Zen 2 APUs/"Renoir" discussion thread

Page 11 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Shivansps

Diamond Member
Sep 11, 2013
3,873
1,527
136
It doesn't seem to be hugely bandwidth starved at DDR4-2133 either. I think its just the benchmark not being very sensitive.

According to AMD the 4750G should be 19% faster in Time Spy compared to 3400G.

The score was increasing at a steady rate and then it flatlines after DDR4-3800. But yeah i would like to see a some games, both normal and shader intensive games like Metro Exodus.
 

amd6502

Senior member
Apr 21, 2017
971
360
136
The score was increasing at a steady rate and then it flatlines after DDR4-3800. But yeah i would like to see a some games, both normal and shader intensive games like Metro Exodus.

And 3800 DDR4 is still not mainstream memory. Right now it's just under $30 for an 8gb stick of 3000 DDR4, while it's over $100 for the same 8gb stick of 3800 DDR4! I say that is good timing and a good call by AMD for the case of saving die space with 8CU.

For the money saved by avoiding more exotic high frequency memory you can buy yourself an RX 570 and then some; so at you are so much better off going the dGPU route anyways.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
The score was increasing at a steady rate and then it flatlines after DDR4-3800. But yeah i would like to see a some games, both normal and shader intensive games like Metro Exodus.

It's a bit misleading. There's a huge jump between 2133-2666-3200, and even decent jump to 3600(which will flatten the curve) and it goes up a tiny bit by 200MHz thereafter.

Here's how the scaling goes-
2666: 54%
3200: 52%
3600: 32%
3800: 45%<--Likely an outlier
4000: 32%
4200: 12%
4400: 12%

I'd put ~30% as being average, neither very bottlenecked, nor the opposite. Very roughly speaking, ideally if you are designing a GPU, you'd want the gains to be roughly split between memory bandwidth, shader firepower, and texture throughput. Then its logical over an average of dozens of games, memory would be roughly be a 1/3rd of the gain, or 30-40% gain for every doubling of bandwidth.

iGPUs are atypical that a very weak GPU is paired with a usually very powerful CPU. It makes benchmarks less reliable. If you look into the frame rates the 3DMark benchmarks are getting, on iGPUs they run at very low frame rates, which is different from what you'll run the games at.

3200MHz but CL 22 vs 4200 MHz CL18. Latency is very important in many games.

Sometimes when its very bottlenecked, the system behaves very weird. So it can result in superlinear scaling.
 
Last edited:

Shivansps

Diamond Member
Sep 11, 2013
3,873
1,527
136
And 3800 DDR4 is still not mainstream memory. Right now it's just under $30 for an 8gb stick of 3000 DDR4, while it's over $100 for the same 8gb stick of 3800 DDR4! I say that is good timing and a good call by AMD for the case of saving die space with 8CU.

The problem is not really the cut down, the segmentation and pricing is, thats were the "saving die space" ends for us, the consumers. Having the 3400G being replaced by the 4300G, and the 3200G getting no replacement is just unacceptable, thats really the problem, if the 4300G were replacing the 3200G at $110 and the 4600G ending at around $160 then it would be something else.

For the money saved by avoiding more exotic high frequency memory you can buy yourself an RX 570 and then some; so at you are so much better off going the dGPU route anyways.

I always said that what it really matters is DDR4-3200 and stock performance, because that is what most people are going to use it. Overclock and faster ram speed is good for us, the people who actually tinker with these things, but we are like what? 1%?

But remember that even at DDR4-3200 Vega 11 will always perform better than Vega 8. The 3200G needs 1.6Ghz to match a stock 3400G, so bottleneck or not, bigger IGP performs better.
 

amd6502

Senior member
Apr 21, 2017
971
360
136
Having the 3400G being replaced by the 4300G, and the 3200G getting no replacement is just unacceptable

I think the 4300g and 4600g will supplement rather than replace the 3200g+3400g, at least for over another year until another Zen3 APU can fill that lower end gap that 3200g is at.

