AMD Ryzen 5 2400G and Ryzen 3 2200G APUs performance unveiled

[whm1974]

Well it looks like if you want to run Linux on these APUs, you are going to have jump through some hoops. Or wait a few months first.
https://www.phoronix.com/scan.php?page=article&item=ryzen5-2400g-vega11&num=1

 

[raghu78]

For a new budget build, the Ryzen APUs perform within 1% overall of the GT 1030 and offers a dGPU upgrade path. I would classify this as a TOTAL WIN for budget gamers building a new system. As the (mostly) proud owner and user of two GT 1030-based systems (SFF i5-2500(non-K) 4GB and SFF i5-3470 8GB), I feel the need to defend the GT 1030 as well as the entire concept of used/low-spec gaming. I bought the GT 1030 primarily for playing back my HDR UHD Blu-ray backups from my server, but I sometimes play Destiny 2 from the couch, Diablo 3, Rocket League, watch Dota 2 matches in-game, etc.

For example, I can play Destiny 2 at 1080p (no scaling) with HDR and a mix of low-medium settings between 32fps (heavy action) and 50fps (nothing happening, no one around). Destiny 2 is pretty well optimized for such a gorgeous-looking and fast-paced MMO. PUBG, this is NOT. If I frame-limit either system to 30fps, the gameplay feels identical to playing it on PS4/Xbone. The GT 1030 is, in my opinion, extremely underrated for a silent, low profile, $69 GPU that also includes full HEVC/HDR acceleration and DP1.4/HDMI2.0. It's a great little card to add to an older system, but it doesn't solve the problem of those older Sandy/Ivy systems lacking NVME, USB 3.1, etc.

With Ryzen APUs, you get the above experience plus better overall CPU performance, overclocking, modern IO, and a dGPU upgrade path for a low cost of entry. It is really fantastic. Even with DDR4-3200, I was able to build a budget system for under $300. I'd probably splurge in a few places, but I think I could still do it for less than $400.

Very well said. People do not understand that in countries like India a lot of PCs are sold for INR 20000 - 30000(USD 300 - USD 450). These PCs are used by kids and teens who play games like Dota 2, CS:GO, GTA V, Rocket League which can easily be played on these APUs at medium-high settings even at 1080p. In fact the APUs can play the more demanding games like Witcher 3, Wolfenstein II, Rise of the Tomb Raider at 720p medium settings. imo these APUs are perfect for budget gaming at 720p.

 

[daveybrat]

So my question is are there any motherboards that come with a bios new enough to support these apu's? If i were to buy a normal B350 motherboard, will it at least post with these new apu's or do i need an older cpu to flash the bios first?

 

[24601]

Are there any plans for any triple or quad channel DDR4 APUs for Ryzen Refresh or Ryzen 2?

The DDR4 price situation should abate by Ryzen 2 timeframe from Chinese fabs starting high volume production of DDR4.

Chinese fabs won't have HBM ready so that's going to stay gouge-arific from the memory fab cartel.

The problem is that by Ryzen 2 timeframe, 2 channel DDR4 APU will be 100% pointless once again compared to just buying dGPU.

The GDDR5 channels for Kaveri were pointless due to bulldozer, but having wider memory bus for Ryzen(2) + Vega/Navi would probably still be a good idea.

 

[raghu78]

Are there any plans for any triple or quad channel DDR4 APUs for Ryzen Refresh or Ryzen 2?

The DDR4 price situation should abate by Ryzen 2 timeframe from Chinese fabs starting high volume production of DDR4.

Chinese fabs won't have HBM ready so that's going to stay gouge-arific from the memory fab cartel.

The problem is that by Ryzen 2 timeframe, 2 channel DDR4 APU will be 100% pointless once again compared to just buying dGPU.

You will never see a 3 or 4 channel memory system on AMD APUs. The cost and complexity of motherboards will go up and its just not worth it. The way AMD will solve the bandwidth problem is with HBM2. When 7nm APUs arrive in H1 2020 the HBM2 memory and packaging tech would have become cost effective and suitable for high volume mainstream.

Samsung made enormous capex investment for DRAM in 2017 to drive DDR4, HBM2 and GDDR6 volumes up in 2018 and you will see that play out. Chinese fabs are never going to be at the cutting edge of memory production. In fact nobody will get close to Samsung in DRAM. Just look at the GDDR6 built at 10nm class Samsung DRAM process vs the Hynix GDDR6 built on 20nm class process.

