AMD Ryzen 7 5700G APU Pictured and Tested

[Uskompuf]

We have received various leaks and benchmarks for AMD's upcoming Ryzen 5000G processors, these were all from engineering samples but we now have our first look at the retail 5700G. The AMD Ryzen 7 5700G features the model number 100-000000263 attributed to earlier rumors and has been tested in CPU-Z scoring 631 points in single-threaded performance along with 6782 points in multi-threaded, and in Cinebench R20 it scored 6040 points. The integrated Vega graphics lack any official drivers but GPU-Z reports a Vega 8 processor with 12 Streaming Multiprocessors and a base clock of 2 GHz. AMD is yet to officially announce any Ryzen 5000G processors so it is unclear how far away their launch is and whether or not they will be made available to the DIY market.



View at TechPowerUp Main Site

 

[thevoiceofreason]

So... still 3400G-like performance in games? DDR5 can't come soon enough.

 

[Sihastru]

They blocked off the serial number, but they left the QR code intact. There's a smart representative of the human race right there.

 

[Lionheart]

Vega 8 graphics again AMD!!! Dafuq!

 

[Melvis]

Vega 8 graphics again AMD!!! Dafu.

I know right.....its a bit of a shame, it would of been nice to see these with RDN1 at least then Id be all over getting one of these but sadly its still better to get a 3400G and call it a day for playing games if your running a tiny ITX build. But if I can see a good FPS gain id be interested but it doesnt seem to be enough considering the price you gotta pay for these over a 3400G

 

[thevoiceofreason]

A beefier GPU wouldn't help with that kind of memory bandwidth anyway. It's already starved with Vega. Maybe if they added on-die cache like with RX6000 series...

 

[Space Lynx]

my dad has a A320 mobo and I believe the first gen APU of ryzen... hoping the BIOS gets an update for this, will be a nice upgrade for him and prob last him 10+ years honestly as he only does web browsing. well really I don't need to upgrade at all for him, but eh. I'd like to if the MSRP is still good at around $99-$139 range.

 

[tabascosauz]

So... still 3400G-like performance in games? DDR5 can't come soon enough.

Do you actually own the current APUs, OC and game on them to back up your statements about how they're supposedly bandwidth-starved? If you actually had hands-on experience with Renoir you'd know it is already a bandwidth god in terms of read, write and copy; Cezanne's Zen 3 improvements may very well take its already decent DRAM latency into the 40ns range.

DDR5 won't help jack shit if AMD continues to saddle these otherwise excellent APUs with Vega 8 (even lower clocked Vega 7 for the 6-core, even worse). If you know what you're doing, you're already going to be running 4000-4400 B-die on these APUs, but it won't actually tangibly help your performance beyond about 3800CL16. Maybe if you only run JEDEC 2133. You can take the memory higher all you want, you can try to OC the iGPU further all you want, Vega 8 isn't going to go anywhere meaningful.

It's like calling for GDDR6X on a GT 1030. Granted, the DDR4 version was über bad, but it literally ran at half the bus width of a normal DDR4-2133 system.

The POS Vega iGPU is too small and too old to ever be of any use. AMD can go on and on about how Vega 8 makes up the difference in clocks, but the performance shows. I can understand the need to keep die size small to mitigate the consequences of their unfortunate decision to put all their products in the same TSMC N7 basket, but still being stuck on Vega in 2021 is just lmao. Or alright, Vega with a die shrink and some new encoding features.

Like, either give us back 7nm Vega 11, or move to RDNA. I don't see why the latter is so hard to do.

 

[Caring1]

Cinebench shows it running @3.8GHz for a score of 5,500 also so I'm going to assume that's a single core overclock to reach the higher score.

 

[ratirt]

Vega 8 graphics again AMD!!! Dafuq!

3400G has Vega 11(11EUs 704 Shaders) graphics and 5700G has Vega 8 (8EUs 512 Shaders) graphics but clocked way higher. It's 2100Mhz for the 5700G vs 1400Mhz for the 3400G.
I'd wait for the actual benchmarks to say it is exactly the same. There will be a boost in the performance but it is still Vega graphics anyway.

 

[thevoiceofreason]

Do you actually own the current APUs, OC and game on them to back up your statements about how they're supposedly bandwidth-starved? If you actually had hands-on experience with Renoir you'd know it is already a bandwidth god in terms of read, write and copy; Cezanne's Zen 3 improvements may very well take its already decent DRAM latency into the 40ns range.

Yes I do, and its graphics performance clearly scales with memory bandwidth (which is a function of frequency and memory bus width) and not latency.

 

[TumbleGeorge]

know right.....its a bit lot of a shame

I just fixed it's for you!

 

[tabascosauz]

Yes I do, and its graphics performance clearly scales with memory bandwidth (which is a function of frequency and memory bus width) and not latency.

