AMD Radeon 780M iGPU Performance Shows Up in Geekbench

[GFreeman]

A new benchmark of the AMD Radeon 780M RDNA3-based iGPU has now been spotted in Geekbench, showing it to be on par with the NVIDIA GeForce GTX 1650. The benchmark was done on the announced AMD Ryzen 7 7840HS CPU, which is a part of AMD's 4 nm Ryzen 7040 "Phoenix" lineup. The CPU in question is an 8-core/16-thread SKU with 40 MB of cache. It works at 3.8 GHz base and 5.1 GHz boost clocks and has a TDP of 35 W to 45 W. The same CPU comes with Radeon 780M RDNA3 iGPU with 12 Compute Units (CUs), for a total of 768 Stream Processors, and works at 2700 MHz.

The Geekbench OpenCL benchmark was fired up on a Chinese laptop with 32 GB of RAM and running on the Windows balanced power plan, where it scored a decent 36,757 points. This score puts it somewhere around the NVIDIA GeForce GTX 1650 and AMD Radeon RX 480 discrete graphics cards. This is an impressive result considering that this is a mid-SKU in the announced Ryzen 7040 series. Of course, we still do not know the specified TDP for this laptop, or what memory was used, so results might be slightly different. Since these laptops are yet to appear in retail, AMD is probably still tweaking the drivers, so we expect these numbers could be even higher for the full release.



View at TechPowerUp Main Site | Source

 

[AnarchoPrimitiv]

Can't wait for mini-pcs with these RDNA3 APUs to come out, though I really wish AMD would offer a desktop version with 16-24CUs

 

[Daven]

So the final, highest SKU might almost get to 1650 Super level of performance. That’s right around the most revered 1060. Impressive.

 

[TumbleGeorge]

Can't wait for mini-pcs with these RDNA3 APUs to come out, though I really wish AMD would offer a desktop version with 16-24CUs

This is the way! I also hope, but...nope! With DDR5 and with this small cache capacity which is not enough to masking good slow RAM speed...Propably around end of 2027 with 9000G series married with DDR6.

 

[chowow]

yes there's a certain amount of people being dreaming of a tiny APU that they can game on,performance of a RX480 okay!!

 

[Lew Zealand]

Let's compare this to an RX 6400:

Same core count of 768 (but RDNA3 vs 2 for whatever that's worth)
2700 vs 2300 MHz
12GB possible vs 4GB
96GB/s (128-bit 6000MT/s DDR5) vs 128GB/s (64-bit 16GHz GDDR6)

And it'll likely be lower than 96GB/sec as who knows if anyone will pair these with higher than 5600 (89.6GB/s).

768 cores seems to be an OK match for this memory bandwidth, adding more will just starve it ever more of memory bandwidth, resulting in smaller gains, and then only in some games.

 

[TumbleGeorge]

More interesting is that how RAM capacity is shared for graphic needs...12GB

 

[Denver]

Let's compare this to an RX 6400:

Same core count of 768 (but RDNA3 vs 2 for whatever that's worth)
2700 vs 2300 MHz
12GB possible vs 4GB
96GB/s (128-bit 6000MT/s DDR5) vs 128GB/s (64-bit 16GHz GDDR6)

And it'll likely be lower than 96GB/sec as who knows if anyone will pair these with higher than 5600 (89.6GB/s).

768 cores seems to be an OK match for this memory bandwidth, adding more will just starve it ever more of memory bandwidth, resulting in smaller gains, and then only in some games.

The fastest supported memory is LPDDR5x-7500, not 6000mhz. 25% faster.

https://www.amd.com/en/product/13036

 

[MaitreWindows]

It's not the first time that this exact CPU is referenced as having 40MB of cache on TPU, but I can't confirm this information. What AMD's site suggests is that it has 24MB (8MB L2 + 16MB L3). Who is right?
EDIT: Link I'm using to support this question -> AMD Ryzen™ 7 7840HS | AMD

 

[TumbleGeorge]

Не за първи път се споменава, че точно този процесор има 40MB кеш на TPU, но не мога да потвърдя тази информация. Това, което предлага сайтът на AMD е, че има 24MB (8MB L2 + 16MB L3). Кой е прав?
РЕДАКТИРАНЕ: Връзка, която използвам, за да подкрепя този въпрос -> AMD Ryzen™ 7 7840HS | AMD

+L1 and +additional local cache of iGPU (?) if it have.

