ApocalypseeNow I know the problem, the laptop Ryzen APU is limited to DDR4-2400MHz while the i7-1065G7 in this review uses LPDDR4X-3733MHz. That is a massive memory bandwidth advantage on the Intel side.

Lets see it at 2400.

eidairaman1Lets see it at 2400.

What would be the point in purposefully reducing the memory speed on the Intel offering? Ideally they should be testing on whatever is available on the market, so if that is 3733 MT/s RAM then that is the performance we care about.

eidairaman1Lets see it at 2400.

What's the point? They designed this architecture to support fast RAM. It's an advantage they've earned.

Why didn't you demand tests of Vega 64 on GDDR5 instead of HBM2? :-)

danbert2000What would be the point in purposefully reducing the memory speed on the Intel offering? Ideally they should be testing on whatever is available on the market, so if that is 3733 MT/s RAM then that is the performance we care about.

So we can see whether the Intel IGP is better than the AMD IGP when the rest of the systems are as close to equal as possible? My Civic is faster than your Mustang when I fill up with gas and you fill up with water.

notbWhat's the point? They designed this architecture to support fast RAM. It's an advantage they've earned.

Why didn't you demand tests of Vega 64 on GDDR5 instead of HBM2? :)

While I concede to the fact that this Intel chip officially supports faster memory, your Vega64 example is quite silly. You can change system RAM around in a computer all day. What you cannot do, however, is pluck memory off of a video card and install your own... of a different type, no less.

hatWhile I concede to the fact that this Intel chip officially supports faster memory, your Vega64 example is quite silly. You can change system RAM around in a computer all day. What you cannot do, however, is pluck memory off of a video card and install your own... of a different type, no less.

So you have a low end IGP with high-end memory, and your IGP has no clue on how to fully tax that high-end memory. Really if you overclocked GDDR ram in general on video cards, does that bring any more performance to the table? Or do you agree that the gains are really minimal?

The only way to extract more out of a IGP is something like a MXM card or a HBM2 stack of chips on top of the chip. Otherwise your always bound to much slower DDR(4) memory which is often castrated in speeds and timings, due to cost / space / TDP savings.

I like that Intel is trying to take IGPs seriously now, but the Iris Plus (GT3e, or even just GT3) graphics aren't what end up in the majority of laptops and desktops. Most people end up with mainstream GT2 graphics with half the shaders, no eDRAM cache, and probably with the cheapest single-channel memory the OEM can find. In terms of units shipped and OEM design wins, Iris graphics make up a tiny sub-5% portion of the market, filling a niche that doesn't exist; Expecting your new laptop to have Iris Plus or Iris Pro is foolish because in terms of design wins and retail availability, you have extremely limited choice of which models come with the relevant intel CPU.

If people want ultraportable graphics performance they buy either a Ryzen-U APU or move on up to the MX150/MX250 options. If they don't want graphics performance they'll be buying something with a ULV Intel processor that likely has a severely downclocked GT2 implementation, starved of performance by the TDP limits of a thin ultraportable. The minute your requirements extend to bulkier laptops, there are dGPUs with vastly superior performance at competitive prices compared to the more expensive higher-end Intel CPUs with Iris graphics, so the Iris graphics niche is tiny and borderline irrelevant.

What Intel really need to do is focus on making their GT2 IGPs perform well with low memory bandwidth and no embedded cache.
That's what AMD did with Raven Ridge; Vega 8 with even slower RAM is acceptable for low-def mainstream gaming as long as the TDP isn't too restrictive (20W gets 90% of the performance available, 15W models struggle once the STAPM threshold is reached). Once you add faster, dual-channel RAM, even the MX150 looks unnecessary - and that's just talking about an 18-month old, lower-end Raven Ridge model. The newer Picasso Ryzen 7 3700U on 12nm is even more efficient and that gets decent performance out of just 15W.

wow this might be huge step on intel, finnaly we have decent intergrated intel :)

Chrispy_I like that Intel is trying to take IGPs seriously now, but the Iris Plus (GT3e, or even just GT3) graphics aren't what end up in the majority of laptops and desktops. Most people end up with mainstream GT2 graphics with half the shaders, no eDRAM cache, and probably with the cheapest single-channel memory the OEM can find. In terms of units shipped and OEM design wins, Iris graphics make up a tiny sub-5% portion of the market, filling a niche that doesn't exist; Expecting your new laptop to have Iris Plus or Iris Pro is foolish because in terms of design wins and retail availability, you have extremely limited choice of which models come with the relevant intel CPU.

That is the problem with a lot of laptops, not just Intel but AMD as well. Most of the reviews uses the best laptops of the bunch when testing for new gen CPU/APU while the majority of other cheaper laptops uses non upgradeable (sometimes) soldered single channel memory, neutered TDP rating with tiny cooler that can't maintain boost for long and throttle when both iGPU and CPU portion is put under load.

hatWhile I concede to the fact that this Intel chip officially supports faster memory, your Vega64 example is quite silly. You can change system RAM around in a computer all day. What you cannot do, however, is pluck memory off of a video card and install your own... of a different type, no less.

But you can dial down HBM2 to minimize Vega's memory advantage.

And it's not like you could magically test Ice Lake-U with desktop DDR4 anyway. In fact, most (if not all) Ice Lake-U notebooks shown to date are ultrabooks with soldered RAM.

ApocalypseeThat is the problem with a lot of laptops, not just Intel but AMD as well. Most of the reviews uses the best laptops of the bunch when testing for new gen CPU/APU while the majority of other cheaper laptops uses non upgradeable (sometimes) soldered single channel memory, neutered TDP rating with tiny cooler that can't maintain boost for long and throttle when both iGPU and CPU portion is put under load.

I wouldn't call that "a problem". It's a property.
In laptops performance is always adjusted for power consumption target or cooling limitations. There's a huge variance in performance on the same components.

Laptop users are more feature-oriented and they tolerate this (and the fact that one should read a review of the particular - not similar - laptop).
Desktop users are more performance-oriented.