AMD Ryzen 9 7940HS analysis - Zen4 Phoenix is ideally as efficient as Apple

[Redaktion]

After quite a delay, the first 4 nm chips are being brought onto the market. We finally got our hands on the new Pheonix series Ryzen 9 7940HS in order to test the processor's performance and check out its efficiency levels using different power levels. Was the wait worth it?

https://www.notebookcheck.net/AMD-Ryzen-9-7940HS-analysis-Zen4-Phoenix-is-ideally-as-efficient-as-Apple.713395.0.html

[DXC0]

This efficiency comparison would make sense with the U lineup.

[NikoB]

As I feared, at the end we have a complete EPIC FAIL from AMD, according to the numbers in this review.

It was supposed to be the king of single-core performance. And the power efficiency king in the multi-threaded test, greatly outperforming the 7945HX in 1W performance.

We all see some kind of mess...

Is it deliberate suppression of energy efficiency, for the sake of sales of the 7x45 series, or a complete failure in the design of cores for "4/5nm"? I don't understand how this could happen.

I was expecting much better power efficiency at 45W PL1 compared to the 7945HX, and significantly faster single-threaded performance with a PL2 of around 70-80W.

I hope it's Asus who screwed up the design somehow... but if the same happens in other laptops, then this is AMD's failure.

The video part is even harder to evaluate against such a background, especially if the memory is worse than lpddr5 7500.

In general, it turns out that in the fall AMD will release Zen4 +, in which at least somehow fix the problems. Hope.

Well, for 2023 the king is 7945HX, but laptop manufacturers are dishonored in models with it, they don't use its 28 pci-e 5.0 lines in any way. I expected that 2-4 USB40 ports would be the minimum in the worst laptop with this SoC, but as we can see even in the top solutions there is no USB40, it's a shame, etc. and so on.

I don't understand why AMD made such a cool pci-e 5.0 controller with 28 free lines with such a slow memory controller - at least they made 4 channels, increasing the memory controller bus to 256GB (64x4). As a result, we would get a balanced solution that is really capable of pumping 28 pci-e 5.0 lines simultaneously and ensuring the operation of the TB5 x 2 analog for a future 4090 upgrade via an eGPU over the pci-e 3.0 x16 bus, through 4 x USB40 aggregation.

I'm even somehow not interested in trying to evaluate the neural network unit and igpu in 7940HS against the backdrop of such a shame AMD.

It's a pity reality shattered all hopes in 2023.

[GameXGR]

As I feared, at the end we have a complete EPIC FAIL from AMD, according to the numbers in this review.

It was supposed to be the king of single-core performance. And the power efficiency king in the multi-threaded test, greatly outperforming the 7945HX in 1W performance.

We all see some kind of mess...

Is it deliberate suppression of energy efficiency, for the sake of sales of the 7x45 series, or a complete failure in the design of cores for "4/5nm"? I don't understand how this could happen.

I was expecting much better power efficiency at 45W PL1 compared to the 7945HX, and significantly faster single-threaded performance with a PL2 of around 70-80W.

I hope it's Asus who screwed up the design somehow... but if the same happens in other laptops, then this is AMD's failure.

The video part is even harder to evaluate against such a background, especially if the memory is worse than lpddr5 7500.

In general, it turns out that in the fall AMD will release Zen4 +, in which at least somehow fix the problems. Hope.

Well, for 2023 the king is 7945HX, but laptop manufacturers are dishonored in models with it, they don't use its 28 pci-e 5.0 lines in any way. I expected that 2-4 USB40 ports would be the minimum in the worst laptop with this SoC, but as we can see even in the top solutions there is no USB40, it's a shame, etc. and so on.

I don't understand why AMD made such a cool pci-e 5.0 controller with 28 free lines with such a slow memory controller - at least they made 4 channels, increasing the memory controller bus to 256GB (64x4). As a result, we would get a balanced solution that is really capable of pumping 28 pci-e 5.0 lines simultaneously and ensuring the operation of the TB5 x 2 analog for a future 4090 upgrade via an eGPU over the pci-e 3.0 x16 bus, through 4 x USB40 aggregation.

I'm even somehow not interested in trying to evaluate the neural network unit and igpu in 7940HS against the backdrop of such a shame AMD.

It's a pity reality shattered all hopes in 2023.

