Quote from: Valantar on June 25, 2020, 10:54:06
...

Amazing. Thanks!

Quote from: dms on June 24, 2020, 23:33:51

Quote from: Valantar on June 19, 2020, 19:14:16

Quote from: dms on June 19, 2020, 11:18:15

Quote from: Valantar on June 18, 2020, 18:33:50
Also, I would assume this prototype laptop uses LPDDR4X (no reason they would cheap out for a prototype), while your Vega 7 benchmark is done with DDR4-3200, so I would assume a small further performance uptick for LPDDR4X Vega.

I would suspect it's the other way around because LPDDR memory has substantially lower bandwidth.

For example, DDR4-3600 has around twice the bandwidth of LPDDR4x-3733.

Per channel, yes, as LPDDR uses 32-bit (or even 16-bit for mobile) channels while DDR uses 64-bit channels. But both Ice Lake and Renoir, and likely also Tiger Lake, can run two LPDDR channels for each of their DDR channels (each controller can act as a single DDR channel or two "virtualized" LPDDR channels). So aggregate bandwidth for any properly configured LPDDR system is significantly higher than the same SoC with DDR due to the higher clocks.

Interesting, I didn't know that. Thanks for clarifying.

Not doubting you but is this well-known or do you have some kind of reference? Most comments I've seen (mainly on reddit tbh) seem to indicate that DDR >>> LPDDR in terms of bandwidth.

No problem :) I wouldn't call it common knowledge, and your misconception is indeed a common one. It's not that odd, given that in previous generations LPDDR has been slower both per channel and in aggregate numbers, mainly as the clock speeds haven't been that high (LPDDR3 only hit 2133MT/s after DDR4 at similar or higher speeds had already been established as an industry standard). It's only with LPDDR4X and its very high clock speeds that LPDDR has properly taken the overall performance crown. Though we don't know how long that will hold as DDR5 is on its way (likely hitting desktops in 2021) - but then again so is LPDDR5, which already exists in a handful of high end phones.

As for the channel layout, AMD detailed it when Renoir was first presented, but I would assume that's the type of detail most publications would skip over as it gets too technical for broader audiences. AnandTech covered it nicely in their launch coverage, and the following AMD slide pretty much sums up everything one needs to know:
(https://images.anandtech.com/doci/15624/2%20AMD%20Ryzen%20Mobile%20Tech%20Day_General%20Session_Architecture%20Deep%20Dive-page-022_575px.jpg)
(large version of the slide is linked in the AT article above.)

DDR4 still has two advantages: capacity (64GB or possibly 128 given dense DIMMs) and expandability (possibility of non-soldered memory), with LPDDR4X topping out at 32GB max and only existing in soldered form (the low voltages likely wouldn't work for a DIMM due to resistance in the contacts etc.).

Quote from: Valantar on June 19, 2020, 19:14:16

Quote from: dms on June 19, 2020, 11:18:15

Quote from: Valantar on June 18, 2020, 18:33:50
Also, I would assume this prototype laptop uses LPDDR4X (no reason they would cheap out for a prototype), while your Vega 7 benchmark is done with DDR4-3200, so I would assume a small further performance uptick for LPDDR4X Vega.

I would suspect it's the other way around because LPDDR memory has substantially lower bandwidth.

For example, DDR4-3600 has around twice the bandwidth of LPDDR4x-3733.

Per channel, yes, as LPDDR uses 32-bit (or even 16-bit for mobile) channels while DDR uses 64-bit channels. But both Ice Lake and Renoir, and likely also Tiger Lake, can run two LPDDR channels for each of their DDR channels (each controller can act as a single DDR channel or two "virtualized" LPDDR channels). So aggregate bandwidth for any properly configured LPDDR system is significantly higher than the same SoC with DDR due to the higher clocks.

Interesting, I didn't know that. Thanks for clarifying.

Not doubting you but is this well-known or do you have some kind of reference? Most comments I've seen (mainly on reddit tbh) seem to indicate that DDR >>> LPDDR in terms of bandwidth.

That keyboard font looks like the one that ASUS uses for its zen books.

Considering that Renoir laptops are still trickling onto the market, it's not at all too late. If these indeed beat Renoir, and arrive in July as rumored, Intel will have a strong competitive position for the next 6-10 months until Ryzen 5000 APUs arrive.