 

[MaitreWindows]

+L1 and +additional local cache of iGPU (?) if it have.

Thanks for the input. I originally thought the same, but L1 is a mere 512KB, and all 3D cache CPUs so far have been suffixed -X3D, which this isn't. Additionally, it would be reasonable to think that it would be listed in the official spec sheet as this important design element would be set in stone by now (as opposed to frequencies that can be revised up until launch), since the CPUs are very likely in mass production.

 

[TumbleGeorge]

Thanks for the input. I originally thought the same, but L1 is a mere 512KB, and all 3D cache CPUs so far have been suffixed -X3D, which this isn't. Additionally, it would be reasonable to think that it would be listed in the official spec sheet as this important design element would be set in stone by now (as opposed to frequencies that can be revised up until launch), since the CPUs are very likely in mass production.

Apparently, further investigation is needed in the case to rule out criminal advertising falsification of the numbers. 40MB cache is also specified in wccfkek.

 

[MaitreWindows]

Apparently, further investigation is needed in the case to rule out criminal advertising falsification of the numbers. 40MB cache is also specified in wccfkek.

You are correct my friend! Wccftech being more of a "leak" website, I get it coming from them, but TPU is definitely news. Therefore, I agree with your suggestion that further investigation is, indeed, warranted.

 

[ymdhis]

Seems about as fast as I expected, if they make a desktop version with more than 12CUs I most likely will consider switching to AM5.

 

[Flanker]

Looks like we are getting there

 

[Minus Infinity]

It's not the first time that this exact CPU is referenced as having 40MB of cache on TPU, but I can't confirm this information. What AMD's site suggests is that it has 24MB (8MB L2 + 16MB L3). Who is right?
EDIT: Link I'm using to support this question -> AMD Ryzen™ 7 7840HS | AMD

I thin what has happened is it was recently reported AMD had downgraded Phoenix specs. Originally L3 cache was supposed to be 32MB and it is now only 16MB. So I agree the total cache is now 24MB as per AMD's website, but was originally planned to be 40MB. TPU needs to update it's articles. Really sucks why AMD did this since Phoenix does not have infinity cache either IIRC.

 

[TumbleGeorge]

I thin what has happened is it was recently reported AMD had downgraded Phoenix specs. Originally L3 cache was supposed to be 32MB and it is now only 16MB. So I agree the total cache is now 24MB as per AMD's website, but was originally planned to be 40MB. TPU needs to update it's articles. Really sucks why AMD did this since Phoenix does not have infinity cache either IIRC.

No, actually the large cache was "accepted" back in the days when rumors hinted that the iGPU could be 24 CU. They turned out to be incorrect, and this has been known for a long time. The decrease of the last days only concerns the expectations for basic and turbo frequencies of the CPU part in the (mobile)APUs, if I'm not mistaken.

 

[Vayra86]

I suppose my memory is weird, but when I see 780M the first thought is Nvidia re-releasing their Kepler Refresh Titan in mobile form.

Its not even strange given the fact they were peddling Turing GPUs not too long ago due to chip shortage.

 

[Denver]

No, actually the large cache was "accepted" back in the days when rumors hinted that the iGPU could be 24 CU. They turned out to be incorrect, and this has been known for a long time. The decrease of the last days only concerns the expectations for basic and turbo frequencies of the CPU part in the (mobile)APUs, if I'm not mistaken.

It's a matter of naming, just like the offboard RDNA3 GPUs, the XTX has 12288 FP32 cores (or 6144SPS).

 

[GhostRyder]

I mean, sounds good to me as I would be happy having a laptop with that setup for some light gaming!