Once someone goes to the $70 GPU the argument starts to fall off. Sure the sub $50 cards are only worth it for someone that just wants another monitor (BUT buy a display port card and monitors)The value to performance of these $50 cards is low.
Intel's integrated offerings already killed them off -- with the notable exception of the recent Macbook 15", where a dGPU is used primarily to support multiple 5k displays.
Speaking from bitter experience of the most beautiful, but also expensive and unreliable Mac Book Pro 17" ($3500 USD), now both the GPU and CPU will de-solder themselves from the motherboard!
100W is quite a bit higher TDP than any previous Mac Mini. The highest-power Minis I'm aware of are:
- The Quad-Core Mac Minis from 2011 & 2012 with 45W TDP. (e.g. i7-3720QM)
- The 2011 Mac Mini with dGPU with 35+26=61W TDP. (i7-2620M + Radeon HD 6630M)
Mind you, we can always dream that Apple will care about those of us who have wanted a Mac without a built-in display that fits between the Mac Mini and the Mac Pro.
Unlikely. A 100W TDP would mean that it would not be possible to run the laptop for any more than an hour at full load (on battery) or it would draw from both the wall and battery with their current power designs at full load. Also, That would be insane to cool in a MBP format (that's about 2x the thermal output of the current 15").
The Surface Book has a full-bore NVidia GTX 1060 in it. It cannot charge while the 1060 is going 100% in a modern AAA video game (yes: it uses up so much power, that if you plug it into the wall-outlet and game, you are LOSING power).
GPUs use up tons of power, but 99% of the time they sit around and are "off". There's no need to turn on the GPU when browsing Facebook, and modern Win10 can keep the graphics card off.
Anyway, 100W TDP is perfectly acceptable for laptops, as long as this is a "worst case" TDP. Besides, gamers do want to game, and many are willing to accept bad battery life for the ability to run Witcher 3 or Tomb Raider on the go.
No one will push a laptop to full load on a battery. Apple can pretty much throttle power on battery mode and no one will complain. As for the wall charger, the design can be changed. Alienwares with gtx 1070 can go upto 115w tdp with AC adapters without any problem. As for cooling, Asus ROG Zephyrus cools a gtx 1080 with a size comparable to macbook pro.
This statement is incorrect, 15" designs handle the 100W TDP just fine. Look at the Dell XPS 15 - it has an Intel H-series CPU (45W) plus a 1050 GPU (55W). While having all this power, it's slim and the cooling solution works OK under load (less reports about thermal throttling than latest MBP has) and with a 130W power brick it can be powered no problem unxer full load.
I hope it isn't bound for the next iMac. There are rumors of a new chassis for the next model but those are suspect in the face of the recent iMac Pro; namely why would Apple want the Pro version to be on an older chassis and therefor appear obsolete?
However if a new chassis is in the works and space is a premium it could lead to an iMac with chips similar to this but that would be disappointing as many are waiting for six core processors. For an Mini, wouldn't that TDP be too high?
Apple has an odd relationship with its desktop models. With the iMac Pro and "promise" of a new mini and Mac Pro there is a lot of hope for a resurgence however short it might be.
This isn't a mobile chip and with 100 watts thermal solution would be challenging in anything but a desktop enclosure. We need to wait and see if there is a mobile version coming and also the benchmarks. Is the watts worth it?
Mac Mini - TDP too high.
iMac - Lots of space for CPU + GPU.
We all thought the whole reason of sticking a dGPU close to CPU was because of space, hence a Notebook type machine. But if anyone could enlighten me why I want a desktop with Integrated dGPU rather then CPU+GPU combo which is likely cheaper, and easier to cool down.
Not to mention you are stuck with KabyLake 4 Core instead of the newer Coffee Lake with 6 Core.
For the same reason companies make SoCs. It saves money. Intel's idea with EMIB (multiple dies in one package) is to save money by using older semiconductor process for things such as the southbridge (ex: USB) that don't need the latest and most expensive process. This way, they'll also get more value from their investment in previous process nodes.
