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

[IRobot23]

Choice is the key work. For many, Bulldozer derivatives were a no go. On the other hand, zen is a very attractive option.

AM4 is very attractive.
You can easily upgrade to 6C and dGPU... but this is now. Raven ridge is only the start.

From the ground zen was design for MCM.

So maybe in 2019 AMD can do chip similar to i7 8808G (HBM+ cpu die + gpu die).

We have to understand that in 2019 7nm node will be here.

 

[Glo.]

We have to understand that in 2019 7nm node will be here.

And if we go by natural shrink and rumored advancement of Zen2 - we are going to get something like 6 core/12Thread CPU with 1536 GCN core GPU. And that has to have HBM memory on package. This APU would replace 2400G.

Which is the reason why AM4 for entry level gaming is the best platform of them all.

 

[Spartak]

Does anybody have a clue what the difference between the 46W and 65W profile will mean for CPU performance? I'm planning to buy the 2400G for my SFF and prefer it to run as silent as possible. Not a gamer so if this basically means that mainly the GPU will be downclocked my experience wouldnt be affected by the lower TDP profile.

 

[Shivansps]

Igp is not there to play AAA games on high resolution, this should be clear, but the fact 2400g could play AAA games at 720p HIGH/ultra is commendable, if you are short on cash it gives you options or a stop gap until a later date, either upgrade the processor or even just add a GPU later on, Ryzen APU gives you that choice, acceptable casual gaming or even some AAA gaming on low resolution.

With intel you have no choice, it's no AAA games or fork out for a GPU.
..oh yes 2400g will easily play Witcher 3 1080p low, probably would be playable with 1080p medium/mix...

How? not even the 1030 cant do that, and the 1030/RX550 are the absolute best case possible.

Nobody spotted that going from 2400 MHz RAM to 3200 MHz brought 10% increase in performance, for 33% increase in memory bandwidth?

It may very well be the sweet spot for the performance/price.

I mentioned like a page ago. Still not enoght, the GT1030 has about the same bandwidth as a DDR4-3200, then again it nearly matches the RX550 that has more than twice the bandwidth,

And if we go by natural shrink and rumored advancement of Zen2 - we are going to get something like 6 core/12Thread CPU with 1536 GCN core GPU. And that has to have HBM memory on package. This APU would replace 2400G.

Which is the reason why AM4 for entry level gaming is the best platform of them all.

Well if you like AM4 platform then go A8-9600 instead of G4560, is not really that much of a difference in perf. A8-9600 w/GT1050 with 8GB SC cheap ram is still A LOT better than a 2200G/2400G with DDR4-3200, and it offers the same upgratibility of AM4.
The Pentium was mentioned as a alternative option with the 1050, but is not the only one.

IF it has 16 ROPs, im expecting 2200G with the iGPU at 1500-1600MHz and 3600MHz memory to match RX550 performance or be faster in some newer games (DX-12/Vulkan) .

That means
BF1 (Single Player) @ 1080p Low = ~60fps (Avg)
Overwatch @ 1080p High = >60fps (Avg)
Rocket league @ 1080p High = > 60fps (Avg)
DOTA 2 @ 1080p High = ~ 60fps (Avg)
Quake Champions @ 1080p Medium = ~ 60fps (Avg)
Witcher 3 @ 1080p Low = 30fps (Avg)

Add a cheap 22-24" FreeSync monitor and this is one of a hell cheap entry level gaming system with tons of features and upgradability path.

Do you how that in order to match a RX550 we are talking about 80 to 100% perf increase over the A12-9800? AMD would have showed that in the slides, for sure. AMD has not matched the performance of a punny R7 250 GDDR5 with their igps yet.

Interesting that AMD showed almost nothing about games in those slides, just a uninteresting IGP fight with a i5-8400 and 3dmark scores that are less affected by bandwidth (AM1 IGP should have kick the ass of Intel IGPs of that era, acording to 3dmark, it never happened because they were starved of bandwidth).
They avoided a direct game comparison with both the GT1030/RX550 AND the A12-9800...

