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

[AtenRa]

Isn't that for 2700U with 10NCUs? 2400G will have all.

Shader engine is not only CU/SP.
Example : VEGA64 has 4 Shader engines same goes for RX 48/47/58/57, FuRy, R9 290 series.

Good comparison is rx 550 vs rx 560, both have 2SE, but RX 560 has double amount of CUs

Ahh you mean the Graphics Pipelines, most probably only one with 11 CUs.

VEGA 64 diagram.

 

[french toast]

Why is there a limitation with the graphics pipelines? Does that extend to rops?

 

[LTC8K6]

IRobot23 and LTC8K are the best.

Catching up on this thread and saw Abott & Costello reincarnated with posts 904<>918. Thanks for the laughs. A good stress release.

You got two tens for a five?

 

[ZGR]

This APU is faster than my i7-5775C at half the price. That is such amazing value!

This APU is perfect for older games, even for 4k. I hate the esports marketing. Competitive esports thrives on 144hz+ and that usually means a GTX 1060/RX 480 or faster.

This APU is solid for 60hz, but that doesn't mean it is only for older pvp games. I can see this APU thriving on single player games.

My old i7 is a 4c/8t and it still has plenty of life left even for an Rx 64 or GTX 1080 ti as well as next gen. This APU seems like such a no brainer in this current climate of expensive dGPU's.

 

[Shivansps]

So 4000 for 2200G and 5000 for 2400G in 3dmark 11 P.

Even the slower RX550 3dmark 11 P result has a higher score https://www.3dmark.com/3dm11/12616711
GT1030 is around 4600 and a GTX1050 2GB scores around 9000.

I hope thats a driver issue, even the 2200G should be scoring around 5000.

3dmark is not really bandwidth limiting but there is a small impact when comparing the same GPU with DDR3 and GDDR5.
DDR3 R7 250 https://www.3dmark.com/3dm11/9842569
2400

GDDR5 R7 250 https://www.3dmark.com/3dm11/8813442
2700

 

[USER8000]

3DMark11 scores with a Ryzen 5 1400 and a GT1030:

https://www.3dmark.com/search#/?mode=advanced&url=/proxycon/ajax/search2/cpugpu/3dm11/P/2230/1138/500000?minScore=0&cpuName=AMD Ryzen 5 1400&gpuCount=0&gpuName=NVIDIA GeForce GT 1030

The 2400G scores for CPU physics and graphics scores look similar to an overclocked Ryzen 5 1400 and a GT1030.

We will need to wait for proper reviews to see if games hold up in the same way.

 

[Shivansps]

Its petty much in line with AMD slide that 2400G = GT1030 at least in 3dmark. No way it will hold out that way in real games, it never happened before, but since i cant find any reason of why the score looks bad compared to a RX550 (2200G should be already scoring about the same and its more than 1K of points below) it may be a driver issue, or a TDP issue, but that would not explain why the 2400G scores better.

 

[IRobot23]

Ahh you mean the Graphics Pipelines

Well...

Looks like VEGA has 16pixels output/clock per shader engine. Just like Tonga, Tahiti, Fiji and Hawaii...

 

[neblogai]

The 2400G scores for CPU physics and graphics scores look similar to an overclocked Ryzen 5 1400 and a GT1030.

We will need to wait for proper reviews to see if games hold up in the same way.

Separate scores better for the 2400G, but 'combined score' much higher for 1400+GT1030 than 2400G. Again- indicating power limit, probably caused by weak Wraith Stealth cooler on a 2400G.

 

[alexruiz]

I was going to quote several posts, but I will just ask this:

- Are those APU scores 3dmark11, or firestrike?
The scores seem low for 3dmark 11. Here s a run of one of my old personal PCs running an A10-780k Kaveri:
https://www.3dmark.com/3dm11/12483614

If they were firestrike, they would be phenomenal

Edit:
Has to be indeed 3dmark11. Firestrike scores by AMD in some of the first posts. Duh!

 

[alexruiz]

On the topic of 3DMark and how it is affected by RAM speed.

1) RAM speed really affects 3DMark on APUs.