Samsung GDDR6 - 18 Gbps , 16 Gb
Hynix GDDR6 - 12-14 Gbps, 8 Gb

https://www.anandtech.com/show/12338/samsung-starts-mass-production-of-gddr6-memory
https://www.anandtech.com/show/12345/sk-hynix-lists-gddr6-memory-as-available-now

Samsung GDDR6 has 15-30% higher speeds and double the capacity of Hynix GDDR6. This is what that tremendous USD 7 billion DRAM capex in 2017 is doing. It pretty much kills Samsung's competition and makes them fight for leftover and lower cost/margin chips.

http://www.icinsights.com/news/bull...i-capex-deliver-knockout-blow-to-competition/

Bill McClean, president of IC Insights stated, “In my 37 years of tracking the semiconductor industry, I have never seen such an aggressive ramp of semiconductor capital expenditures. The sheer magnitude of Samsung’s spending this year is unprecedented in the history of the semiconductor industry!”

http://www.icinsights.com/data/articles/documents/1025.pdf

"Samsung’s current spending spree is also expected to just about kill any hopes that Chinese companies may have of becoming significant players in the 3D NAND flash or DRAM markets. As our clients have been aware of for some time, IC Insights has been extremely skeptical about the ability of new Chinese startups to compete with Samsung, SK Hynix, and Micron with regards to 3D NAND and DRAM technology. This year’s level of spending by Samsung just about guarantees that without some type of joint venture with a large existing memory suppler, new Chinese memory startups stand little chance of competing on the same level as today’s leading suppliers."

Samsung is the new Intel of the semiconductor industry. They are going to dominate DRAM and NAND and will grow and strengthen foundry over the next 3-5 years.

 

[24601]

Memory fab cartel always sets gouge-ing prices at 300-700% markup until Chinese fabs catch up to the same technology.

Chinese fabs won't catch up to HBM(2) by Ryzen 2 timeframe, so HBM(2) will be whatever the memory fab cartel wants to price it (aka, completely out of the realm of usefulness in a budget APU).

AMD clearly doesn't want to use GDDR on their budget APUs due to un-palatability of the price marketing when you cannot exclude memory prices from the cost a dumb uninformed end user will see.

That leaves triple and quad channel DDR4 as the only options left.

Unless AMD just plans on letting their non-semi-custom APUs die again.

 

[xblax]

So my question is are there any motherboards that come with a bios new enough to support these apu's? If i were to buy a normal B350 motherboard, will it at least post with these new apu's or do i need an older cpu to flash the bios first?

Gigabyte has released a few new mainboards this February. They come with Raven Ridge compatible bioses out of the box. However, It seems like they are only available in Europe yet.

https://www.reddit.com/r/Amd/comments/7x0zjq/psa_list_of_mainboards_with_guaranteed_raven/

 

[raghu78]

Memory fab cartel always sets gouge-ing prices at 300-700% markup until Chinese fabs catch up to the same technology.

Chinese fabs won't catch up to HBM(2) by Ryzen 2 timeframe, so HBM(2) will be whatever the memory fab cartel wants to price it (aka, completely out of the realm of usefulness in a budget APU).

AMD clearly doesn't want to use GDDR on their budget APUs due to un-palatability of the price marketing when you cannot exclude memory prices from the cost a dumb uninformed end user will see.

There is no cartel going to be left. Samsung has already 45% of worldwide DRAM market share. With the massive DRAM capex they are going to be more than 60% market share very soon.

https://www.dramexchange.com/WeeklyResearch/Post/2/4828.html

The market share situation by the end of 2018 is going to be very different. Samsung is going to kill the competition by the sheer scale and capacity of their DRAM expansion.

 

[24601]

There is no cartel going to be left. Samsung has already 45% of worldwide DRAM market share. With the massive DRAM capex they are going to be more than 60% market share very soon.

https://www.dramexchange.com/WeeklyResearch/Post/2/4828.html

The market share situation by the end of 2018 is going to be very different. Samsung is going to kill the competition by the sheer scale and capacity of their DRAM expansion.

How can you be so dense?

How does that allow AMD to acquire the high bandwidth memory at the price they need it at to make the margins they need at the price they can sell their product?

Do you not even know what I am talking about?

Do you know what a cartel is?

How would having less players lower the price?

 

[raghu78]

How can you be so dense?

How does that allow AMD to acquire the high bandwidth memory at the price they need it at to make the margins they need at the price they can sell their product?

Do you not even know what I am talking about?

Do you know what a cartel is?

How would having less players lower the price?

You have a poor understanding of the challenges with current HBM2 tech and the concept of economies of scale. HBM2 technology has 2 barriers - one is complexity and hence lower yield / higher cost (requirement of TSVs) and the other is low volume and hence higher cost. The first is a packaging tech challenge which the foundries and OSATs are already working on. The second is a volume and scale problem. By 2020 you will have cheaper packaging tech without need for TSVs. Amkor is already well into SLIM and SWIFT development to replace TSVs.

http://www.semicontaiwan.org/zh/sit.../data15/docs/7_5._tsv_less_semicon_taiwan.pdf
http://ieeexplore.ieee.org/document/7999664/

The second challenge of volume is going to be solved by the enormous ramp of DRAM production volumes by Samsung especially on bleeding edge tech like HBM2 and GDDR6. As the production volumes increase massively and the economies of scale start to play the cost will also come down over time. I am betting by H1 2020 we will see both HBM2 memory and packaging tech become cost effective and suitable for high volume mainstream APUs.

