Athlon 200GE - the ultimate great place-holder CPU?

[VirtualLarry]

So the Vega 3 CU iGPU in the A200GE beats a HD 4850 (512MB, I assume) video card, in Metro Last Light? Wow, not that bad. Those cards were, back in the day, fairly powerful.

 

[Face2Face]

So the Vega 3 CU iGPU in the A200GE beats a HD 4850 (512MB, I assume) video card, in Metro Last Light? Wow, not that bad. Those cards were, back in the day, fairly powerful.

It's actually a 4850 1GB model. Note that frame times are a bit worse. I have a mixture of games, everything from FEAR and CRYSIS to GTA 5 and Fortnite. I'll post them here when my vid goes live. Shooting for this weekend...we'll see.

 

[SPBHM]

So the Vega 3 CU iGPU in the A200GE beats a HD 4850 (512MB, I assume) video card, in Metro Last Light? Wow, not that bad. Those cards were, back in the day, fairly powerful.

that's over 10 years ago... and it was already in "legacy support" when the game you mentioned was released...
kind of cool I guess since it's a 100W+ card, but...

 

[Face2Face]

that's over 10 years ago... and it was already in "legacy support" when the game you mentioned was released...
kind of cool I guess since it's a 100W+ card, but...

Correct. Driver support ceased months before the release of MLL and it's pretty much the ongoing theme. In older games where TersaScale 1.0 cards are still optimized they pull ahead, and the opposite for newer titles that have GCN optimizations.

 

[Shivansps]

Also, ran GFXBench Metal with my iPhone XR and GFXBench Vulkan with my 200GE w/VEGA3.

1080p Aztec Ruins (Normal Tier) Offscreen

200GE/Vega 3 w/Vulkan
View attachment 3088
iPhone XR A12 - Metal
View attachment 3089

AMD 200GE APU = 32 FPS
Apple A12 SoC = 68 FPS

Also ran Manhattan 3.0 Offscreen 1080p both using OpenGL

AMD 200GE APU = 42 FPS
Apple A12 SoC = 115 FPS

And this is why i think AMD went too far in decreasing GPU CUs from BR to RR on the low end, only the 9500E had 4CU, the rest, incluiding the 9500 had 6CU. The reason is very simple, this is already better than Intels HD630, why bother?

 

[DrMrLordX]

And this is why i think AMD went too far in decreasing GPU CUs from BR to RR on the low end, only the 9500E had 4CU, the rest, incluiding the 9500 had 6CU. The reason is very simple, this is already better than Intels HD630, why bother?

Even with 6CU, apple's A12 would still score higher.

 

[Abwx]

Even with 6CU, apple's A12 would still score higher.

In 3DMark - Ice Storm Unlimited Physics 1280x720 offscreen Apple A12 score 27400 while the 3 CUs of a mobile Ryzen 2200U score 36000-48000, but for some reasons 6CUs should be worse..??..

https://www.notebookcheck.net/Apple-A12-Bionic-SoC.331518.0.html

https://www.notebookcheck.net/AMD-Ryzen-3-2200U-SoC.277344.0.html

 

[DrMrLordX]

In 3DMark - Ice Storm Unlimited Physics 1280x720 offscreen Apple A12 score 27400 while the 3 CUs of a mobile Ryzen 2200U score 36000-48000, but for some reasons 6CUs should be worse..??..

https://www.notebookcheck.net/Apple-A12-Bionic-SoC.331518.0.html

https://www.notebookcheck.net/AMD-Ryzen-3-2200U-SoC.277344.0.html

You know what I'm citing. Don't be obtuse. It's in this thread, on this previous page:

https://forums.anandtech.com/thread...-great-place-holder-cpu.2553764/post-39730379

Funny how that flew under your radar, but my post didn't . . .

 

[Abwx]

You know what I'm citing. Don't be obtuse. It's in this thread, on this previous page:

https://forums.anandtech.com/thread...-great-place-holder-cpu.2553764/post-39730379

Funny how that flew under your radar, but my post didn't . . .

I saw this post, but if the A12 was that good then the results would be similar in 3D mark, probability is that Apple s GPU is a half baked one and does not support all functionalities that are supported by Vega, hence the mediocre result in 3D Mark and the inflated result in GFX Bench..