I didn't say its performance isn't reliant on memory bandwidth. You said they're bandwidth-starved. As we're not talking about mobile Cezanne here, these desktop APUs are only bandwidth "starved" if you're below 1900MHz MCLK and FCLK, in which case you're already doing something wrong considering these chips thrive in the 2000-2300MHz IF range, and 2x8GB SR kits that can hit those speeds aren't prohibitively expensive either. Neither did I say that latency was an issue for the APUs.

No one's asking for a 5700XT here. All we want is a badly needed newer arch on the same number of CUs, or a larger core to get back to where we were + some higher 7nm clocks. If we weren't in the middle of a (partially self-inflicted) N7 supply shortage, there'd be no excuse afforded to AMD for this shit, "hey we took away some of your CUs which we justified by giving you more cores, but have some more clockspeed to level the playing field......sorta...kinda...not really......and now we're going for round 2!"

 

[BSim500]

Are these going to actually be sold to consumers, or the same vapourware (outside OEM channels) that the 4000 Renoir's were?

 

[Caring1]

Are these going to actually be sold to consumers, or the same vapourware (outside OEM channels) that the 4000 Renoir's were?

I'm crossing my fingers they go retail as I'm waiting for one for an iTX build.

 

[TumbleGeorge]

I'm crossing my fingers they go retail as I'm waiting for one for an iTX build.

#metoo

 

[1d10t]

Vega again, oh come on AMD he's already moved to Intel, why you still keeping his legacy? Put your RDNA in there already

On serious side, I don't expect any major architectural changes to the APU. Intel is prime example, they have a very good performance XE inside their Tiger Lake, and yet we only got ancient UHD on the desktop CPU.

Are these going to actually be sold to consumers, or the same vapourware (outside OEM channels) that the 4000 Renoir's were?

20 bucks they will come in bundle.

 

[TumbleGeorge]

Vega again, oh come on AMD he's already moved to Intel, why you still keeping his legacy? Put your RDNA in there already

But this maybe is new Vega. There maybe are 2-3-4-5 different Vega generations?

 

[1d10t]

But this maybe is new Vega. There maybe are 2-3-4-5 different Vega generations?

Yeah, you may call it "rebranded" .Vega actually still GCN, you can tinker with its TMU, ROPs , clock and many encoder but in my opinion RDNA at lower clock would provide much better effiency at least, and performance at best.

 

[watzupken]

Do you actually own the current APUs, OC and game on them to back up your statements about how they're supposedly bandwidth-starved? If you actually had hands-on experience with Renoir you'd know it is already a bandwidth god in terms of read, write and copy; Cezanne's Zen 3 improvements may very well take its already decent DRAM latency into the 40ns range.

DDR5 won't help jack shit if AMD continues to saddle these otherwise excellent APUs with Vega 8 (even lower clocked Vega 7 for the 6-core, even worse). If you know what you're doing, you're already going to be running 4000-4400 B-die on these APUs, but it won't actually tangibly help your performance beyond about 3800CL16. Maybe if you only run JEDEC 2133. You can take the memory higher all you want, you can try to OC the iGPU further all you want, Vega 8 isn't going to go anywhere meaningful.

It's like calling for GDDR6X on a GT 1030. Granted, the DDR4 version was über bad, but it literally ran at half the bus width of a normal DDR4-2133 system.

The POS Vega iGPU is too small and too old to ever be of any use. AMD can go on and on about how Vega 8 makes up the difference in clocks, but the performance shows. I can understand the need to keep die size small to mitigate the consequences of their unfortunate decision to put all their products in the same TSMC N7 basket, but still being stuck on Vega in 2021 is just lmao. Or alright, Vega with a die shrink and some new encoding features.

Like, either give us back 7nm Vega 11, or move to RDNA. I don't see why the latter is so hard to do.

The problem is this, at lower resolution, most graphic card don't run into memory bandwidth limitation that quickly. So when you try to bench a game at 720p, the higher CU count helps. As you start pushing 1080p even at low settings, the memory bandwidth starts becoming a bottleneck. So if you are thinking if they come up with a Vega 11 or 20 but stuck with DDR4 3200 as base, and its going to make games playable at 1080p at medium quality settings or higher, then you may be disappointed. Instead you may see that the increase in CU will suffer significant diminishing returns. While I think 8 CUs is too little as well and AMD should have introduced Vega 11 again with the higher clock speed, it is what it is. In my own experience, the Vega 11 on my Ryzen 5 3400G actually benefited more from me bumping the memory frequency to 3600, as oppose to leaving it at the official memory frequency that Zen+ supports with a 250 Mhz overclock on the GPU itself. If you combine both, the Vega 11 is actually faster in 3D Mark benchmarks than a Nvidia MX350.

 

[Julhes]

 

[watzupken]

But this maybe is new Vega. There maybe are 2-3-4-5 different Vega generations?

Unfortunately, Vega is architecturally the same regardless of which APU generation it is bundled with. Only differences is that its produced on a more refined 7nm, and also an update in video decode capabilities if I am not mistaken. Mostly minor changes, but no changes to the core architecture.

 

[tabascosauz]

But this maybe is new Vega. There maybe are 2-3-4-5 different Vega generations?

