• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."
  • Community Question: What makes a good motherboard?

Zen 2 APUs/"Renoir" discussion thread

Page 10 - 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
3,110
786
136
This PCI-E issue seems to be contradicting what all the B550 motherboard webpages seems to be indicating at the moment.

1‎ x PCI Express x16 slot (PCIEX16), integrated in the CPU:
  1. 3rd Generation AMD Ryzen™ processors support PCIe 4.0 x16 mode
  2. New Generation AMD Ryzen™ with Radeon™ Graphics processors support PCIe 3.0 x16 mode
    * For optimum performance, if only one PCI Express graphics card is to be installed, be sure to install it in the PCIEX16 slot.
A lot of boards are using the main 16x split to wire extra M2 NVME slots and they promise they they would work at 3.0 if "new gen APU" is installed. If AMD changed stuff at the last moment again i see a huge issue coming up.
 

Asterox

Senior member
May 15, 2012
415
556
136
It isn't all that unusual for a GPU. My 1060 has been running at 2.05 GHz since '16.

It's insane for an APU though.
Especially it is still a Vega GPU, squeezed into a 65W TDP frame.

"Well we can compared that to a slightly bigger APU"(Sony PS5), powered by RDNA 2 GPU or above 2ghz GPU.
 

leoneazzurro

Senior member
Jul 26, 2016
329
437
136
The chip itself has 16 available PCI 3.0 lanes. Configurations are likely to be 1x8, 1x4, 1x4, 2x4, 1x4, 1x4 with the first the lanes dedicated to the external VGA cards, the second to NVME and the third to other. The chipset then will follow its specification: if it's a B550 it wil likely have 4x general purpose PCI 3.0 lanes and either 2 additional PCI3.0 lanes or 2 SATA lines. The APU has 4 less PCI lanes than a 3000 series CPU, so it will not have the full 1x16 or 2x8 configuration for graphics cards and the lanes from CPU are 3.0 instead of 4.0 because of the mobile power requirements. BTW, if a motherboard maker reports incorrect information, or it refers to a future unannounced product, I don't see how AMD is to blame. What these APUs are capable of, and what they aren't capable of, was clear since day 1 of the architecture description.
 
Last edited:

Geranium

Member
Apr 22, 2020
73
95
61
"All the APUs support DDR4-3200, and have eight PCIe 3.0 lanes. On the PCIe lanes, this is because the mobile chip was built with eight PCIe 3.0 lanes, to save on power in a mobile environment. AMD sees these chips being used mostly on their own without a separate discrete graphics card, given that the company already has the Ryzen 3 CPU family for those that want discrete graphics. "

So they cut PCIe lanes from 20 on the 3400G to only 8 on the 4700G? This has to be x8 3.0 for the GPU and x4 3.0 for storage, right? Perhaps an oversight by the author?
You are getting it wrong. Both Raven Ridge(2400G) and Picasso(3400G) has 16 PCI-e 3.0 lane in 8+4+4 Configuration, not 20.
New Renoir has 20 PCI-e 3.0 lane in 8+4+4 and +4 configuration. The extra 4 lane is for connecting Wifi 6 or 10G LAN or Thunderbolt(if Intel ever opens it) directly to CPU. Though it looks like only 16 lane activate for desktop for compatibility reasons.
 
Last edited:

FriedMoose

Member
Dec 14, 2019
48
28
51
You are getting it wrong. Both Raven Ridge(2400G) and Picasso(3400G) has 16 PCI-e 3.0 lane in 8+4+4 Configuration, not 20.
New Renoir has 20 PCI-e 3.0 lane in 8+4+4 and +4 configuration. The extra 4 lane is for connecting Wifi 6 or 10G LAN or Thunderbold(if Intel opens it) directly to CPU.
Wikichip lists the 3400G as having 20 PCIe 3.0 lanes:

 

Geranium

Member
Apr 22, 2020
73
95
61
Wikichip lists the 3400G as having 20 PCIe 3.0 lanes:

That looks like wrong.
Official AMD specs sheet : https://www.amd.com/en/products/specifications/embedded/8191+11961
Raven Ridge PCIe Lane Count.PNG
Enterprise ones have most PCI-e active.

https://www.amd.com/en/products/apu/amd-ryzen-5-3400g
Even there it says PCI-e 8x for general purpose.
 