The same A10-7850k system of which I posted the 3DMark11 score would net only ~1800 Pscore with single channel DDR3-1600. Graphics score would drop by almost half. Need to find that result...


2) RAM speed doesn't affect as much if discrete GPU.

RAM speed still affects 3DMark runs, but nowhere as much as it does for APUs. I have 4 more results, 3dmark11 and firestrike of a Ryzen 5 1300X, GTX 1050 Ti system that I built for a customer. One set of runs is DDR4-3200 16-18-18-35 dual channel 2x4GB, the other set of runs is DDR4-2400 14-16-16-31 single channel 1x8GB. The scores are higher on the dual channel system, but not that much.

DDR4-3200 dual channel results:
https://www.3dmark.com/3dm11/12621879
https://www.3dmark.com/fs/14776506

DDR4-2400 single channel results:
https://www.3dmark.com/3dm11/12622776
https://www.3dmark.com/fs/14772205

I guess we will just have to wait and see...

 

[Shivansps]

On the topic of 3DMark and how it is affected by RAM speed.

1) RAM speed really affects 3DMark on APUs.

The same A10-7850k system of which I posted the 3DMark11 score would net only ~1800 Pscore with single channel DDR3-1600. Graphics score would drop by almost half. Need to find that result...


2) RAM speed doesn't affect as much if discrete GPU.

RAM speed still affects 3DMark runs, but nowhere as much as it does for APUs. I have 4 more results, 3dmark11 and firestrike of a Ryzen 5 1300X, GTX 1050 Ti system that I built for a customer. One set of runs is DDR4-3200 16-18-18-35 dual channel 2x4GB, the other set of runs is DDR4-2400 14-16-16-31 single channel 1x8GB. The scores are higher on the dual channel system, but not that much.

DDR4-3200 dual channel results:
https://www.3dmark.com/3dm11/12621879
https://www.3dmark.com/fs/14776506

DDR4-2400 single channel results:
https://www.3dmark.com/3dm11/12622776
https://www.3dmark.com/fs/14772205

I guess we will just have to wait and see...

here you go:
A10-7850 with dual channel: https://www.3dmark.com/3dm11/9117205
A10-7850 single channel: https://www.3dmark.com/3dm11/8523238

As expected, graphics alone do go down a bit(around 25%), but combined score gets hit like a ton of bricks, probably because there is no enoght bandwidth to run both things at the same time.

Seen this, then is possible for the 2200G to be so sevely bottlenecked in bandwidth that it performs a full 1000 points in graphics below the RX550. Then forget about OC it, the APU performance will be determined by bandwidth, bandwidth and bandwidth.

 

[AtenRa]

We dont know what RAM was used in the above results, could be 2667MHz only.

 

[alexruiz]

3DMark11 scores with a Ryzen 5 1400 and a GT1030:

https://www.3dmark.com/search#/?mode=advanced&url=/proxycon/ajax/search2/cpugpu/3dm11/P/2230/1138/500000?minScore=0&cpuName=AMD Ryzen 5 1400&gpuCount=0&gpuName=NVIDIA GeForce GT 1030

The 2400G scores for CPU physics and graphics scores look similar to an overclocked Ryzen 5 1400 and a GT1030.

We will need to wait for proper reviews to see if games hold up in the same way.

I don't know if more info such as motherboard used, cooling solution and specially, RAM configuration was disclosed.
What RAM speed should be assumed? DDR4-2400 dual channel?

 

[piesquared]

This APU is faster than my i7-5775C at half the price. That is such amazing value!

This APU is perfect for older games, even for 4k. I hate the esports marketing. Competitive esports thrives on 144hz+ and that usually means a GTX 1060/RX 480 or faster.

This APU is solid for 60hz, but that doesn't mean it is only for older pvp games. I can see this APU thriving on single player games.

My old i7 is a 4c/8t and it still has plenty of life left even for an Rx 64 or GTX 1080 ti as well as next gen. This APU seems like such a no brainer in this current climate of expensive dGPU's.

