AMD Ryzen 7 7840U Low-Power Processor Beats Previous-Gen Flagship Ryzen 9 6900HX

[btarunr]

AMD's 4 nm "Phoenix" silicon could serious turn the company's fortunes around in the ultra-thin notebook space. The 28-Watt Ryzen 7 7840U surfaced on Cinebench R23 screenshots, where it is shown beating the previous-generation 55 W flagship, the Ryzen 9 6900HX. If this is any indication of performance across the board, then the 15-28 W models of Ryzen 7040-series "Phoenix" could unleash an open-season against competing 15-28 W-category 13th Gen Core processors that have lower P-core counts, such as 2P+8E. The 7840U has all eight "Zen 4" CPU cores enabled, along with a fast RDNA3 graphics architecture based iGPU. In the screenshot, the 7840U is shown with a Cinebench R23 multi-threaded score of 14285 points, a score that is higher than that of the "Zen 3+" based 6900HX "Rembrandt," and a touch below the 45 W Core i7-12800H, which means it could have the upper hand over several 13th Gen and 12 Gen SKUs in the 15-28 W category.



View at TechPowerUp Main Site | Source

 

[TumbleGeorge]

Sweet results! Haven't patience to see price/performance numbers!

 

[ratirt]

That is a huge achievement if true. Wonder, how good these CPUs scale with power. hopefully up to a 120W it makes sense to bump the power.

 

[Rais]

The issue as always for AMD is the availability in the most decent notebook models.

 

[watzupken]

I actually don't think RDNA3 is significantly faster than RDNA2. The reason for the jump in gaming performance is likely attributed to the faster Zen 4 CPU cores, that really helps when running at resolutions of 1080p or lower. The other reason could be the use of faster LDDDR5 that may be used with these low powered chips.

 

[Nostras]

That is a huge achievement if true. Wonder, how good these CPUs scale with power. hopefully up to a 120W it makes sense to bump the power.

I seriously doubt it considering that AMD's mobile low power chips have always sucked a lot in power scaling. They're at their most efficient below 50W and plateau at about 65W or so. At 120W you're most definitely just better off getting a dragon range CPU.

 

[neblogai]

I actually don't think RDNA3 is significantly faster than RDNA2. The reason for the jump in gaming performance is likely attributed to the faster Zen 4 CPU cores, that really helps when running at resolutions of 1080p or lower. The other reason could be the use of faster LDDDR5 that may be used with these low powered chips.

CPUs do not matter for iGPU performance. What matters, is RAM speed, and GPU clocks. Which will be higher for these RDNA3 chips, as they are made on 4nm, thus will be able to work at higher clocks at same limited U-series power, compared to 6nm Rembrandt.

 

[hs4]

1. The HP Dragonfly Pro with 7736U is set to PL1=40W, PL2=51W
2. The 7700X achieve this performance somewhere in the 40-50W range
combining above two, the PL1=40W setting is the most likely.

 

[AnarchoPrimitiv]

I'm really excited for these APUs to become available in mini PCs....that 780M RDNA3 iGPU with 12 CU's seems like it could make for an excellent emulation machine and even for 1080p gaming. The 680m was already good at AAA gaming at 900p-1080p with settings adjustments, at least according to the reviews from ETA prime (the best youtube channel for mini PC reviews) so I would not be surprised in a mini PC that can turn up the wattage and use some 6000Mhz+ DDR5 (which ETA prime showed gave good performance improvements in mini PCs equipped with the 6000 series APUs) could achieve 60fps @ 1080p, and definitely do so with FSR. Either way, I think it'd make a great HTPC to run dolphin, project 64, etc and do some AAA gaming on.....though I hope AMD actually releases desktop APUs for the DIY market with the 780m iGPU in it (but I'd really like it if they increased rhe CUs to 16 or even higher).