 

[24601]

You have a poor understanding of the challenges with current HBM2 tech and the concept of economies of scale. HBM2 technology has 2 barriers - one is complexity and hence higher cost (requirement of TSVs) and the other is low volume and hence higher cost. The first is a packaging tech challenge which the foundries and OSATs are already working on. The second is a volume and scale problem. By 2020 you will have cheaper packaging tech without need for TSVs. Amkor is already well into SLIM and SWIFT development to replace TSVs.

http://www.semicontaiwan.org/zh/sit.../data15/docs/7_5._tsv_less_semicon_taiwan.pdf
http://ieeexplore.ieee.org/document/7999664/

The second challenge of volume is going to be solved by the enormous ramp of DRAM production volumes by Samsung especially on bleeding edge tech like HBM2 and GDDR6. As the production volumes increase massively and the economies of scale start to play the cost will also come down over time. I am betting by H1 2020 we will see both HBM2 memory and packaging tech become cost effective and suitable for high volume mainstream APUs.

I give up talking to you Samsung stock holder/flogger.

 

[raghu78]

I give up talking to you Samsung stock holder/flogger.

Your problem is a lack of understanding of what ails the adoption of HBM2 tech today and how the OSATs and foundries are already working to solve it. Samsung will play a crucial role in the adoption of current bleeding edge HBM2 memory in high volume mainstream products by 2020. So will the OSATs like Amkor. Adios amigo.

 

[CHADBOGA]

The second challenge of volume is going to be solved by the enormous ramp of DRAM production volumes by Samsung especially on bleeding edge tech like HBM2 and GDDR6. As the production volumes increase massively and the economies of scale start to play the cost will also come down over time. I am betting by H1 2020 we will see both HBM2 memory and packaging tech become cost effective and suitable for high volume mainstream APUs.

So non-Intel monopolies bring only joy to consumers?

 

[goldstone77]

I give up talking to you Samsung stock holder/flogger.

raghu78 is correct in his assessment's. With all the posting and talk of TSMC, GlobalFoundries, and Intel people have neglected to see what Samsung has been doing. They have the most aggressive process node roadmap as well. They will attempt EUV at 7nm this year, and if they are able to do that they will have the lead on the competition and clear path to 3nm!

 

[tamz_msc]

So non-Intel monopolies bring only joy to consumers?

Hmm, so why doesn't Intel do anything about the memory situation?

 

[goldstone77]

Hmm, so why doesn't Intel do anything about the memory situation?

They tried with 3D Xpoint, and it didn't go anywhere. They have bigger issues with just trying to make 10nm happen.

 

[frozentundra123456]

So non-Intel monopolies bring only joy to consumers?

Yea, pretty ironic isnt it.

 

[3DVagabond]

And that's why the cheaper 2200G looks more attractive from a budget stance...

The money saved can go into the ram.

Well, the 2400 sold out 1st but now the 2200 is back ordered too. It's really semantics arguing what's the better deal. They're both selling like penny candy.

 

[OrangeKhrush]

I am looking at catching it while the iron is hot. A 1200 is locally priced for me at 2300K while the 2400G is 2300K also and significantly better. I will give it to my young step brother who is now starting to get old enough to understand gaming. the titles he plays also fall hook line and sinker into the Raven Ridge domain.

 

[OrangeKhrush]

So non-Intel monopolies bring only joy to consumers?

Totally unrelated but this is a blast from the past, back in the day when Toms Forums were active and people actually cared to go there. I just remembered the avatar

 

[SirDinadan]

Gigabyte has released 4 BIOSes in a week time for the GA-AB350N-Gaming WIFI. Why there are so many and which one should I go for. Thx!

 

[The Stilt]

Gigabyte has released 4 BIOSes in a week time for the GA-AB350N-Gaming WIFI. Why there are so many and which one should I go for. Thx!

T20h is the review bios, F22b is the public version of the same.

 

[FlanK3r]

Roger,
finally I got 2400G alos. But my board is only C6H (where are the outputs) and next candidate is propably TUF B350M-Plus Gaming. What do you think about this board?

Do you want to write deep OC Guide how to tweak Raven Ridge? I like read your things man ,-)

 

[SirDinadan]

T20h is the review bios, F22b is the public version of the same.

Thank you! So technically, they should be the same - F22b is listed but the file is not there.

 

[TheELF]

@The Stilt just run a virtual drive on your ram,use 2 8gb modules and make a 8Gb drive,should be close enough to dual channel 4Gb.
Starwind ram disk is free and good.