Doesn't change that it's Vega. In Renoir, they die shrunk to 7nm, added some newer VP9 and HEVC capabilities, and clocked higher. In Cezanne, they're only improving volt/freq scaling, reducing power draw and adding something that's basically the opposite of Nvidia Reflex Boost (nvidia keeps clocks high when not needed for input lag, amd reduces "unnecessary" gpu clocks for power). So still Vega at heart.

The problem is this, at lower resolution, most graphic card don't run into memory bandwidth limitation that quickly. So when you try to bench a game at 720p, the higher CU count helps. As you start pushing 1080p even at low settings, the memory bandwidth starts becoming a bottleneck. So if you are thinking if they come up with a Vega 11 or 20 but stuck with DDR4 3200 as base, and its going to make games playable at 1080p at medium quality settings or higher, then you may be disappointed. Instead you may see that the increase in CU will suffer significant diminishing returns. While I think 8 CUs is too little as well and AMD should have introduced Vega 11 again with the higher clock speed, it is what it is. In my own experience, the Vega 11 on my Ryzen 5 3400G actually benefited more from me bumping the memory frequency to 3600, as oppose to leaving it at the official memory frequency that Zen+ supports with a 250 Mhz overclock on the GPU itself. If you combine both, the Vega 11 is actually faster in 3D Mark benchmarks than a Nvidia MX350.

Benched a little while ago on either I think AGESA 1100C or 1200 - the more casual games I play on my HTPC benefit massively (25-40%) between DDR4-2133 and DDR4-3600, but literally zero difference between 3600 16-16-16, 4000 16-16-16 and 4200 17-18-18. Tell me that's still a memory bottleneck...it's the Vega 7 that is constantly pegged.

And no, they're not CPU-bound in the slightest. One of them is 60-fps hardcoded and how it behaves on my 5900X, now *that's* CPU bound.

Starting to sound like a broken record, but while it's a given that 2133 to 3600 is going to be a huge difference, up where the IF limit of these new APUs actually lies it really appears that AMD needs to either give back our CUs or give us RDNA.

 

[watzupken]

I didn't say its performance isn't reliant on memory bandwidth. You said they're bandwidth-starved. As we're not talking about mobile Cezanne here, these desktop APUs are only bandwidth "starved" if you're below 1900MHz MCLK and FCLK, in which case you're already doing something wrong considering these chips thrive in the 2000-2300MHz IF range, and 2x8GB SR kits that can hit those speeds aren't prohibitively expensive either. Neither did I say that latency was an issue for the APUs.

No one's asking for a 5700XT here. All we want is a badly needed newer arch on the same number of CUs, or a larger core to get back to where we were + some higher 7nm clocks. If we weren't in the middle of a (partially self-inflicted) N7 supply shortage, there'd be no excuse afforded to AMD for this shit, "hey we took away some of your CUs which we justified by giving you more cores, but have some more clockspeed to level the playing field......sorta...kinda...not really......and now we're going for round 2!"

I am not sure what kind of playing field are you looking at. This CPU mentioned here is a desktop variant, and I don't see Intel posing any challenge here. Rocket Lake comes with a meagre 32 EU config of Xe graphics, and it is quite significantly behind in terms of iGPU performance. Even on the mobile side of things, Vega 8 is still competitive against Xe graphics even with its max 96 EUs config. Both iGPUs trade blows in games, so there is no one that performs decisively better. In fact, I think Intel will also be stuck with Xe graphics for quite a number of years.

I do understand the disappointment as I am looking forward to RDNA2 iGPU too. But if it is not ready, its not ready. I believe AMD have made a calculated decision on not bringing in RDNA2. So we just gotta wait.

 

[tabascosauz]

I am not sure what kind of playing field are you looking at. This CPU mentioned here is a desktop variant, and I don't see Intel posing any challenge here. Rocket Lake comes with a meagre 32 EU config of Xe graphics, and it is quite significantly behind in terms of iGPU performance. Even on the mobile side of things, Vega 8 is still competitive against Xe graphics even with its max 96 EUs config. Both iGPUs trade blows in games, so there is no one that performs decisively better. In fact, I think Intel will also be stuck with Xe graphics for quite a number of years.

I do understand the disappointment as I am looking forward to RDNA2 iGPU too. But if it is not ready, its not ready. I believe AMD have made a calculated decision on not bringing in RDNA2. So we just gotta wait.

Literally no one talked about Intel in that post, nor did anyone claim that UHD750 is anything but shit. Literally nobody. The desktop distinction was to make sure 3200 JEDEC SO-DIMMs wouldn't be used as a lazy example of "bandwidth starved".

AMD's only objective for Cezanne was to gain mobile marketshare. The Zen 3 cores will do that when paired with RTX 30 mobile. Result? Renoir + Zen 3.

Renoir was pretty reasonably priced, grey market non-warranty notwithstanding. I wonder if this gen whether retail or grey market will see APUs being introduced to scalpers for the first time. I hope not.