Red Hawk

Diamond Member
Jan 1, 2011
3,269
168
106
NO, you are completely wrong. First off the second number was never about the number of cores or threads, your own examples are wrong, the 1300X and 2300X are 4/4 CPUs, not 4/8, then you forget about the Ryzen 3 3100 that is a 4/8 CPU.

Also who said that you cant have a 4/8 4200G when you have a 4/4 3200G??, that is called PROGESS, the 2nd number was never about cores/threads, it is used to indicate what SKU is better than the other in the same generation, thats it.
What you cant have is a $140 4200G and a $220 4400G because everyone will notice what you are doing.

and finally, there is nothing wrong with the 4400G naming considering these APU will share naming with the Zen3 CPU, the problem only happens when you want to have a 4400G that is more expensive than a 4600X. The 4400 indicates that is the lowest of the Ryzens 5, and would be correct.
Ok sure, the -300 moved from 4c/4t to 4c/8t with the 3300X. The 4300G still fits that. And yeah, these CPUs don't represent a big step like last year's Ryzen 3rd gen. At worst, that makes them a bit boring and disappointing. But there's no real room to rag on the naming and price, not when you have to pay a bunch more for comparable performance and features from Intel.
 
Last edited:
  • Like
Reactions: Tlh97 and Thibsie

Shivansps

Diamond Member
Sep 11, 2013
3,110
786
136
Ok sure, the -300 moved from 4c/4t to 4c/8t with the 3300X. The 4300G still fits that. And yeah, these CPUs don't represent a big step like last year's Ryzen 3rd gen. At worst, that makes them a bit boring and disappointing. But there's no real room to rag on the naming and price, not when you have to pay a bunch more for comparable performance and features from Intel.
The point is that, while i do agree that 4600G fits better for the 6C/12T APU, there is nothing wrong with keeping the 4400G name, and there is zero reasons for changing the 4200G name to 4300G because these number never had anything to do with core or thread number.
 

uzzi38

Golden Member
Oct 16, 2019
1,227
2,263
96

Korean review, but shows something very interesting.

Even the 4650G with Vega7 at 1900mhz is hugely bottlenecked by memory bandwidth.

DDR4-3200cl22 -> DDR4-4200cl18 with no changes to iGPU clocks shows a 50% perf uplift.

Looks to me like trimming CUs wasn't a bad idea at all
 

Geranium

Member
Apr 22, 2020
73
95
61
Errr.... isn't that basically what is in USB4?

I thought that was because they opened up TB3?

The PHY from TB3 is also in DP2.0 now too.
Thunderbolt is open (sort of) on paper, not on practice. Have you seen any 3rd party controller for TB3/TB4, like 3rd party USB controller?
 

Shivansps

Diamond Member
Sep 11, 2013
3,110
786
136

Korean review, but shows something very interesting.

Even the 4650G with Vega7 at 1900mhz is hugely bottlenecked by memory bandwidth.

DDR4-3200cl22 -> DDR4-4200cl18 with no changes to iGPU clocks shows a 50% perf uplift.

Looks to me like trimming CUs wasn't a bad idea at all
Whiout seeing the details of each game i cant be sure, but those numbers look too low for 3200mhz.

As for Firestrike, it is just 400 point over the stock 3400G with DDR4-3200
 

uzzi38

Golden Member
Oct 16, 2019
1,227
2,263
96
Whiout seeing the details of each game i cant be sure, but those numbers look too low for 3200mhz.