 

[Daven]

CPUs do not matter for iGPU performance. What matters, is RAM speed, and GPU clocks. Which will be higher for these RDNA3 chips, as they are made on 4nm, thus will be able to work at higher clocks at same limited U-series power, compared to 6nm Rembrandt.

Some reviews found the iGPU faster in the 3DV variant vs non-3DV variant desktop processors. So what you say is not 100% true. Some CPU features like higher clocks and cache will increase game performance between otherwise identical iGPUs.

 

[hs4]

I'm really excited for these APUs to become available in mini PCs....that 780M RDNA3 iGPU with 12 CU's seems like it could make for an excellent emulation machine and even for 1080p gaming. The 680m was already good at AAA gaming at 900p-1080p with settings adjustments, at least according to the reviews from ETA prime (the best youtube channel for mini PC reviews) so I would not be surprised in a mini PC that can turn up the wattage and use some 6000Mhz+ DDR5 (which ETA prime showed gave good performance improvements in mini PCs equipped with the 6000 series APUs) could achieve 60fps @ 1080p, and definitely do so with FSR. Either way, I think it'd make a great HTPC to run dolphin, project 64, etc and do some AAA gaming on.....though I hope AMD actually releases desktop APUs for the DIY market with the 780m iGPU in it (but I'd really like it if they increased rhe CUs to 16 or even higher).

The time spy garphics score of 780M is said to be 2.5k - 3k, much lower than the score expected from the RDNA3 thermal limit (4k), but close to the score expected from memory bottleneck (3.3k) by DDR5-6400 (dual channel 100 GB/s). So that the score won't be increased if the number of CU is increased to 16.

Intel will introduce new iGPU and its upper tier will be almost the same to A370M which has 112 GB/s memory bandwidth. The score of A370M is around 3.6k, so the memory bottleneck likely limits the performance around 3k. Finally, both AMD and Intel's iGPU will perform similar by this bottleneck.

 

[TumbleGeorge]

We know that AMD APUs, as complex chips that do not contain Infinity Fabric and do not have the corresponding penalty, would work better with RAM, including, since the discussion also mentions future desktop versions, which have enough time to have suffered some RAM controller improvements compared to the first Ryzen 7000X batches produced. How about the desktop 7000G(Phoenix), would they work well with DDR5 7200+?

 

[hs4]

We know that AMD APUs, as complex chips that do not contain Infinity Fabric and do not have the corresponding penalty, would work better with RAM, including, since the discussion also mentions future desktop versions, which have enough time to have suffered some RAM controller improvements compared to the first Ryzen 7000X batches produced. How about the desktop 7000G(Phoenix), would they work well with DDR5 7200+?

It will simply increase bandwidth by 12%, probably 12 CUs will be too much. It will be necessary to increase the number of channels by using unified memory like Apple Silicon, or by loading dedicated HBM like Instinct MI or Xeon MAX. Otherwise, motherboard prices would be even higher, comparable to those for Threadripper (quad channel).

 

[TumbleGeorge]

It will simply increase bandwidth by 12%, probably 12 CUs will be too much. It will be necessary to increase the number of channels by using unified memory like Apple Silicon, or by loading dedicated HBM like Instinct MI or Xeon MAX. Otherwise, motherboard prices would be even higher, comparable to those for Threadripper (quad channel).

Official support of Ryzen 7000X for DDR5 5200 JEDEC. DDR5 7200 is 38.46% faster. I think that is enough to justify 16 CU iGPU in Ryzen 7000G APUs. We already see offers for DDR5 RAM which is faster than 7200(up to 8000 in this moment and probably more in this year before it's end).

 

[R0H1T]

We know that AMD APUs, as complex chips that do not contain Infinity Fabric and do not have the corresponding penalty, would work better with RAM, including, since the discussion also mentions future desktop versions, which have enough time to have suffered some RAM controller improvements compared to the first Ryzen 7000X batches produced. How about the desktop 7000G(Phoenix), would they work well with DDR5 7200+?

That's wrong, they do have IF but monolithic dies for some APU's at least till last gen.

