News Ryzen AI 9 HX 370 approaches Ryzen 7945HX3D Geekbench performance while chilling in Silent Mode

[Admin]

An Asus ProArt 16 laptop was tested in Geekbench sporting AMD's brand new Ryzen AI 9 HX 370 processor. The chip scored impressive results, nearly matching AMD's previous-gen 16-core mobile CPU despite running in the laptop's silent mode.

Ryzen AI 9 HX 370 approaches Ryzen 7945HX3D Geekbench performance while chilling in Silent Mode : Read more

 

[abufrejoval]

I'm also seeing this Strix Point as quite literally a game changer, because it's the first real console capable design that is sold as a PC with either Steam deck, NUC/console or laptop incarnations and quite a potential range with little more than power presets.

Hopefully availabilty and prices for the various incarnations will be good enough to actually fulfill all that potential and vendors won't cut into expandability via some odd USB/TB, NVMe, WLAN or RAM options.

 

[Deleted member 2731765]

It was also tested in Geekbench 5 and OpenCL tests. In Geekbench 5.4.5, the chip scored 1847 points in single-core and 14,316 points in multi-core tests.

Finally, in the OpenCL test, the RDNA 3.5 (Radeon 890M) iGPU scored 41,995 points. Here the 890M's graphics performance is higher than 680M and 780M igpu scores, almost 38-59% faster.

And almost closing the gap with some entry level discrete mobile/desktop GPUs from the past, 1650 Ti/RTX 2050. But we can't solely rely on OpenCL scores.

The performance is low, since this could also be an early sample, apart from the Silent Mode test environment being used.

The 2.0 GHz base clock matches the specs, but the log file shows a boost clock of around 4.2 GHz, which is far below its highest-rated clock rate of 5.1 GHz.

 

[Deleted member 2731765]

The overall performance is still low. The log file shows a boost clock of around 4.2 GHz, which is far below its highest-rated clock rate of 5.1 GHz.

 

[Deleted member 2731765]

Article::

"Compared to its predecessor, the Ryzen 9 8940HS, the AI-branded chip is 6.5% faster in the single-core test and 17% faster in the multi-core test. The Ryzen 9 8940HS pumped out scores of 2,380 and 11,775, respectively, for the single and multi-core tests."

BTW, the correct/actual SKU product name is Ryzen 9 8945HS APU which is the flagship of the "Hawk Point" lineup. Not 8940HS.

The 2,380 and 11,775 scores exactly match with this SKU.

https://www.amd.com/en/products/processors/laptop/ryzen/8000-series/amd-ryzen-9-8945hs.html

 

[Notton]

I want to see the AI 9 HX 370 beating the 7945HX3D, not catching up to it.
If the chip is that fast without 3D cache... would it be faster if it had it, or are we going to run into thermal problems?

 

[TechyIT223]

Wait for proper third party benchmarks as well

 

[Deleted member 2731765]

If the chip is that fast without 3D cache... would it be faster if it had it, or are we going to run into thermal problems?

As per leaks, Strix Point lineup won't come in any X3D flavor. I expect the next-gen "Fire Range" APUs to succeed the current 7045X3D "Dragon Range" lineup, including the flagship Ryzen 9 7945HX3D.

The cTDP value set by AMD for this new Ryzen AI chip is 15-54W ! Whereas, Dragon Range's configurable TDP (cTDP) is 45/55-75W. So a direct comparison of these two chips would be tricky somtimes.

At least for the iGPU part/graphics, all Dragon Range chips only sport 2 CU core count 610M processor, vs 16 CU on this top Strix Point AI 370 part.

 

[Alvar "Miles" Udell]

I'd trust Cinebench more on this type of test than Geekbench as it taxes the CPU quite heavily under its multicore test, but these kind of results should be expected and -demanded- and not be a surprise. Generational IPC uplift should always be measured in double digits, and with process nodes changing essentially every generation then higher efficiency should also be expected.

 

[usertests]

I'm also seeing this Strix Point as quite literally a game changer, because it's the first real console capable design that is sold as a PC with either Steam deck, NUC/console or laptop incarnations and quite a potential range with little more than power presets.

By "console capable" I assume you're talking about the graphics meeting or exceeding the performance of an Xbox Series S. That could be the case if 890M graphics is ~40% faster than 780M. The CPU performance is overkill and doesn't matter, except for benefiting lower-power devices.

I want to see the AI 9 HX 370 beating the 7945HX3D, not catching up to it.
If the chip is that fast without 3D cache... would it be faster if it had it, or are we going to run into thermal problems?

It was running in a low power mode.

Just from some idiot math, a 16% IPC increase should have 12-core Zen 5 feeling like 14-core Zen 4. There are Zen 5c cores in play here with lower peak clock speeds, but with thermal constraints, 7945HX3D all-core might not be that high anyway. Zen 5's lack of significant clock speed increases probably means it's using less power and maintaining clocks well, from moving to a TSMC N4 node. However, Phoenix/Hawk was already on N4.