It's pushed to 5.2Ghz while Apple chilling at below 4, no wonder the efficiency is so bad, not even counting for the fact that it's X86. With such a small IPC uplift from Zen 3+, they are forced to push it beyond any one who cares about efficiency should, to get a gen over gen kind of uplift. I'd rather take 16 cores at a mere 2.5 GHz over 8 running at 4.0, the latter is bound to be more efficient, as 4 is 60% higher but requires multiple times more power. This might be offsetting the inefficiencies of having two CCD's. 7940HS/U could be much better at 15/28 Watts, I hope, there are leaks of AMD possibly combining 16 core Zen5 mobile and 40CU "RNDA 3.5", hope that's not wasted on dGPU laptops lol.

[Poster]

So after the wait of just a review, overall it's not a substantial gain over the 6000 series. I then should look to buy a laptop from one of the few 6000HS or rebadged 7035HS series that will last me till... Oh well, such is as things in the world - a disappointment. That idle power consumption though, it's abysmal!

[flow x13 owner]

I don't get the reported idle power consumption numbers.

On my 7940HS flow x13 (received 3 days ago), the CPU power package (as reported by HWinfo) drops below 1W at idle.

Is the 3W reported here a G14 specific issue?

[:)]

Soldiers of Rohan! Hold! Remain steadfast!

Do not buy anything from these companies. They're trying to make a killin' during a recession. Literally blasphemy.

Stuff is still way too ridiculously priced for what they're. (I mean, €800 for ROG Ally in EU? Ayy Lmao. Do they realize that this sheety gtx 1650 tier graphics is barely enough to run anything in 2023? At most, these weak rdna3 igpu's should be relegated to sub < €400 chromebooks by now)

Hold onto your old toaster with your dear life and wait until Zen 5 APU's coming mid-to-late 2024.

Hopefully prices will go back to more reasonable levels by then. If not, oh well. At least we tried.

:)

[Andres Callegari]

Those people highlighting single core performance live in the prehistoric cave times or directly/indirectly paid by Intel contrafund to spread Intel beneficial misinformation to naive people.

It's like praising a car above others because it has higher performance when disabling the other 3 wheels. Utterly useless..

Very unrealistic real world usage, which is misinformation deeply marketed by Intel, since it is behind its competitors, under-performing in real world usage, and using synthetic benchmark programs that benefit them only. Reminder that Intel disables compiler optimizations when it detects competitors hardware and hides that and other legally required disclose information by burying the links/information in their information.

Comparison must be with all cores enabled, on real life used software, and benchmarking the power used by the hardware.

If a sport vehicle uses double the fuel to run 0-60 mph in 3 seconds, it can't be compared favorably with a vehicle that achieves the same using half the fuel. Yet, Intel marketing is an expert in confusing customers that its hardware consuming twice the power is a good thing, its ok, and  makes them better than the compatition.!!

[Paviko]

I can't believe these numbers. It's the biggest letdown of 2023. Going from TSCM 7 to TSCM 4 gives absolutely nothing. 190 points/W for 4nm versus 182 points/W for 7nm. I was waiting for Zen 4 notebooks, but it's not worth it.

[h4]

To author
I take two objections to this article.

The first: please stop phishing.
The title says "Zen4 Phoenix is ideally as efficient as Apple" and the conclusion says "Zen4 Phoenix didn't quite live up to expectations at times".

The second: measuring actual battery life is required for the conclusion, I think.
AMD often had longer battery life on WiFi v1.3, etc. (although ADL and Zen3+ were comparable under Office-like loads).

I can't believe these numbers. It's the biggest letdown of 2023. Going from TSCM 7 to TSCM 4 gives absolutely nothing. 190 points/W for 4nm versus 182 points/W for 7nm. I was waiting for Zen 4 notebooks, but it's not worth it.

As you can see, the performance at the same power consumption is 20% higher, which is exactly what node advancement brings. However, AMD chose to prioritize performance and increase the clock to offset this.

In the ADL vs Zen3+ era, the Zen3+(8C) won at around 35 W as the dividing line and below that, the ADL(6P8E) was superior. Based on these measurements, in MTL vs Zen4, MTL(6P8E) would win at 15 W or higher and Zen4(8C) would win at 15 W or lower. However, at this idle power consumption, there may be a penalty of 9 W or less.

Those people highlighting single core performance live in the prehistoric cave times or directly/indirectly paid by Intel contrafund to spread Intel beneficial misinformation to naive people.
...
Very unrealistic real world usage

With 6 cores (12 threads) or more, MS Office and Photoshop rely on single-threaded performance; you can run benchmarks that include them, like PCMark or Crossmark, or you can make a huge Excel file yourself to see. The additional two cores are only beneficial at some times, such as Windows Update.