To guess what is still produced look at what laptops are still in demand and selling. We have Raven2 and Picasso filling the low end to mid mainstream laptop processors (while Renoir is filling the mid to upper mainstream and elite notebooks. Gradually Picasso will be phased out in favor of Raven2 (just like Bristol Ridge was phased out for Stoney ridge). Around that time, a new Zen3 (maybe Zen2) mobile focused APU will replace Picasso, and soon after that we'll have our 3200g replacement derived from that.

On desktop we have Athlon 3000ge at $50 , and likely by time Renoir arrives on AM4 the 3200g / 3400g will shift down in price and will complement in filling the price ranges with Renoir; my guess for street prices (after any initial retailer gauging / supply issues):

3000ge at $50

3200g at $80

2400g at $95 (as in AF variant / Picasso based)

3100 at $100

3400g at $115

3300x at $150

4300g at $160, and 3600 slighty above.

4600g at $190 and 3600x around that price too.

List prices would typically be $0 to $30 higher.

That would make those price intervals from $50 to $200 pretty well filled.
 

mohit9206

Golden Member
Jul 2, 2013
1,381
511
136
I think the 4300g and 4600g will supplement rather than replace the 3200g+3400g, at least for over another year until another Zen3 APU can fill that lower end gap that 3200g is at.

To guess what is still produced look at what laptops are still in demand and selling. We have Raven2 and Picasso filling the low end to mid mainstream laptop processors (while Renoir is filling the mid to upper mainstream and elite notebooks. Gradually Picasso will be phased out in favor of Raven2 (just like Bristol Ridge was phased out for Stoney ridge). Around that time, a new Zen3 (maybe Zen2) mobile focused APU will replace Picasso, and soon after that we'll have our 3200g replacement derived from that.

On desktop we have Athlon 3000ge at $50 , and likely by time Renoir arrives on AM4 the 3200g / 3400g will shift down in price and will complement in filling the price ranges with Renoir; my guess for street prices (after any initial retailer gauging / supply issues):

3000ge at $50

3200g at $80

2400g at $95 (as in AF variant / Picasso based)

3100 at $100

3400g at $115

3300x at $150

4300g at $160, and 3600 slighty above.

4600g at $190 and 3600x around that price too.

List prices would typically be $0 to $30 higher.

That would make those price intervals from $50 to $200 pretty well filled.
May i know why do you think the 4300G will be $160 and not like say around $120?
 

Olikan

Platinum Member
Sep 23, 2011
2,023
275
126
Some gaming benchmarks, Raven ridge 11CU slight beat to Renoir 7CU at same memory, within error margin, 7CU Renoir wins in the end by better memory overclock...

Raven ridge, is no match to full Renoir 8CU, even at lower memory speed...

 

Shivansps

Diamond Member
Sep 11, 2013
3,873
1,527
136
Some gaming benchmarks, Raven ridge 11CU slight beat to Renoir 7CU at same memory, within error margin, 7CU Renoir wins in the end by better memory overclock...

Raven ridge, is no match to full Renoir 8CU, even at lower memory speed...


That is exactly what we expected a few months ago, that the Vega 7 at 1.9Ghz will be head to head with Vega 11 at 1.4Ghz.

And we can already see what is going to happen, you will have simple games like LOL that will run better even on Vega 6 due to the high clock, and games like PU Battlegrounds that are more shader intensive that is GPU bottlenecked by Vega 6 and it takes Vega 7 at 1.9Ghz with DDR4-3600 to match Vega 11 1.4ghz with just DDR4-3200.

As i said from the beginning, is not always about memory bandwidth. Also, look at the 4750G numbers with DDR4-3200, they are excellent across the board.

These stock numbers for the 4350G are unacceptable as a 3400G replacement that much is clear. the 4650G im going to lower the bar and say it is ok, even with the slight perf loss on DDR4-3200.
 
Last edited:

Gideon

Golden Member
Nov 27, 2007
1,710
3,927
136
Well it seems thst these are on premise and for sale at the biggest Estonian retailer!

Prices include 20% VAT. CPUs are OEM/BULK but do come with a cooler it seems:

4350G -145€
4650G - 192.90€
4750G - 309.90€

The last two seem pretty good value, considering 3600 costs 194 € and 3700x is 329.90€. 4350G could be a tad cheaper as 3400G is 147€ and 3200G is 92.36€.