 

[DrMrLordX]

if the A12 was that good then the results would be similar in 3D mark,

Or maybe, just maybe, GFXBench is optimized well for Metal and 3DMark isn't?

edit: it may be more than just poor optimization for Metal from 3DMark. Consider this:

1800x @ 4.0 GHz, DDR4-3333, VegaFE @ 1585 MHz:
3DMark Ice Storm Unlimited 206724 (391489 graphics, 77955 Physics)
GFXBench 5.0 Vulkan Aztec Ruins 1080P Offline 30785.6 Frames (478.781 fps)

Improvement multiple moving from 3CU to my system:
3DMark: ~4.86x (based on average score of 2200u from notebookcheck)
GFXBench 5.0 Vulkan: ~14.93x (based on 200GE tested by @Face2Face)

If there is anything "wrong" with a 200GE running GFXBench, I would say it has nothing specifically to do with it being a Vega. There's something more going on here.

For reference, GFXBench 5.0 reports that a 1080Ti running Aztec Ruins 1080P Offline in Vulkan mode scores 40223.1 . There's no shortage of performance on NV hardware either.

 

[Shivansps]

Guys you only need to look at 2300U results to know how 6CUs performs. At any rate, the A12 is performing petty much like a Vega 11 or better, and that looks like too much to me.

 

[DrMrLordX]

Guys you only need to look at 2300U results to know how 6CUs performs. At any rate, the A12 is performing petty much like a Vega 11 or better, and that looks like too much to me.

I do not think 3DMark is optimized to run all that well on anything but NV or AMD hardware. That is its base. I'm not sure if people think all that much of GFXBench, but I would still not trust 3DMark numbers on most ARM platforms.

Also, comparing my system to the 2200u scores from notebookcheck clued me in that there might be something . . . funny about those Ice Storm Unlimited scores anyway. Not sure if they're useful. My VegaFE has 64 CUs versus the 3 of the 200GE/2200u, but I only saw less than 5x improvement in the overall score. Keep in mind that the CPU I used has 4 times as many cores at +33% clockspeed from the same uarch.

Then there's this score here:

https://hwbot.org/submission/3962259_havli_3dmark___ice_storm_unlimited_radeon_vega_3_81453_marks

200GE scoring twice what notebookcheck cites as being the average for the 2200u. Even if there is some overclocking (which I think there isn't; it's just got decent RAM, that's all), there's no way that system should score that high compared to mine. The physics score being 46036 compared to mine (~77000) is absurd. Unless it is single-threaded physics, which would be really funny.

 

[Face2Face]

Alright, finally finished up my video, but I'll post my results here. Any questions about my testing methodology, feel free to let me know. Also. memory was set to 2666Mhz.
Feel free to watch the video here if you want -

 

[Face2Face]

HD 2600 XT is 256MB, not 512MB

 

[DrMrLordX]

@Face2Face hats off to you for posting results here. Some people don't like sitting through videos. It makes sense that people want videos in this day and age, but still, being able to get down to the raw data is nice.

That being said, the Fortnite and GTAV results are pretty surprising. Also CS:GO. What was with the 1% and .1% minimums? That game tanked on the 200GE.

 

[Face2Face]

@Face2Face hats off to you for posting results here. Some people don't like sitting through videos. It makes sense that people want videos in this day and age, but still, being able to get down to the raw data is nice.

That being said, the Fortnite and GTAV results are pretty surprising. Also CS:GO. What was with the 1% and .1% minimums? That game tanked on the 200GE.

No problem at all. Sitting though a 15min video isn't for everyone - Anyway, I used the community made benchmark which uses unrealistic smoke in a couple of areas. This translates into unrealistic frame time data via 1% and 0.1% lows. The only reason I use it, is because it's consistent and AVG FPS is very much like in-game. I normally include FT graphs in my videos, but thought it would be too much for all of these cards. I'll attach one here, so you can see the FT data for yourself.

 

[DrMrLordX]

Ah, looks like all the cards get tripped up in those two same spots.

 

[Shivansps]

What happening there is the drivers are not optimised for new games or Windows 10, hell AMD dosent even provide ANY drivers for HD4xxx or less for Windows 10, and even the HD5xxx and HD6xxx are already on a 3 year old driver.

That said, the only thing VEGA 3 has in favor compared to a HD4850 is the pixel fill rate and more vram. But the results are petty much what i would expect.

 

[Insert_Nickname]

I'm somewhat surprised how well Crysis and Metro LL run actually. Both needed a beefy GPU back then to even run acceptable.

The fact we're at the point were you can run Crysis@30FPS on an entry-level IGP is impressive. Just shows what 10 years worth of improvements can do.

That said, the only thing VEGA 3 has in favor compared to a HD4850 is the pixel fill rate and more vram. But the results are petty much what i would expect.

Quite apart from being DX11/12, OpenGL 4.6 and Vulkan compliant with Shader Model 6. Besides having modern video acceleration and consuming a fraction of the power, of course.

Sorry but I couldn't resist...