 

[mikk]

Interesting that AMD showed almost nothing about games in those slides, just a uninteresting IGP fight with a i5-8400 and 3dmark scores that are less affected by bandwidth (AM1 IGP should have kick the ass of Intel IGPs of that era, acording to 3dmark, it never happened because they were starved of bandwidth).
They avoided a direct game comparison with both the GT1030/RX550 AND the A12-9800...

This is typical for AMD because AMD GPUs and expecially APUs are more competitive in 3dmark, it's more a GPU computing benchmark rather than real world gaming where bandwidth, TDP budget, driver quality and CPU performance can make big differences. That's why AMDs marketing team usually uses 3dmark in those marketing slides.

 

[tamz_msc]

Well if you like AM4 platform then go A8-9600 instead of G4560, is not really that much of a difference in perf. A8-9600 w/GT1050 with 8GB SC cheap ram is still A LOT better than a 2200G/2400G with DDR4-3200, and it offers the same upgratibility of AM4.
The Pentium was mentioned as a alternative option with the 1050, but is not the only one.

Why would anybody like to be stuck with a Bristol Ridge APU with much slower Construction cores? I thought that consensus from previous discussions on the late availability for Bristol Ridge for desktop was that it was not very viable as an upgrade path if it bought at this juncture?

2200G provides the best upgrade path as of now, especially considering it's 100$ price, even if the iGPU isn't of much use, since it gives you a good quad-core CPU as well.

 

[french toast]

How? not even the 1030 cant do that, and the 1030/RX550 are the absolute best case possible.


I mentioned like a page ago. Still not enoght, the GT1030 has about the same bandwidth as a DDR4-3200, then again it nearly matches the RX550 that has more than twice the bandwidth,



Well if you like AM4 platform then go A8-9600 instead of G4560, is not really that much of a difference in perf. A8-9600 w/GT1050 with 8GB SC cheap ram is still A LOT better than a 2200G/2400G with DDR4-3200, and it offers the same upgratibility of AM4.
The Pentium was mentioned as a alternative option with the 1050, but is not the only one.



Do you how that in order to match a RX550 we are talking about 80 to 100% perf increase over the A12-9800? AMD would have showed that in the slides, for sure. AMD has not matched the performance of a punny R7 250 GDDR5 with their igps yet.

Interesting that AMD showed almost nothing about games in those slides, just a uninteresting IGP fight with a i5-8400 and 3dmark scores that are less affected by bandwidth (AM1 IGP should have kick the ass of Intel IGPs of that era, acording to 3dmark, it never happened because they were starved of bandwidth).
They avoided a direct game comparison with both the GT1030/RX550 AND the A12-9800...

1030? 1080p medium?
Sure it can... ~28fps.
https://youtu.be/bNm8-G8Kz_g

With high performance ram 2400g can play Witcher 3 medium 1080p at console level FPS probably with similar pitfalls (novigrad/crookback bog)

 

[kallisX]

if i remember correctly, previous AMD APU`s significantly clocked down the cpu when gpu power was needed. so performence could be worse than the ryzen they are replacing in certain conditions?

 

[Shivansps]

1030? 1080p medium?
Sure it can... ~28fps.
https://youtu.be/bNm8-G8Kz_g

With high performance ram 2400g can play Witcher 3 medium 1080p at console level FPS probably with similar pitfalls (novigrad/crookback bog)

Are you kidding me? Thats unplayable. I tested that with a 7400 and a 1030, it can manage 30-40 at medium on 720P and thats about it. Hardly something to call home about in a game that is 3 years old already.

A8-9600 with 8GB DDR4-2400 and a 1050 will beat the crap out of 2400G in games at the same price of a 2400G+2x4 3200, plus it has the full 8GB of ram instead of having to share with the IGP. If you dont want the A8-9600, you can have the 2200G+1050, a little more money but it still has more ram as well.