The Ryzen G series is impressive. They now have relative compute performance to competing chips on the market, with far and away better graphics, similar to low end discrete graphics boards. Precision Boost 2.0 and some of the other SenseMI tech built in make this chip quite attractive even for a general purpose computer with bias towards gaming.

 

[neblogai]

..
Seen this, then is possible for the 2200G to be so severely bottlenecked in bandwidth that it performs a full 1000 points in graphics below the RX550. Then forget about OC it, the APU performance will be determined by bandwidth, bandwidth and bandwidth.

Here is another, older 2400G score- that confused me a lot:
https://www.3dmark.com/compare/3dm11/12257178/3dm11/12554398/3dm11/12593862#
Similar graphics score, lower physics score- but much higher combined score.

 

[piesquared]

Results are starting trickle out. Add another 30% from overclocking and this thing could approach 1050/ RX 560 territory! Beast.

http://moepc.net/?post=4249

Ryzen Mobile @ 15W is amazing also.

 

[IRobot23]

IF this is true, than looks like great deal. Yet games can show different picture.

 

[Shivansps]

Results are starting trickle out. Add another 30% from overclocking and this thing could approach 1050/ RX 560 territory! Beast.

Sure, lets all play 3dmark. With those numbers, it will have a hard time trying to match a GT1030 in every game and we are talking about the expensive 2400G, i sure hope so slower ram was used or there is some wierd driver issue.

 

[i-know-not]

An RX 550 has the same mem bandwidth and ROPs as an RX 560, and the 3DMark 11P score divided by FLOPs is 15-20% higher than that of the 560. Being overprovisioned with bandwidth and ROPs, the RX 550 is very much an outlier for GCN GPUs, and thus a naive comparison of any other GCN GPU to an RX 550 will give inaccurate conclusions.

 

[alexruiz]

Results are starting trickle out. Add another 30% from overclocking and this thing could approach 1050/ RX 560 territory! Beast.

http://moepc.net/?post=4249

Ryzen Mobile @ 15W is amazing also.

Ok, he is using dual channel DDR4-2933.
Sticks with Samsung b-die chips are cl14, SK-Hynix chips are cl16. Even assuming samsumg b-die RAM for best case scenario, it is still very promising.

Edit: It it truly supports DDR4-3200 at full speed, that will be great.
Edit 2: His bristol ridge number seems low. I posted the score of one of my personal A10-7850k Kaveri at ~2700. Bristol ridge should not be slower.

 

[neblogai]

Ok, he is using dual channel DDR4-2933.
Sticks with Samsung b-die chips are cl14, SK-Hynix chips are cl16. Even assuming samsumg b-die RAM for best case scenario, it is still very promising.

Edit: It it truly supports DDR4-3200 at full speed, that will be great.
Edit 2: His bristol ridge number seems low. I posted the score of one of my personal A10-7850k Kaveri at ~2700. Bristol ridge should not be slower.

Where did you get the info regarding it using 2933 memory? AFAIK- this is just aggregated info from anonymous 3DMark scores Tum_Apisak collected.

 

[USER8000]

Ok, he is using dual channel DDR4-2933.
Sticks with Samsung b-die chips are cl14, SK-Hynix chips are cl16. Even assuming samsumg b-die RAM for best case scenario, it is still very promising.

Edit: It it truly supports DDR4-3200 at full speed, that will be great.
Edit 2: His bristol ridge number seems low. I posted the score of one of my personal A10-7850k Kaveri at ~2700. Bristol ridge should not be slower.

2933MHZ is the maximum officially rated memory speed.

 

[raghu78]

Rx Vega M GL perf on par with GTX 1050 Ti. Not bad for a 65w CPU+GPU package considering GTX 1050 Ti is around same TDP for just GPU. 2400G perf is quite good for an APU. But 2200G is looking to be the killer budget APU of 2018. Unbeatable value for $99.

 

[alexruiz]

Where did you get the info regarding it using 2933 memory? AFAIK- this is just aggregated info from anonymous 3DMark scores Tum_Apisak collected.

In test setup he had DDR4-2933 for Raven Ridge APUs.