 

[TumbleGeorge]

they do have IF

Do you sure for APU?
Hmm, I see IF between CPU and iGPU but for what AMD not using direct connection and whether this usage affects the same way as the connection to the i/o chiplet. Does memory controller ratio matter? I don't think that's the usage in this case?

 

[R0H1T]

Yes.

AMD Ryzen 7 5700G Review​

 

[TumbleGeorge]

Yes.

AMD Ryzen 7 5700G Review​

In the schematic, the RAM controllers are not connected to the CPU and to the iGPU via Infinity Fabric, but there is a direct connection, so there are differences with the 7000X.

 

[R0H1T]

Yes but you said there's no IF in APU's which is wrong, this is/was last gen APU based on GCN. APU's have been monolithic unlike Zen4 last year ~ remember 7950x is also an APU. So you're half right, presumably talking about the chiplet penalty?

 

[TumbleGeorge]

Yes but you said there's no IF in APU's which is wrong, this is/was last gen APU based on GCN. APU's have been monolithic unlike Zen4 last year ~ remember 7950x is also an APU. So you're half right, presumably talking about the chiplet penalty?

7000G also will be monolithic for sure. And yes I talking about IF penalties when is using between I/O chiplet which contains RAM controllers and other chiplets in N00X CPU series. Role of IF in N00G series is a little different and it doesn't seem to be on the signal path from the RAM to the logic parts.

 

[dragontamer5788]

"Infinity Fabric" is just a marketing name. Back in Zen1, there already was a difference between IFOP and IFIS. I don't know how AMD has changed "Infinity Fabric", but rest assured that its complex, constantly changing, and proprietary (closed documentation, secret to AMD).

On-Die communications is faster than Die-to-Die communications, which is faster than socket-to-socket communications. But all three communications will be called "Infinity Fabric" at the marketing level... possibly the Engineering level (if there are shared concepts between them).

For an example of how Infinity Fabric worked in Zen1, see here: https://en.wikichip.org/wiki/amd/infinity_fabric#Scalable_Control_Fabric_.28SCF.29

https://en.wikichip.org/w/images/a/a6/amd_zeppelin_basic_block_4-dies.svg

------------

EDIT: Raven Ridge's infinity fabric was apparently like this:

"SDF" is the on-die Data plane, or a more detailed part of Infinity Fabric. (Citation: https://fuse.wikichip.org/news/1596/hot-chips-30-amd-raven-ridge/3/)

UMC being the memory / DDR4 controllers, CCX being the CPU cores. So UMC was connected to the same internal SDF switch as CPU, meaning CPU is going to access DDR4 a bit faster than the GPU. At least for Raven Ridge.

 

[neblogai]

Some reviews found the iGPU faster in the 3DV variant vs non-3DV variant desktop processors. So what you say is not 100% true. Some CPU features like higher clocks and cache will increase game performance between otherwise identical iGPUs.

Those reviews failed to get proper performance of non-3D chip igpus, and that is how they got `performance increase` from 7950x3D igpu. It was PCGamer and Tomshardware, and both redid their tests, and corrected their mistake. So, the point still stands: gaming on igpus does not benefit from faster CPUs.

 

[Sonic3389]

The issue as always for AMD is the availability in the most decent notebook models.

This I stick with the HP Spectre x360 range as it's their top end convertible ultrabook series. It's never had an AMD option and I don't believe it ever will.

 

[Punkenjoy]

I stopped to be really hyped by those ultrabook iGPU. They aren't there to be really powerful. They are there to just be good enough and that is mostly it.

If AMD wanted to have a real usable APU they could probably do it. The bandwidth could still be an issue, (even the 6500XT have over 140 GB/s of bandwidth with its 64 bit bus) but with a smaller process node, same size or slightly more infinity cache, it could probably still make those cards obsolete and provide some decent gaming performance. Just something near or slightly above the Xbox series S GPU and that would be great.