The 2400G may look well in a direct comparison with the 8400, but the fact is no one buys the 8400 to play with the igp -or- the GT1030. Not sure why everyone tries to force the 2200G and specially the 2400G intro something they are not, those new APU will shine on non-gaming Ryzen builds, the same ones that are right now pairing a R3 1200 and R5 1400 with a awful GT710/HD5450, and sometimes RX550/GT1030.

Expending that much money intro attemping to convert 2200G and 2400G intro IGP gaming builds makes not sence outside ITX and Brix-like.

 

[AtenRa]

Do you how that in order to match a RX550 we are talking about 80 to 100% perf increase over the A12-9800?

A12-9800
8 CUs = 512 Processors (1,13GFLOPS at 1108MHz)
32 TMUs at 1108MHz = 35,45 GTexels/s
8 ROPs at 1108MHz = 8,86GPixels/s
up to DDR-4 2400 = 38,4GB/s Bandwidth

Ryzen 2200G (OC to 1550MHz + DDR-4 3600)
8 CUs = 512 Processors (1,58GFLOPS)
32 TMUs at 1550MHz = 49,6 GTexels/s
16 ROPs at 1550MHz = 24,8GPixels/s
DDR-4 3600 = 57,6GB/s Bandwidth

For comparison,

RX550 (1183MHz)
8 CUs = 512 Processors (1,21GFLOPS)
32 TMUs at 1183MHz = 37,86 GTexels/s
16 ROPs at 1183MHz = 18,93,8GPixels/s
GDDR-5 128Bit 7000MHz = 112GB/s Bandwidth

GT1030 (1468MHz boost)
384 Cores = (1,127GFLOPS)
24 TMUs at 1468MHz = 35,23 GTexels/s
16 ROPs at 1468MHz = 23,48,8GPixels/s
GDDR-5 64-Bit 6008MHz = 48,06GB/s Bandwidth

Ryzen 2200G at 1550MHz and DDR-3600 has higher specs than all the above, plus a new and better GPU architecture than A12-9800 and RX550. RX550 has almost double the Memory bandwidth vs the Ryzen 2200G but it doesnt use it because its severely under power for that huge bandwidth.
In paper Ryzen 2200G at 1550MHz and 3600 ram it should even be faster than GT1030.

 

[Glo.]

Well if you like AM4 platform then go A8-9600 instead of G4560, is not really that much of a difference in perf. A8-9600 w/GT1050 with 8GB SC cheap ram is still A LOT better than a 2200G/2400G with DDR4-3200, and it offers the same upgratibility of AM4.
The Pentium was mentioned as a alternative option with the 1050, but is not the only one.

It has nothing to do with "liking" any platform but EVERYTHING with practicality.

If you are on tight budget, you buy AM4, which has cheaper MoBo, very competitive CPUs/APU with Intel offerings, and has something that Intel platform does not have. Future proof. Coffee Lake is the last gen made on 1151 platform, and if you will want to upgrade soon - you will have to buy new MoBo. With AM4 - it will not happen until 2020.

If you want future proof. You can go either: 2400G+ 3200 MHz RAM, and be done with it, or you take 2400 MHz RAM, GTX 1050 and 2200G. And the best thing is the situation I mentioned. If you go with 2400G, and will want to update to Zen2 APU with potentially HBM2 - you immediately get much better hardware.

 

[Insert_Nickname]

Well if you like AM4 platform then go A8-9600 instead of G4560, is not really that much of a difference in perf. A8-9600 w/GT1050 with 8GB SC cheap ram is still A LOT better than a 2200G/2400G with DDR4-3200, and it offers the same upgratibility of AM4.
The Pentium was mentioned as a alternative option with the 1050, but is not the only one.

A8-9600 with 8GB DDR4-2400 and a 1050 will beat the crap out of 2400G in games at the same price of a 2400G+2x4 3200, plus it has the full 8GB of ram instead of having to share with the IGP. If you dont want the A8-9600, you can have the 2200G+1050, a little more money but it still has more ram as well.