To me 4650G looks particularily good value for ultra budget builders, cause unlike 3400G, you could couple it with a decent GPU down the line and still get excellent performance. 3400G would be CPU limited in quite a few heavily multithreaded titles
 

moinmoin

Diamond Member
Jun 1, 2017
4,994
7,765
136
Well it seems thst these are on premise and for sale at the biggest Estonian retailer!

Prices include 20% VAT. CPUs are OEM/BULK but do come with a cooler it seems:

4350G -145€
4650G - 192.90€
4750G - 309.90€

The last two seem pretty good value, considering 3600 costs 194 € and 3700x is 329.90€. 4350G could be a tad cheaper as 3400G is 147€ and 3200G is 92.36€.

To me 4650G looks particularily good value for ultra budget builders, cause unlike 3400G, you could couple it with a decent GPU down the line and still get excellent performance. 3400G would be CPU limited in quite a few heavily multithreaded titles
Looks very good, personally I'd even prefer Pro chips if they were normally available to consumers (I'd love to have the capability of encrypting RAM). I just wish there were more info on what particular cooler is included.
 

FriedMoose

Member
Dec 14, 2019
48
28
51
Some gaming benchmarks, Raven ridge 11CU slight beat to Renoir 7CU at same memory, within error margin, 7CU Renoir wins in the end by better memory overclock...

Raven ridge, is no match to full Renoir 8CU, even at lower memory speed...

So it seems like four times the L3 cache makes a much bigger difference than a monolithic die. You have to greatly overclock both core and memory to match a stock 3700X.
 
Reactions: Tlh97 and Olikan

amd6502

Senior member
Apr 21, 2017
971
360
136
May i know why do you think the 4300G will be $160 and not like say around $120?

Look at what the 3300x is retailing for. Relative to the bottom bin 3100 ($99) it was severely underpriced; supply could not keep up with demand at the suggested retail price of $120, unless the binning process were to get very wasteful and use mostly half disabled but fully functional dual CCX dies. Hence the price on the 3300x should have been closer to the street price of the 3600 ($164), say $140. Add a Vega 6 iGPU and that bumps it up to around $160.

On second thought, it looks like the 4300g wil be the same 2+2 layout as the 3100, except for better frequencies, 4GHz boost and 200MHz higher base clock. The 4350g will likely be 4+0 layout with supposed 4.1GHz boost and same base as 4300g.

So I change my guess, the 4350g will be priced where the 3600 is at now ($164), and the 4300g I guess to be at $150 (which is exactly where the 3400g is sitting at right now).

I see it as much less likely that the 4300g would be priced far below (<$140) the 3600 Matisse value sweet spot (which by launch would be $160). For a sub $140 pricing on the 4300g, the yields for Renoir would have to be really poor, or the frequency bar for the 4600g cores would have been set way too high---this doesn't seem to be the case (they are supposedly 3.7GHz/4.2GHz). So $150 sounds right.

For $140 and below pricing there will be plenty of choices. 2600, 3300x, 3400g, 1600AF, 3100, 3200g, 3000ge

(I think it would also eventually make sense to add a bottom bin SKU for Renoir. Same freqs as 3100, except 2 or 3CU of GPU, priced at $110. Or otherwise, just disable GPU and add this to 3100 production.)
 
Last edited:

Shivansps

Diamond Member
Sep 11, 2013
3,873
1,527
136
So it seems like four times the L3 cache makes a much bigger difference than a monolithic die. You have to greatly overclock both core and memory to match a stock 3700X.

That is a huge suprise to me, because this bad:



may even be worse than Zen+ bad. I really expected these APU to fare well in dGPU gaming... I really want to see CPU tests of the 3400G vs the 4300G now because it may not be that much faster.
 

maddie

Diamond Member
Jul 18, 2010
4,787
4,771
136
That is a huge suprise to me, because this bad:



may even be worse than Zen+ bad. I really expected these APU to fare well in dGPU gaming... I really want to see CPU tests of the 3400G vs the 4300G now because it may not be that much faster.
Yes. Assuming these results are replicated and confirmed, the prices for these new APUs are way too high.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
may even be worse than Zen+ bad. I really expected these APU to fare well in dGPU gaming... I really want to see CPU tests of the 3400G vs the 4300G now because it may not be that much faster.