Like others have said, it's a bit late for Intel. Also, AMD can probably counter this just by going RDNA, not even RDNA 2, and with DDR4x

When Intel really releases these chips, AMD already will have newer chips, so Intel is now always a generation too late and only wins on single core performance. Also drivers from AMD are for graphics much much better, so this game is no example of "all world" experience.

The nice thing is that most 2021 PCs or Notebooks will have near MX350 Zgpuperformance without additional ram chips or 10 - 25W additional chips.

Quote from: dms on June 19, 2020, 11:18:15

Quote from: Valantar on June 18, 2020, 18:33:50
Also, I would assume this prototype laptop uses LPDDR4X (no reason they would cheap out for a prototype), while your Vega 7 benchmark is done with DDR4-3200, so I would assume a small further performance uptick for LPDDR4X Vega.

I would suspect it's the other way around because LPDDR memory has substantially lower bandwidth.

For example, DDR4-3600 has around twice the bandwidth of LPDDR4x-3733.

Per channel, yes, as LPDDR uses 32-bit (or even 16-bit for mobile) channels while DDR uses 64-bit channels. But both Ice Lake and Renoir, and likely also Tiger Lake, can run two LPDDR channels for each of their DDR channels (each controller can act as a single DDR channel or two "virtualized" LPDDR channels). So aggregate bandwidth for any properly configured LPDDR system is significantly higher than the same SoC with DDR due to the higher clocks.

Quote from: DavidC1 on June 19, 2020, 00:18:51

Quote from: Bogdan Solca on June 19, 2020, 00:08:01We tested a Microsoft Surface Book 3. Keep in mind that this test here was for that specific level I mentioned, while the test you are referring to is for another less demanding map.

Thank you. You haven't tested the device yet though? Doesn't show up in the database. What PL1 setting if I may ask?

We were already reviewing this model and we decided to test the BF5 map on it since it was handy :) The full review should drop tomorrow. As for the PL1 setting, all games are tested with full power specs.

Quote from: Valantar on June 18, 2020, 18:33:50
Also, I would assume this prototype laptop uses LPDDR4X (no reason they would cheap out for a prototype), while your Vega 7 benchmark is done with DDR4-3200, so I would assume a small further performance uptick for LPDDR4X Vega.

I would suspect it's the other way around because LPDDR memory has substantially lower bandwidth.

For example, DDR4-3600 has around twice the bandwidth of LPDDR4x-3733.

Quote from: Bogdan Solca on June 19, 2020, 00:08:01We tested a Microsoft Surface Book 3. Keep in mind that this test here was for that specific level I mentioned, while the test you are referring to is for another less demanding map.

Thank you. You haven't tested the device yet though? Doesn't show up in the database. What PL1 setting if I may ask?

Quote from: DavidC1 on June 18, 2020, 23:55:49

QuoteNow, we compared these results with the current Ice Lake Iris Plus G7 iGPU on the i7-1065G7,

And with what 1065G7 system?

The crappy HP Pavilion 15 with single channel memory, or the Dell Inspiron with DDR4-2666(along with who knows what other subpar settings)?

The Iris Plus G7 in the 1065G7 gets 17 fps in BF V 1080p High according to your results.

We tested a Microsoft Surface Book 3. Keep in mind that this test here is for that specific map Shrout is showing in his clip, while the test you are referring to is for another less demanding map.

QuoteNow, we compared these results with the current Ice Lake Iris Plus G7 iGPU on the i7-1065G7,

And with what 1065G7 system?

The crappy HP Pavilion 15 with single channel memory, or the Dell Inspiron with DDR4-2666(along with who knows what other subpar settings)?

The Iris Plus G7 in the 1065G7 gets 17 fps in BF V 1080p High according to your results.

This looks promising, but Intel desperately needs to step up their driver game for this to be anything but a flop.

Also, I would assume this prototype laptop uses LPDDR4X (no reason they would cheap out for a prototype), while your Vega 7 benchmark is done with DDR4-3200, so I would assume a small further performance uptick for LPDDR4X Vega. Nonetheless, finally we're seeing some actual competition in iGPU performance! This is great!

Can't wait for 11th gen Tiger lake laptops in premium category. I would love to see some and alternative s as well but apparently the market is rigged. So all I can hope for is that Intel gets it right this time.