AMD hasn't added 3D V-Cache to mainstream APUs yet. 7945HX3D is a 7950X3D desktop CPU repackaged for BGA laptops and mini PCs. We have no idea if Strix Point, Strix Halo, etc. would be given 3D V-Cache. "Fire Range" will likely get 3D V-Cache, being the Zen 5 version of Dragon Range.

V-Cache would increase gaming performance for some games, but not every workload would benefit.

 

[thestryker]

It was also tested in Geekbench 5 and OpenCL tests. In Geekbench 5.4.5, the chip scored 1847 points in single-core and 14,316 points in multi-core tests.

Finally, in the OpenCL test, the RDNA 3.5 (Radeon 890M) iGPU scored 41,995 points. Here the 890M's graphics performance is higher than 680M and 780M igpu scores, almost 38-59% faster.

OpenCL doesn't really stress memory bandwidth so it's meaningless for determining IGP graphical performance. Once we have real graphics benchmarks it will give us insight as to how much performance is being left on the table though.

I'm also seeing this Strix Point as quite literally a game changer, because it's the first real console capable design that is sold as a PC with either Steam deck, NUC/console or laptop incarnations and quite a potential range with little more than power presets.

Unless there's some extra cache that hasn't been disclosed Strix Point is basically just 4 more CUs on a tweaked RDNA 3. It will likely be memory bandwidth constrained just like everything with the 680M and 780M are.

 

[Thunder64]

OpenCL doesn't really stress memory bandwidth so it's meaningless for determining IGP graphical performance. Once we have real graphics benchmarks it will give us insight as to how much performance is being left on the table though.

Unless there's some extra cache that hasn't been disclosed Strix Point is basically just 4 more CUs on a tweaked RDNA 3. It will likely be memory bandwidth constrained just like everything with the 680M and 780M are.

Could they put a small amount of inifinity cache on die for the iGPU to help with bandwidth, or would that blow up the die size too much?

 

[Deleted member 2731765]

Yes of course, we can't judge any iGPU performance just by OpenCL. Let's wait for some real gaming/graphics benchmarks. But this is all we have for now, at least some scores to ponder over, hehe.


Edit: For some odd reason, forum member Quotes are not working at my end.

 

[thestryker]

Could they put a small amount of inifinity cache on die for the iGPU to help with bandwidth, or would that blow up the die size too much?

They absolutely can, and the latest rumors have pointed towards just that for the Strix Halo SKUs. It's also possible Strix Point has some as well, but I think they would have been advertising that in their slides if that were the case.

 

[Thunder64]

They absolutely can, and the latest rumors have pointed towards just that for the Strix Halo SKUs. It's also possible Strix Point has some as well, but I think they would have been advertising that in their slides if that were the case.

I agree about Strix Point not having it. I could see 32MB on Strix Halo though. Certainly seems more realistic than vchace IMHO.

 

[Deleted member 2731765]

Yup.

As per one AMD's leaked slide/document, the future STRIX HALO lineup would have 32 MB of MALL cache that will be eliminating bandwidth bottlenecks for this iGPU.

A 144 page document was recently leaked. Whether this specs info is accurate and "final" can't be said with surety, but there is some legitimacy.

""It is worth noting that the Strix Halo APU will add a 32MB MALL cache to the SoC. Its function is similar to the current Infinity Cache, which can reduce the use of memory bandwidth and improve GPU performance. A Taiwanese manufacturer revealed that the Strix Halo GPU The performance is even comparable to the RTX 4060 Laptop, and the NPU computing power can reach up to 60 TOPS.""

Source:

AMD 下代 Strix Halo APU 處理器 40 CU 的 RDNA 3.5 GPU 16 核心 Zen 5

【勁到爆炸 🤣】Twitter 上流出了一份 144 頁的 AMD 官方文件，証實了明年推出的 Zen 5 APU 處理器規格將會非常強勁，並且首次出現巨大化的 Strix Halo APU 將會採用 Chiplet 多晶片設計，GPU 部份將會擁有高達 40CU (20xWGP) 的 RDNA 3.5 繪圖核心，主流級顯示卡也會被它輕鬆 KO。

www.hkepc.com

A Strix Halo render:

 

[abufrejoval]

Unless there's some extra cache that hasn't been disclosed Strix Point is basically just 4 more CUs on a tweaked RDNA 3. It will likely be memory bandwidth constrained just like everything with the 680M and 780M are.

IMHO the Steam deck has created a new class of consoles, the potentially (but not necessarily) mobile ones that sit below current consoles, but offer a far greater range of usability (you can take it to your friends and play jointly) and simply enough power to have fun.

In other words: that race, where the next generation console always needs to be far faster, offer more resolution etc. than the older, may be over.

Mobile gaming is the biggest global money maker, vastly exceeding consoles and gaming PCs. And for those who want a better gaming experience than on mobile phones, the mobile consoles may just satisfy the vast majority of those gamers, that so far would have had to go with a console or gaming PC.

And with a Halo part or next generation devices, there will be again more options and an upward path.

Basically I believe they're likely to siphon off so much attention and budgets, the niche for the gaming high-end is getting too small to sustain its growth.