You do not want Excavator in a build with dedicated graphics. Excavator is not really suited as a gaming CPU*, so BR shouldn't be considered unless you're using the IGP. Besides, why would you use a 9600 when the AM4 Athlon 950 is available, clocked higher and cheaper? That way you can get into the AM4 platform goodness, without too much investment.

*Even the older FM2+ Kaveri Athlons are faster.

 

[Peter Watts]

2200G directly competes against Core i3 8100 and not Pentium G4560/4600 etc.

80-90% of EOM systems (SFF, USFF from HP, Lenovo etc) are shipping with iGPUs only. For home/Office use those two CPUs are directly competitive.
Until the 2C 4T + 3CU Ryzen SKUs release, Pentium has no real competition (unless you take Bristol Ridge 9600/9700).

Also to add that both 2200G and 2400G are direct competitors against Intels NUC based systems.

You seem to have a lot of experience with APU´s just looking at your signature, do you think it´s worth the money to buy the 2400G over the 2200G? (Of course they aren´t out yet and benchmarks should be done first.... But theoretically speaking). 169 vs 100 dollars is a significant price difference in my humble opinion, and maybe that 69 dollars could be better spent buying dual channel ram. (Speaking for myself and most likely other builders on a budget).

 

[mohit9206]

A12-9800
8 CUs = 512 Processors (1,13GFLOPS at 1108MHz)
32 TMUs at 1108MHz = 35,45 GTexels/s
8 ROPs at 1108MHz = 8,86GPixels/s
up to DDR-4 2400 = 38,4GB/s Bandwidth

Ryzen 2200G (OC to 1550MHz + DDR-4 3600)
8 CUs = 512 Processors (1,58GFLOPS)
32 TMUs at 1550MHz = 49,6 GTexels/s
16 ROPs at 1550MHz = 24,8GPixels/s
DDR-4 3600 = 57,6GB/s Bandwidth

For comparison,

RX550 (1183MHz)
8 CUs = 512 Processors (1,21GFLOPS)
32 TMUs at 1183MHz = 37,86 GTexels/s
16 ROPs at 1183MHz = 18,93,8GPixels/s
GDDR-5 128Bit 7000MHz = 112GB/s Bandwidth

GT1030 (1468MHz boost)
384 Cores = (1,127GFLOPS)
24 TMUs at 1468MHz = 35,23 GTexels/s
16 ROPs at 1468MHz = 23,48,8GPixels/s
GDDR-5 64-Bit 6008MHz = 48,06GB/s Bandwidth

Ryzen 2200G at 1550MHz and DDR-3600 has higher specs than all the above, plus a new and better GPU architecture than A12-9800 and RX550. RX550 has almost double the Memory bandwidth vs the Ryzen 2200G but it doesnt use it because its severely under power for that huge bandwidth.
In paper Ryzen 2200G at 1550MHz and 3600 ram it should even be faster than GT1030.

You don't know that. You don't know whether the rx550 can or cannot use that extra available bandwidth.
Also I'm pretty sure no way 2200g igpu can be faster than 1030.its straight up impossible even if u use 5000mhz ram if such a thing existed.

 

[PeterScott]

Here is an interesting interview that has answers about Raven Ridge. James Prior basically says that:
1) Raven Ridge is on 14nm+, not 14nm.
2) It has the same architectural improvements that Zen+ will have.
3) The difference is basically Zen+ being on 12nm.
https://youtu.be/7MPxN6MEgbk?t=308

1) Watched it. He never said that. Which is good.

Because there is no 14nm+. Basically GF renamed what was going to be called 14nm+ to 12nm.

 

[neblogai]

1) Watched it. He never said that. Which is good.

Because there is no 14nm+. Basically GF renamed what was going to be called 14nm+ to 12nm.

Not that I want to imply anything, or read much into it, but his sentence (@ 5:45) is:
" We knew we were going to target 14nm with the first Ryzen CPUs, and then we were going to target 14nm+ with Raven Ridge..".