As for Firestrike, it is just 400 point over the stock 3400G with DDR4-3200
Fire Strike is a poor indicator of performance when memory bottlenecked. Most synthetics tend to be, the only thing I ever trust is Time Spy Graphics Test 2. Even GT1 scales with core clocks on memory bound GPUs.
 

Shivansps

Diamond Member
Sep 11, 2013
3,110
786
136
Fire Strike is a poor indicator of performance when memory bottlenecked. Most synthetics tend to be, the only thing I ever trust is Time Spy Graphics Test 2. Even GT1 scales with core clocks on memory bound GPUs.
Interesting, do you know if there are the details settings of the games tested somewhere? Because 30fps in Fortnite, even at 1080p medium is a bad result for 3200mhz ram.
 

uzzi38

Golden Member
Oct 16, 2019
1,227
2,263
96
Interesting, do you know if there are the details settings of the games tested somewhere? Because 30fps in Fortnite, even at 1080p medium is a bad result for 3200mhz ram.
Unfortunately I don't speak Korean, so no
 

soresu

Golden Member
Dec 19, 2014
1,535
762
136
Thunderbolt is open (sort of) on paper, not on practice. Have you seen any 3rd party controller for TB3/TB4, like 3rd party USB controller?
It's not long been done as things go - it's into new specs now which are on their way.

USB4 and DP2 are incoming.

I wouldn't expect an open TB4 controller though.
 

Olikan

Platinum Member
Sep 23, 2011
2,003
234
106
"For iGPU, expects 2500mhz OC on average, he showed 2650mhz"


This smells good for RDNA2 cards
 

leoneazzurro

Senior member
Jul 26, 2016
329
437
136
When the CPU is the bottleneck sure.

But in all of the above cases, the GPU is almost certainly the bottleneck.
I mean, latency is important for the effective bandwitdth of the iGPU, too. That because the improvements in the scores are higher than the simple difference in pure bandwidth.
 

Shivansps

Diamond Member
Sep 11, 2013
3,110
786
136
"For iGPU, expects 2500mhz OC on average, he showed 2650mhz"


This smells good for RDNA2 cards
oof the 4750G with DDR4-3600 rams is only 7% faster on average than the 3400G?

The memory results shows that Vega 8 at 2100mhz is not longer memory starved at DDR4-3800. At least for Fire Strike, i would like to see some games.
 
Last edited:

IntelUser2000

Elite Member
Oct 14, 2003
7,249
1,839
136
OUCH the 4750G with DDR4-3600 rams is only 7% faster on average than the 3400G?

The memory results shows that Vega 8 at 2100mhz is not longer memory starved at DDR4-3800.
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.
 
  • Like
Reactions: lightmanek

amd6502

Senior member
Apr 21, 2017
830
274
106
DrMrLordX said:

What was AMD's net margin in 2015? 2016?
Losing a LOT of money. %-21 and %-8.

Or breaking even while sinking sh*tloads into Zen R+D. Money that won't be earned back again until years later.

First off the second number was never about the number of cores or threads, your own examples are wrong, the 1300X and 2300X are 4/4 CPUs, not 4/8, then you forget about the Ryzen 3 3100 that is a 4/8 CPU.
True, not officially. However, usually so. 3100 is an exception because that is based on bottom of the barrel near-discard die salvage and has increased core to core latencies due to the 2+2 core layout. Even so, AMD marketing screwed up once again (no big surprise) as it would have been much more aptly named 3250 or so.

Let's be honest: AMD isn't prioritizing die real estate to iGPUs right now. It's not clear when they'll go back in that direction, if ever.
Not clear but very likely in sync with 1. Zen3, and 2. when mainstream memory frequencies catch up. So, if you absolutely must have a bigger iGPU than 3400g or 8CU Renoir, then wait for Zen3 APU ~ H1 2021.
 
  • Like
Reactions: Tlh97

ASK THE COMMUNITY