And then these mobile consoles are PCs, NUCs or laptops, too. At a level of performance, competence, compactness and versatility, where there is simply no need to buy a PC or laptop for kids, students or for non-corporate work.

Smartphones to my chagrin have kept shy of qualifiying as desktop replacements, Strix will pull out all the stops.

Of course this is all my personal projection, but I know that if my kids wanted a gaming rig and a student laptop today, it would be a single device, not two.

(They actually got my workstation cast-offs, but that's beside the point)

 

[thestryker]

Mobile gaming is the biggest global money maker, vastly exceeding consoles and gaming PCs.

FWIW Console/PC has been catching up and mobile makes up less than 50% of gaming revenue now.

IMHO the Steam deck has created a new class of consoles, the potentially (but not necessarily) mobile ones that sit below current consoles, but offer a far greater range of usability (you can take it to your friends and play jointly) and simply enough power to have fun.

The handheld PCs are interesting, but none of them have really nailed the balance yet. I think Valve's custom SoC is the closest, but it could use a refresh with current technology.

I like my Ally a lot, but there are fundamental issues that can't really be addressed by a refresh like the Ally X. AMD keeps the highest IGP SKU for the highest CPU so their APUs are generally overkill on the CPU side.

It's possible Lunar Lake may end up being the right mix since it's high performance, optimized for efficiency, and doesn't have excess cores that aren't necessary for a handheld device.

I certainly think there's a good sized market there, and await better devices which will hopefully appear this year/next.

 

[usertests]

As per one AMD's leaked slide/document, the future STRIX HALO lineup would have 32 MB of MALL cache that will be eliminating bandwidth bottlenecks for this iGPU.

I was going to ask you why the recent leaks/docs keep calling it a MALL (Memory Access at Last Level) cache instead of just Infinity Cache, but I searched and found that they did that for RX 6000 too, so nothing unusual here:

https://www.tomshardware.com/news/amd-radeon-rx-6000-big-navi-mall

 

[Deleted member 2731765]

Yeah, MALL was mentioned before as well . Good find. Me covered it as well long back, when I used to work for this site ! I myself forgot it. Stumbled upon it via google search, lol.

AMD’s Big Navi 21 GPU lineup rumored to get a new ‘Power Saving’ feature dubbed as MALL

AMD appears to be reportedly working on a new feature called Memory Access at Last Level (MALL) for its upcoming next-gen RDNA2-based Big Navi graphics cards.

www.dsogaming.com

 

[TechyIT223]

Yup.

As per one AMD's leaked slide/document, the future STRIX HALO lineup would have 32 MB of MALL cache that will be eliminating bandwidth bottlenecks for this iGPU.

A 144 page document was recently leaked. Whether this specs info is accurate and "final" can't be said with surety, but there is some legitimacy.

""It is worth noting that the Strix Halo APU will add a 32MB MALL cache to the SoC. Its function is similar to the current Infinity Cache, which can reduce the use of memory bandwidth and improve GPU performance. A Taiwanese manufacturer revealed that the Strix Halo GPU The performance is even comparable to the RTX 4060 Laptop, and the NPU computing power can reach up to 60 TOPS.""

Source:

AMD 下代 Strix Halo APU 處理器 40 CU 的 RDNA 3.5 GPU 16 核心 Zen 5

【勁到爆炸 🤣】Twitter 上流出了一份 144 頁的 AMD 官方文件，証實了明年推出的 Zen 5 APU 處理器規格將會非常強勁，並且首次出現巨大化的 Strix Halo APU 將會採用 Chiplet 多晶片設計，GPU 部份將會擁有高達 40CU (20xWGP) 的 RDNA 3.5 繪圖核心，主流級顯示卡也會被它輕鬆 KO。

www.hkepc.com

A Strix Halo render:

Noice. Can we get access to the. Full PDF document though?

 

[Deleted member 2731765]

Can we get access to the. Full PDF document though?

Nah, we can't share "full" embargoed/confidential docs with anyone.

 

[Deleted member 2731765]

Some off topic stuff, but of course related to AMD chips.

AMD Ryzen 5 9600X was tested in the AIDA64 Cache and Memory benchmark.

This chip is an engineering sample so final performance is going to be different. In the test, the CPU scored some huge gains over the Ryzen 5 7600X regarding L1 and L2 bandwidth.

L1 Cache Bandwidth in GB/s (Read / Write / Copy / Latency):

• Ryzen 5 9600X (ES): 3756.4 / 1884.4 / 3755.9 / 0.8ns
• Ryzen 5 7600X: 2029.6 / 1026.9 / 2048.1 / 0.7ns

L2 Cache Bandwidth in GB/s (Read / Write / Copy / Latency):

• Ryzen 5 9600X (ES): 1874.6 / 1795.1 / 1859.7 / 2.8ns
• Ryzen 5 7600X: 1028.5 / 1017.0 / 1017.6 / 2.6ns

L3 Cache Bandwidth in GB/s (Read / Write / Copy / Latency):

• Ryzen 5 9600X (ES): 782.08 / 771.46 / 772.32 / 10.1ns
• Ryzen 5 7600X: 847.82 / 854.86 / 822.01 / 9.7ns