All joking aside, take a look at the specs for the HD3000 generation. The Vega3 (192:12:4) core is actually a lot stronger then midrange chips (120:8:4 and 40:4:4) back then, it only looses hardware wise to the HD38xx series (320:16:16), and even then outperforms them.

For the HD4000 cards, the Vega3 is hanging in despite having less hardware (192:8:4 vs 800:40:16) just shows how much things have advanced in the last 10 years. A full Vega11 (704:44:16) is going to mop the floor with these older cards. Even the Vega8 (512:32:16) will likely outperform anything pre-HD5000.

 

[Shivansps]

I was talking about brute force to run those old games that shouldt be using these improvements. You can see there that DX9 games are the ones that perform best on these older cards, this probably means DX10 driver is not optimised. Im sure this will also happen on DX11 with HD5000.

 

[Face2Face]

What happening there is the drivers are not optimised for new games or Windows 10, hell AMD dosent even provide ANY drivers for HD4xxx or less for Windows 10, and even the HD5xxx and HD6xxx are already on a 3 year old driver.

That said, the only thing VEGA 3 has in favor compared to a HD4850 is the pixel fill rate and more vram. But the results are petty much what i would expect.

They do have a windows 10 for HD 2000-4000 TeraScale cards, but it cannot be downloaded from their site. You have to DL and install it using Windows update. I haven't dug into the details of it yet, but it's probably similar to Catalyst 13.1.

https://www.amd.com/en/support/kb/release-notes/rn-rad-win-legacy

 

[Face2Face]

Ah, looks like all the cards get tripped up in those two same spots.

Yeah. Now that I look at the graph, it may make sense to run the bench for 40ish seconds instead. Then it won't capture the odd smoke issues and produce more realistic frame times. They don't change much besides those two points.

 

[VirtualLarry]

I finally got my A200GE rig built last night! In an MSI B350 Bazooka micro-ATX mobo, with Corsair Vengeance LPX DDR4-2667 2x4GB of RAM, a Patriot Scorch 256GB M.2 PCI-E NVMe.

Had some initial difficultly, that could have been down to BIOS version, or could have been RAM seating issues. Anyways, I swapped in a Ryzen 3 1200, flash the BIOS to 1J0, and then put the APU back in, and it finally booted, after re-seating the RAM too.

It powered off suddenly after one of the first boots, which was a little disconcerting.

I tried overclocking to 4.00Ghz off the bat, with "Auto" voltage, boy, it didn't like that.

So, I tried 3.80Ghz, it seemed OK, but it froze during Win10 installation.

After that, I boosted the voltage manually to 1.40V vcore (fixed), and that seemed to allow it to install, although, it "felt slower" after boosting the voltage.

Still running the stock heatsink. When I finally got into Window 10, all installed, and installed CPU-Z and HWMonitor, it was running at 3.80Ghz, and around 59C, under stress-test from CPU-Z.

So, I tried boosting the clock to 3.90Ghz, still seems to run OK, and a short CPU-Z stress-test while browsing in Firefox, went OK. (That can often cause crashes, if unstable.)

One thing that has me puzzled. Windows 10 didn't auto-install video drivers. So I manually went to AMD.com to get them. They recommended 19.1.1 for the A200GE APU.

I installed those, and upon reboot, the screen kept flashing to black, for like 2 minutes. Wasn't sure what was up. Now, when I boot or switch resolutions, it shows static on and off, like 3-5 times, then it's OK.

Not sure, will try replacing HDMI cable soon. It does it even when clocked down to 3.20Ghz.

Posting from that system.

 

[cbn]

Virtual Larry,

How is the browsing and 4K60 Youtube?

(Also how much RAM does it use for 4K60 Youtube?)

 

[VirtualLarry]

HardwareUnboxed, 4K60 video, Radeon VII unboxing/teardown. Clicked on that video from YT search page, clicked full-screen, boom, 4K60 right away, no "zooming up", like used to happen. "Stats for Nerds" confirmed, 4K viewport, 4K content, running 60FPS, VP9. Beautiful!

Browsing performance is pretty good too.

Installing Win10 did seem slightly... slower, than some of my other rigs, for some reason, but that may have had something to do with my having to force power-off during the first installation attempt, and then deleting and re-creating the partitions, maybe it was TRIMming in the background or something.

Either way, it's working great now. I should probably do some A/B testing with the OC speeds, to see if I notice much of a difference. It did seem like my Ryzen 3 2200G APU rig was slightly snappier, overall, but this one is decent.

Edit: Note that I set the "Content Process Limit" in Firefox to "2", because I only have two real cores on this APU (active, at least). I could possibly set that to "4" to test out too, I suppose.