If you still doubt it, please download HWiNFO and log the utilization of each core. I finally came to the above conclusion using this method.

Certainly your argument makes sense if you are comparing Tiger lake vs Cezanne, but in the 8C vs 6P8E era, single threaded performance is gaining weight.

[usacomputer]

At last the AMD Zen 4 7040 begin to arrive.
And Intel fans don't want to acknowledge AMD's Great Work
It is very clear who is thinking of buying an ultrabook weighing at least 1kg, the only option is AMD Zen 4 7040, this time AMD has broken all the molds with its RDNA 3 together with Artificial intelligence that will be very useful for the new version of Windows 12, Office and the new Direct XII APIs all focused on Artificial intelligence which many hate but is currently the future of many and agony for others.
The news is somewhat impartial as always against AMD but for 90% of the population and you are not a gamer this Zen 4 7040 APU is ideal to be able to Play AAA games, render 3D, videos etc... and for artificial intelligence at a price of €1000
It is clear that next year Apple with the New M3 will surpass AMD and Intel but at what price €3000 I do not want to spend that much money enough with Zen 4 7040 of €1000

[prajaybasu]

In comparison, the current Raptor Lake generation Intel Core i7-13700H (which is practically a slightly faster Alder Lake Core i7-12700H)

13700H is not the same silicon as the 12700H. It supports DDR5-5200 and it's using "Intel 7 Ultra" process w/ Raptor Cove. It's more efficient because it's running at a higher frequency while consuming less power than the 12700H. That's the point of Raptor Cove, they fixed the V-F curve which is quite important because these P and H series CPUs have their voltage control fully locked since 12th gen (except 12900HK).

The only recycled silicon for Raptor Lake Mobile is the HX series, only 13850HX and higher variants are new silicon. Anything below that in HX is ADL with only DDR5-4800 support.

Brushing off Raptor Lake and overhyping TSMC N4 (= N5+), no wonder the author is disappointed.

[NikoB]

Soldiers of Rohan! Hold! Remain steadfast!

Do not buy anything from these companies. They're trying to make a killin' during a recession. Literally blasphemy.

Stuff is still way too ridiculously priced for what they're. (I mean, €800 for ROG Ally in EU? Ayy Lmao. Do they realize that this sheety gtx 1650 tier graphics is barely enough to run anything in 2023? At most, these weak rdna3 igpu's should be relegated to sub < €400 chromebooks by now)

Hold onto your old toaster with your dear life and wait until Zen 5 APU's coming mid-to-late 2024.

Hopefully prices will go back to more reasonable levels by then. If not, oh well. At least we tried.

:)

The dollar for 15 years has depreciated in purchasing power by 2 times.

The pre-top video card AMD 2008 - HD4850 (and then NVidia was weaker, because AMD was really the absolute top on the planet in the consumer segment for the summer of 2008 - HD4870) cost $220-240.

Now an analogue in the class of 2023 vs 2008 HD4850 but from NVidia - GTX4080 costs $1200-1300.

Thus, for exactly 15 years, the prices for video cards have speculatively increased, minus the devaluation of the dollar by 2 times, $1250/$230/2 ~ 2.7 times!

This is madness! Where is the technological progress? Even taking into account the devaluation of the dollar by 2 times in 15 years, the price of 4080 should be no more than $460-550 right now. At worst.

[LL]

Come to real world Andres Callegari please many processes are still done single core in 3D for example.

[Neenyah]

Those people highlighting single core performance live in the prehistoric cave times or directly/indirectly paid by Intel contrafund to spread Intel beneficial misinformation to naive people.

It's like praising a car above others because it has higher performance when disabling the other 3 wheels. Utterly useless..

Very unrealistic real world usage, which is misinformation deeply marketed by Intel, since it is behind its competitors, under-performing in real world usage, and using synthetic benchmark programs that benefit them only. Reminder that Intel disables compiler optimizations when it detects competitors hardware and hides that and other legally required disclose information by burying the links/information in their information.

Comparison must be with all cores enabled, on real life used software, and benchmarking the power used by the hardware.

The hell are you talking about? Is Adobe Photoshop real enough for you? 13:20 in the vid 👉 youtu.be/nTLs3-LQGjg

64C/128T Threadripper 5995WX gets DEMOLISHED (-27.17%. -36.61%, +9.35%, -54.88% (!)) against the 13900K. Guess why? Because single core performance is the most important metrics in most (not all, just most) of the professional software. Check the rest of the video.