[Hardware Unboxed] Ryzen 2400G/2200G Vega Graphics Performance

[Tapoer]

According to tomshardware when at gaming the iGPU of 2200 and 2400 are always at their max clocks, they can't use more power.

 

[Muhammed]

When the differences are so small when using the highest settings, it means both of them provide an identical end-user experience. The GT 1030 typically winning by <10% doesn't matter in the end at this performance level(30fps)

Testing at highest settings is only viable for med tier GPUs and above. The lowest end of GPUs are not tested at high settings, and are not used at high settings at all by the consumer. Medium is more than enough. In which case the 1030 is 30% faster.

 

[Glo.]

I don't think the power limits are quite working if it's only running at 45/50w in gaming when is rated at 65w and well capable of drawing more? A bit odd.

65w == 1030 + the t rated i7.


Sent from my iPhone using Tapatalk

That T rated CPU will use even around 80W of power, so it is not 35W TDP CPU.

https://www.computerbase.de/2017-01/intel-core-i7-7700t-test-kaby-lake/4/

None of 35W TDP CPUs will be truly 35W if the clocks will be in range of 65W chips. i7 8700T is supposed to have HT and 2.4 GHz base clock with boost clock of 4.0 GHz, similarly to i5 8400. Expect it to consume the same amount of power under load as 8400 does.

End of off-topic.

 

[Qwertilot]

They balance it over time by dropping clocks. Of course if the zen cores in these things can support the - entirely unstressfull - gaming loads on a tiny power draw then so can everything Intel make

I remain a bit perplexed why they just don't let the iGpu dynamically over clock to use up the whole 65w power budget.

Isn't that a big theoretical plus of an apu?


Sent from my iPhone using Tapatalk

 

[frozentundra123456]

Clocking the cpu low then is sense. Wonder why the iGpu can't take more power though? Maybe the bandwidth throttles it enough that there's no point.

The perf/watt is definitely reasonable, just not a major win either


Sent from my iPhone using Tapatalk

Yea, performance/watt is supposed to be one of the big pluses for an APU. Certainly wasnt for intel, and it seems like RR is not clearly better than a discrete solution either. The real test will be in mobile. On the desktop, within this relatively low power envelope, it is not a big deal.

 

[BSim500]

Intel + NV can’t match the APUs combination of:

• CPU performance
• GPU performance
• Price
• Power usage.

^ The first three are true. Power consumption isn't quite as clear cut and a lot of people trying to do APU vs low-end dGPU power comparisons don't seem to be grasping the nuances of how modern dGPU's work (which tech benchmarks don't show). Eg, I have an i5-7500 + GTX 1050Ti in my HTPC (fed with a Corsair SF450), and I'm getting around 2.2-2.5x the fps at same whole system 27w idle / 98w max gaming load wattage as multiple reviews show the 2400G typically draws:-
https://www.purepc.pl/procesory/test_amd_ryzen_3_2200g_najtansze_apu_z_ukladem_radeon_vega?page=0,17
http://www.guru3d.com/articles_pages/amd_ryzen_5_2400g_review,6.html
https://www.computerbase.de/2018-02/ryzen-3-2200g-5-2400g-test-amd-raven-ridge/3/
https://pclab.pl/art76962-16.html

If I set MSI Afterburner's power limiter to 70%, it falls to 75w (whole system max load) yet is still pulling double the fps at 75w vs 2400G at 90-100w. Likewise, if a game maxed out an APU and got 50fps @ 98w, on a GTX 1050Ti it's likely to get 100-120fps at same 95-100w. However, if you enable VSync and cap both at 60fps, the former's power draw remains the same (because it's under 60fps) whilst the 1050Ti's power plummets by 30-40w due to the GPU equivalent of "cool & quiet" kicking in, ie, the dGPU will down-clock from 1900MHz @ 1.05v to 1025MHz @ 0.625v and 1050Ti gaming temps for esports / older games + VSync regularly end up an amusing 23-35c @ 60fps load even with those tiny single-fan "star" shaped heatsinks at 30% (900rpm) fan speed. Same with GTX 1060's - that 120w (GPU itself) is at full maxed out load. For "esports" style games capped at 60Hz, it's still barely pulling 30-40w.

These Ryzen APU's are decent chips, highly competitive and unmatched for dGPU-less performance. The 2400G is certainly more power efficient vs itself / R5 1500X + an RX550 but the Vega architecture is quite a bit less power efficient vs Pascal in terms of perf-per-watt, even in APU form. The main APU benefit is pricing and form factor (2400G = fastest chip without dGPU (ie, can fit in slim dGPU less ITX cases like the Inwin Chopin) / 2200G = fastest combo for $99). But if you stuck either one in an ITX case with a Pico-PSU, you'd absolutely need exactly the same 150-180w Pico-PSU + brick for the 2200G/2400G as you would a i3-8100 + 1050Ti (allowing load +30-50% overhead for more stable power delivery during transient spikes, etc).

I hope this explains things better as real-life is nowhere near as simple as "65w CPU + 70w GPU = 135w total so that must pull twice the power of 65w APU + no dGPU".

