes that at least, even if it's muted a bit by the performance hit.

 

[software_engineer]

Well we know the L3 is important for gaming now, are there any tests that other than gaming show 'less L3 + high clock memory(fclk)' vs 'more L3 + less clock memory(fclk)' difference?

Compilation workloads benefit greatly from large a L3 cache. You can see that the 3700X has a substantial advantage over the overclocked 4750G for the LLVM compilation benchmark, the 3700X completing the compilation in around 80% or the time required by the 4750G.