We'll see.

If you go back to the results, there are iGPU comparisons.

Except, it could also be a good CPU comparison. On Overwatch its getting over 60 fps, and in League of Legends, over 100. Maybe the faster CPU is also playing a part in why the Vega 6 in Ryzen 4000 is competitive with Vega 11 in Ryzen 3000, and using slower memory.

Alternate explanation is that Cerny's claims about clock speeds playing a bigger part than CUs being right. It does make sense. I mean clock speeds benefit everything while CUs only a subset.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,729
136
I really expected these APU to fare well in dGPU gaming...
Not surprising at all. The IMC on Renoir really doesn't do much to improve on the latency compared to Matisse. In fact it even regresses according to initial investigations. On top of that 1/4th L3 cache is going to hurt a lot.

Looks like AMD really dropped the ball quite hard when it comes to memory latency on Renoir. I did not expect a regression compared to Matisse.
 

arandomguy

Senior member
Sep 3, 2013
556
183
116
So it seems like four times the L3 cache makes a much bigger difference than a monolithic die. You have to greatly overclock both core and memory to match a stock 3700X.

I don't know if it's necessarily the 4x that makes the difference due to diminishing returns but they were likely looking to really keep sizes/power down but not increasing L3 at all with Renoir. Zen/+ doubled L3 to Zen 2 on non APUs. APU cache remained static at 2x4MB, which is really on the relatively small side given the core/thread count.

Not surprising at all. The IMC on Renoir really doesn't do much to improve on the latency compared to Matisse. In fact it even regresses according to initial investigations. On top of that 1/4th L3 cache is going to hurt a lot.

Looks like AMD really dropped the ball quite hard when it comes to memory latency on Renoir. I did not expect a regression compared to Matisse.

I'd speculate it is trade off with decoupling IF speed to memory speed versus the fixed ratio on all other designs up to this point .
 

arandomguy

Senior member
Sep 3, 2013
556
183
116
It should not decouple at 3600MHz like in the review link I posted above.

I thought Renoir's IF clock operates differently in that IF and memory speeds are decoupled. IF isn't being run at fixed ratio to memory regardless of memory speed.

This isn't about the ratio multiplier switch that Zen 2 CPUs have above 3600.

Coming to the Infinity Fabric, AMD has made significant power improvements here. One of the main ones is decoupling the frequency of Infinity Fabric from the frequency of the memory – AMD was able to do this because of the monolithic design, whereas in the chiplet design of the desktop processors, the fix between the two values has to be in place otherwise more die area would be needed to transverse the variable clock rates. This is also primarily the reason we’re not seeing chiplet based APUs at this time. However, the decoupling means that the IF can idle at a much lower frequency, saving power, or adjust to a relevant frequency to mix power and performance when under load.


Their might be design trade offs to achieve that in which they don't want to disclose. In this case claw back (or more) of any latency gains moving to a single chip. It's ultimately a power oriented design objective foremost.
 
Reactions: spursindonesia

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,729
136
I thought Renoir's IF clock operates differently in that IF and memory speeds are decoupled. IF isn't being run at fixed ratio to memory regardless of memory speed.
That makes sense for low power laptop parts, not desktop. If that is indeed the case then it's a big drawback.
 

DrMrLordX

Lifer
Apr 27, 2000
21,797
11,144
136
That makes sense for low power laptop parts, not desktop. If that is indeed the case then it's a big drawback.

You would think coupled/decoupled mode would be tweakable at any available IF speed assuming everything was running synced up (eg. DDR4-3600, IF 1800) when using Renoir on the desktop.
 

jpiniero

Lifer
Oct 1, 2010
14,832
5,444
136
I suppose there could be an issue with it being PCIe 3.0x8.


Videocardz got the official MSRP. $309 for the 7, $209 for the 5, and $149 for the 3.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |