Question Lenovo IdeaPad S540 testing - 4800U vs 10710U+MX250

[uzzi38]

The power of Renoid...

ENGLISH – Lenovo IdeaPad S540 13ARE vs 13IML – AMD Ryzen 7 4800U vs Intel Core i7 10710U – next lab501

Seriously, some of the results in here are crazy. Especially the power efficiency ones in there at the end.

 

[uzzi38]

Sigh. Just One time i didn't provide a link because it was 2AM here and it was a obscure hard-to-find early review and instantly I must be pulling info out of my ass right?

Here, referenced by no other than @chiakokhua:

https://twitter.com/x/status/1247174427155902466

There's a very detailed analysis of mem-speed and FCLK there, which is quite readable with Google translate (lots of AIDA64 pictures at source too):




Besides it should be expected that zen2 runs like zen2 not some magical fairy dust just because it's on a single Die

It's worth remembering that monolithic dies still use IF to hook up the CCXes to memory controllers etc (if you can find the diagrams these links are called CCMs). As such FCLK is still a thing, however with Zen and Zen+ it's just tied to MCLK (or was it UCLK, I forget now), so the end user never knew it could be adjusted seperately like you can with Zen 2. FCLK has always been the limiting factor in getting higher memory speeds with Zen, it's a bit of a shame that AMD didn't improve it further with Renoir to try and get LPDDR4X-4266 running in full 1:1 mode though, even if only when the la0top was plugged in.

 

[AtenRa]

Sigh. Just One time i didn't provide a link because it was 2AM here and it was a obscure hard-to-find early review and instantly I must be pulling info out of my ass right?

Here, referenced by no other than @chiakokhua:

https://twitter.com/x/status/1247174427155902466

There's a very detailed analysis of mem-speed and FCLK there, which is quite readable with Google translate (lots of AIDA64 pictures at source too):




Besides it should be expected that zen2 runs like zen2 not some magical fairy dust just because it's on a single Die

thx for the link.

 

[coercitiv]

High value proposition of the 4500U in a review specially made for the average Joe:

 

[mtcn77]

If only...

1:1 mode using LPDDR4X is impossible.

It doesn't have to, all dependency in fsb clock is ruled out. Now it can be as 'much' as 1.6:1.

 

[uzzi38]

It doesn't have to, all dependency in fsb clock is ruled out. Now it can be as 'much' as 1.6:1.

Yeah, that's the hope. That OEMs set it to run higher at stock.

 

[mtcn77]

Yeah, that's the hope. That OEMs set it to run higher at stock.

Do OEM's set IF clock? I don't know if it were a thing. We have discussed this with seronx, the result was IF could run @1066 to keep this 3200MHz memory rate. Comparatively with zen 3, the generational gap is present. Also, 2400 is likely, I haven't seen its mention.

 

[uzzi38]

Do OEM's set IF clock? I don't know if it were a thing. We have discussed this with seronx, the result was IF could run @1066 to keep this 3200MHz memory rate. Comparatively with zen 3, the generational gap is present. Also, 2400 is likely, I haven't seen its mention.

OEMs can set anything they want out of spec. AMD won't be very happy, and in some cases they actively enforce certain conditions are met, but I think this is not one of them. I think.

 

[AtenRa]

I still cannot figure out why you are talking about IF when memory controllers are decoupled from IF in Renoir.
And from the Chinese review, the LPDDR4X seems to be running at 4266 (2133MHz)

Also, I cannot see which laptop/s they used in the Chinese review.

 

[mtcn77]

And from the Chinese review, the LPDDR4X seems to be running at 4266 (2133MHz)

It will meet current generational requirements 2400:1066x3200mhz= 7200MHz?(ddr4x tops 6400MHz max).

 

[mtcn77]

There is also a report on ddr5 outlook. It will top at 8400MHz and AMD is most likely to be ready for that as well in the next 5000 generation.

 

[AtenRa]

It will meet current generational requirements 2400:1066x3200mhz= 7200MHz?(ddr4x tops 6400MHz max).

No idea what you mean with that

 

[mtcn77]

No idea what you mean with that

There was a mention of IF max clock. 2400 it was. Looking in the same mindframe with 1.6:1 ratio, it enables 7200MHz ddr(double data rate). Btw, 3200/1066 equates 1.5, sorry about that.

 

[mtcn77]

Woot, I'm most wrongly assuming too much;

It uses a low-power double-data rate four x(LPDDR4X) interface for 4266 MHz.
Fabric clock in Renoir is probably forced to asymmetrical 2:1, rather than 1:1.

.dram_speed_mts = 1600.0 => .fabricclk_mhz = 600.0 // DRAM @ 1600 MHz = IF @ 600 MHz
^-- probably a weird idle.
.dram_speed_mts = 2133.0 => .fabricclk_mhz = 1066.0 // DRAM @ 2133 MHz = IF @ 1066 MHz
.dram_speed_mts = 4266.0 => .fabricclk_mhz = 1600.0 // DRAM @ 4266 MHz = IF @ 1600 MHz
^-- If it was 3200 MHz, fabric clock would probably be also 1600 MHz.

Good guy, Seronx.
Max IF:FSB is 0.375:1.

 

[uzzi38]

I still cannot figure out why you are talking about IF when memory controllers are decoupled from IF in Renoir.

???

Just wondering, have you ever given this a read?

Infinity Fabric (IF) - AMD - WikiChip

Infinity Fabric (IF) is a proprietary system interconnect architecture that facilitates data and control transmission across all linked components. This architecture is utilized by AMD's recent microarchitectures for both CPU (i.e., Zen) and graphics (e.g., Vega), and any other additional...

en.wikichip.org

 

[AtenRa]

???

Just wondering, have you ever given this a read?

Infinity Fabric (IF) - AMD - WikiChip

Infinity Fabric (IF) is a proprietary system interconnect architecture that facilitates data and control transmission across all linked components. This architecture is utilized by AMD's recent microarchitectures for both CPU (i.e., Zen) and graphics (e.g., Vega), and any other additional...

en.wikichip.org

What im saying is that the IF frequency and the Memory Frequency in Renoir are decoupled, so how is the memory bandwidth related to IF frequency ???

 

[uzzi38]

What im saying is that the IF frequency and the Memory Frequency in Renoir are decoupled, so how is the memory bandwidth related to IF frequency ???

Yeah, you should probably read the article.

CCXes are connected to the memory controller via the Infinity Fabric. If your IF is running at 1600mhz, then your CCXes are receiving data at the same rate.

 

[TESKATLIPOKA]

4800U doesnt scale up at 25W TDP because this Laptop uses DDR-4 3200 (51.2GB/s) and not LPDDR-4X 4266 (68.3GB/s)

I asked the autor of that review for average clockspeed of IGP and he replied to my comment this, you can find It in the original review.

The measurements on page 14 were done during 10 runs of Cinebench R15.
During gaming, the average clock speed of the IGP was 1140MHz for 15W and 25W, and 950MHz for 12.5W. The peak was the same every time – 1750MHz

The biggest problem is not using DDR4 instead of LPPDRR4, but the identical average clockspeed for both TDPs! The only question is why the higher TDP didn't increase the average clockspeed of IGP and It stayed at 65% of max turbo clockspeed.

 

[TESKATLIPOKA]

I just looked at the bandwidth for Vega 64. The bandwidth is 484GB/s for a 64CU Vega with an average clockspeed ~1500-1600mhz(just a guess).
From this I got 43-46GB/s for 8CU Vega IGP with average clockspeed of 1140mhz. Only If the IGP was working close to It's max turbo you would need LPDDR4X 4266Mhz, but that's not happening with 4800U. I didn't include the CPU part in my calculations, because I don't know how much bandwidth It really needs or what optimizations AMD did.
What I wrote isn't really accurate so take It with a grain of salt.