 

[frozentundra123456]

A12-9800
8 CUs = 512 Processors (1,13GFLOPS at 1108MHz)
32 TMUs at 1108MHz = 35,45 GTexels/s
8 ROPs at 1108MHz = 8,86GPixels/s
up to DDR-4 2400 = 38,4GB/s Bandwidth

Ryzen 2200G (OC to 1550MHz + DDR-4 3600)
8 CUs = 512 Processors (1,58GFLOPS)
32 TMUs at 1550MHz = 49,6 GTexels/s
16 ROPs at 1550MHz = 24,8GPixels/s
DDR-4 3600 = 57,6GB/s Bandwidth

For comparison,

RX550 (1183MHz)
8 CUs = 512 Processors (1,21GFLOPS)
32 TMUs at 1183MHz = 37,86 GTexels/s
16 ROPs at 1183MHz = 18,93,8GPixels/s
GDDR-5 128Bit 7000MHz = 112GB/s Bandwidth

GT1030 (1468MHz boost)
384 Cores = (1,127GFLOPS)
24 TMUs at 1468MHz = 35,23 GTexels/s
16 ROPs at 1468MHz = 23,48,8GPixels/s
GDDR-5 64-Bit 6008MHz = 48,06GB/s Bandwidth

Ryzen 2200G at 1550MHz and DDR-3600 has higher specs than all the above, plus a new and better GPU architecture than A12-9800 and RX550. RX550 has almost double the Memory bandwidth vs the Ryzen 2200G but it doesnt use it because its severely under power for that huge bandwidth.
In paper Ryzen 2200G at 1550MHz and 3600 ram it should even be faster than GT1030.

Yea, with ram prices the way they are now, it makes a lot of sense to buy DDR4 3600 for a budget system.

 

[FIVR]

1) Watched it. He never said that. Which is good.

Because there is no 14nm+. Basically GF renamed what was going to be called 14nm+ to 12nm.

GF 12nm has different libraries and better scaling. It's not like intel who just renames the exact same process (14nm) for PR purposes.


If GF 12nm is just "14+" then there is no intel 14nm+ or ++. It doesn't even exist.

 

[PeterScott]

GF 12nm has different libraries and better scaling. It's not like intel who just renames the exact same process (14nm) for PR purposes.


If GF 12nm is just "14+" then there is no intel 14nm+ or ++. It doesn't even exist.

None of the process naming has much meaning anymore.

But there is no GF 14nm+ process. In fact or in name.

Whether it was renamed 12nm, or it was scrapped in favor of 12nm, GF 14nm+ does NOT exist.

The video is speaking of the plans they had for 14nm+, which is NOT the case anymore. If you look at the original roadmaps, they were targeting 14nm+ with later products.

But the roadmaps were changed to 12nm, when for whatever reason 14nm+ was replaced with "12nm".

Also note that the claims of density improvments is NOT vs GF 14nm, it is vs 16nm.

 

[FIVR]

None of the process naming has much meaning anymore.

But there is no GF 14nm+ process. In fact or in name.

Whether it was renamed 12nm, or it was scrapped in favor of 12nm, GF 14nm+ does NOT exist.

The video is speaking of the plans they had for 14nm+, which is NOT the case anymore. If you look at the original roadmaps, they were targeting 14nm+ with later products.

But the roadmaps were changed to 12nm, when for whatever reason 14nm+ was replaced with "12nm".

They are simply taking cues from intel. Intel likes to add a "+" to its process every time they get better yields or slightly improved power characteristics in their products (which every process benefits from over time). If their process improves over time, why not call it "14+" and reap the same benefits as intel?

Intel started this BS of "+" yet for some reason I don't see you calling them out.

 

[IRobot23]

Yea, with ram prices the way they are now, it makes a lot of sense to buy DDR4 3600 for a budget system.

Yes, 3600MHz is too expensive. 3466 or 3200MHz is good.

To be fair:
You cannot compare pentiun g4560(slow ram) + gt 1030 vs 2200G + fast ram (3200mhz).

i3 8100 vs 2200G is comparison by cpu power and gt 1030 vs 2200G with fast ram is gpu comparison.