Edit: I literally just measured this with a Kill-A-Watt, but for real life example figures of above effect, with VSync on and a solid 1080p/Ultra/60fps, I'm pulling (at the wall, whole system draw) = 76w for Skyrim (inc 2K texture mods), 61w for Bioshock Infinite, 55w for The Witness, 43w for Portal, 34w for Torchlight, etc. Peak power with GPU at 100% in heavier modern AAA games is 98w. Ironically, removing the 1050Ti and running on the Intel HD630 doesn't lower power draw much as Intel's iGPU's are not only +6x slower they're also almost 5x less efficient, ie, what loads 25-50% usage @ 0.625v on the 1050Ti's shaders will load up 100% usage at 1.0v on the HD630 iGPU shaders.

 

[Muhammed]

Some more 1030 vs Vega 11 disparities:

https://www.sweclockers.com/test/25...-2200g-raven-ridge-med-vega-grafik/18#content

 

[tamz_msc]

^If it's about making it a GPU-only comparison, why not test with the same CPU?

 

[frozentundra123456]

That T rated CPU will use even around 80W of power, so it is not 35W TDP CPU.

https://www.computerbase.de/2017-01/intel-core-i7-7700t-test-kaby-lake/4/

None of 35W TDP CPUs will be truly 35W if the clocks will be in range of 65W chips. i7 8700T is supposed to have HT and 2.4 GHz base clock with boost clock of 4.0 GHz, similarly to i5 8400. Expect it to consume the same amount of power under load as 8400 does.

End of off-topic.

Those numbers look very strange. If "neutral" means idle, as I assume it does, then they are saying a 35 watt cpu draws 33 watts at idle? That makes no sense.

 

[Hitman928]

Those numbers look very strange. If "neutral" means idle, as I assume it does, then they are saying a 35 watt cpu draws 33 watts at idle? That makes no sense.

Pretty sure that's system power.

 

[SirDinadan]

Why this wasn't covered in the first review video? Or are those messy frame times are pass on APUs and not on dGPUs?
God job playing competitively in BF1 and CoD with those wild frame times. You will have a hard time aiming consistently with a rifle in mid to long-range engagements.
The spikes in DiRT Rally are really bad - when you are going for it and push down the throttle those hitches will throw you off.
GTA V is a mess - I reported the same issue to AMD (stutter, inconsistent performance, LoD popping) a year ago. Yeah, drivers are fine now ...

 

[Face2Face]

Why this wasn't covered in the first review video? Or are those messy frame times are pass on APUs and not on dGPUs?
God job playing competitively in BF1 and CoD with those wild frame times. You will have a hard time aiming consistently with a rifle in mid to long-range engagements.
The spikes in DiRT Rally are really bad - when you are going for it and push down the throttle those hitches will throw you off.
GTA V is a mess - I reported the same issue to AMD (stutter, inconsistent performance, LoD popping) a year ago. Yeah, drivers are fine now ...

I'd be interested to see what his (ms) cap is set to in Afterburner. If it's left at 100ms, then those frame times are indeed terrible. Those full stack lines are very noticeable stutter (game pausing) I don't tend to notice large stutters until I see 30ms-40ms of more, but it really depends on your FPS as well. I hope to get a 2200G to do some of my own testing to see how well it stacks up to the games against some previous generation of cards.

 

[SirDinadan]

Those full stack lines are very noticeable stutter

Based on my DiRT 4 test the spikes are ~25-30 ms large and are very noticeable as the gameplay is super smooth otherwise. The sudden jump from 15 ms to 30 is very distracting - immediately felt the difference when switched from Vega 8 to the dGPUs.

 

[TheELF]

GTA V is a mess - I reported the same issue to AMD (stutter, inconsistent performance, LoD popping) a year ago. Yeah, drivers are fine now ...

It has nothing to do with drivers,it's just a crappy port with stutter and non syncing build in,it's easily fixable though.

 

[Face2Face]

Based on my DiRT 4 test the spikes are ~25-30 ms large and are very noticeable as the gameplay is super smooth otherwise. The sudden jump from 15 ms to 30 is very distracting - immediately felt the difference when switched from Vega 8 to the dGPUs.

Definitely at 60 FPS. At around 30 FPS, it's not too bad IMO (25-30 ms spikes), but still noticeable none the less. Cool Website and Channel. I'll get subbed

It has nothing to do with drivers,it's just a crappy port with stutter and non syncing build in,it's easily fixable though.

Not the same cause of stutter though. It's obvious in the video he's completely CPU limited (Celeron)

 

[SirDinadan]

Nice trick, although this is somewhat disappointing. Ryzen 3 with its 4 cores can't handle the game? And yeah, what a crappy port from RockStar.

 

[TheELF]

Not the same cause of stutter though. It's obvious in the video he's completely CPU limited (Celeron)

It's exactly the same,look at the hardware unboxed video again, the 2200g is at non stop 100% usage as well(in the parts he get's the heavy stutters) .

 

[ZGR]

My i5 4690k was at 100% CPU usage in GTA V. That game needs more than 4 threads out of the box. Once you add mods like 50x traffic and peds, an 8 core is the minimum for 60fps.