So, maybe you guys should just realize what actually 2200G and 2400G offers for its money. The end.

When we will have benchmarks we will talk whats best etc.

 

[frozentundra123456]

Are you kidding me? Thats unplayable. I tested that with a 7400 and a 1030, it can manage 30-40 at medium on 720P and thats about it. Hardly something to call home about in a game that is 3 years old already.

A8-9600 with 8GB DDR4-2400 and a 1050 will beat the crap out of 2400G in games at the same price of a 2400G+2x4 3200, plus it has the full 8GB of ram instead of having to share with the IGP. If you dont want the A8-9600, you can have the 2200G+1050, a little more money but it still has more ram as well.

The 2400G may look well in a direct comparison with the 8400, but the fact is no one buys the 8400 to play with the igp -or- the GT1030. Not sure why everyone tries to force the 2200G and specially the 2400G intro something they are not, those new APU will shine on non-gaming Ryzen builds, the same ones that are right now pairing a R3 1200 and R5 1400 with a awful GT710/HD5450, and sometimes RX550/GT1030.

Expending that much money intro attemping to convert 2200G and 2400G intro IGP gaming builds makes not sence outside ITX and Brix-like.

Yea, I totally agree. Like I said before, an APU only makes sense for a budget gaming build for esports, older/less demanding games. The upgrading game is just a losing battle, no matter how hard the APU fans try to hype it. Better off to just save up a bit longer and get the performance you want at the beginning. After all, gaming is a hobby, not a need. Waiting a few month is not the end of the world. The only possible way to justify the upgrade path is with the 2200, due to the low initial price. But to upgrade the 2400 with a discrete card is just absurd. Better to buy a 1600 for only 30 dollars more and add a discrete card from the beginning.

I also seriously doubt the speculative claims about how well an APU will play W3. I have played it with a HD7770. I know it is old now, but faster than any pre Zen apu, and probably competiitve with Zen APUs. 1080p was totally unplayable. In fact, at 900p, low, it was a very borderline experience. Yes, I could have made a video showing I could get "30FPS", by picking the right area when nothing was going on, but there were a lot of times it dipped into the low/mid 20s. And that is ave FPS, I dont even know how low the minimums were.

 

[frozentundra123456]

Yes, 3600MHz is too expensive. 3466 or 3200MHz is good.

To be fair:
You cannot compare pentiun g4560(slow ram) + gt 1030 vs 2200G + fast ram (3200mhz).

i3 8100 vs 2200G is comparison by cpu power and gt 1030 vs 2200G with fast ram is gpu comparison.

So, maybe you guys should just realize what actually 2200G and 2400G offers for its money. The end.

When we will have benchmarks we will talk whats best etc.

I agree we have to wait for real benchmarks from independent testing sites. However, I totally disagree with you about the comparisons. An APU us a package, so total performance of the package (cpu plus gpu) is what matters. So the comparisons you cite certainly *are* valid.

 

[FIVR]

G4650 is a dead product in 2018. Intel might as well limit themselves to producing 8100 and above unless they want to pay board manufacturers to drop their prices down to below AM4 levels.


2 cores? Junk IGP? No upgrade path? Nobody is going to buy that stuff.

 

[PeterScott]

They are simply taking cues from intel. Intel likes to add a "+" to its process every time they get better yields or slightly improved power characteristics in their products (which every process benefits from over time). If their process improves over time, why not call it "14+" and reap the same benefits as intel?

Intel started this BS of "+" yet for some reason I don't see you calling them out.

IMO it's pretty reasonable to to mark a updated process with +, to signify changes but maintaining the same approx density.

It's a lot less reasonable to change the claimed size, when really it is just the same old process with similar tweaks. Like TSMC 12nm FFN that they use to build GV100, which appears to be essentially the same density as their 16nm process, and I strongly suspect GF 12nm, is really just similar tweaks to 14nm.