Routing HBM to the GPU using PCB traces is expensive, and in the low-to-mid end, saving money on PCB and cooling matters. If EMIB is going to be used anyway, why not have the GPU in there as well. With 1024-bit 4GB HBM on package, memory bandwidth won't be a problem, and the i7-8809G will likely be faster than a PS4, which means it'll be good enough for 1080p gaming.
KabyLake was probably chosen because this design was finalized before Coffee Lake. The next iteration will likely have more cores.
1. It will only make sense if Intel were actually manufacturing the GPU themselves as well. In this case they dont. And EMIB is an Intel specific technology, AMD doesn't use it when they package HBM with their dGPU.
2. AMD makes a profit before selling it to Intel, and Intel has to make a profits for integrating it with EMIB, i doubt the cost saving of using EMIB together with HBM, if that was true, out done the profits margin of both companies.
Weird how they include an HD 630 on the die. This eats up a fair chunk of die area, and eats into the thermal budget. It also means that the displays will probably be wired to the HD 630, so the Radeon can power down to save power. Taking out the HD 630 could have lowered the power usage to a level more palatable to a laptop computer at full blast.
This means that you have to pay a penalty to use the Radeon to render, but send its framebuffer result back to the HD630 for displaying. This is also the case current Nvidia Optimus or AMD Enduro systems. Draw backs from this approach are frame tear / lack of proper V-Sync and added latency.
If this ends up being used by Apple however, it is likely that a hardware demux will be used to connect the display to both graphics chips, and switch it back and forth on demand. Other vendors don't do this, because it is more expensive, and passing the display from one GPU to another requires the pixel clock rate (among other details) to be shared between the different GPUs, which is not something that the vendor driver (Intel/Nvidia/AMD) stacks support.
Note that such a hardware demux is used in previous generations of MacBook Pros, and it is not publicly documented, which means very poor and/or very late support in Open Source operating systems such as Linux.
Some system builders (Lenovo and Asus have done this, probably others) wire the internal display of laptops to the Intel GPU, and the external HDMI/DP to the discrete graphics. In the past, this allowed you to power more displays (when Intel GPUs were restricted to 2 displays max), and it still allows you to avoid the Enduro/Optimus hassle for external displays, when you are likely to be powering the laptop with the A/C adapter anyway. It also meant that vendor/OEM specific drivers needed to be used in some cases, though.
In Windows 10 and Linux, this can be nearly for free without any additional hardware. Linux calls this dma-buf, and Windows 10 basically requires it for any WDDM 2.x compliant drivers: GPU to GPU direct DMA mapping without CPU involvement (directly routed over PCI-E and not through the CPU memory controller when possible).
Basically, it means Optimus style and hardware-muxer style setups are obsolete, and display framebuffer and render framebuffer are entirely disconnected naturally; and also does not require any overt cross-architecture support (ex: Nvidia Optimus support in Nvidia's driver requires explicit Intel support; dma-buf solutions don't).
Sure, but then you still don't have V-Sync, which causes terrible and easily noticeable tearing on almost anything (at least in the case of Nvidia, it seems to be much better with AMD from my annecdotal testing). I doubt Apple would find tearing acceptable on their products, honestly.
That isn't related to dma-buf-type usage, though. That would be purely one of the drivers not signalling the end of the frame to the other (since all graphical APIs have suitable methods for this, and WDDM itself natively understands the concept).
Also, in Windows 10, the window manager's compositor (equivalent thereof) does all the event signaling fine; ie, you could easily have a situation where a game is rendered on one GPU, displayed on the other, and the display GPU + monitor has VESA Adaptive-sync enabled, and the game correctly triggers new sync windows as if it had been running on the same GPU.
Does this work in practice? Who knows, but its supposed to, especially with fixes that 1703 and 1709 added.
Is this intended/expected to work on desktop setups with discrete GPUs too? I believe Optimus et al were always laptop-only, but some desktop users might care about heat/power/fan noise.
It's expected to work. I've tested it on mine, Haswell i7's own iGPU, and a full sized GCN Radeon, with a game rendering on the Radeon but displaying on the iGPU.
Multi-monitor setup with 2 on the Radeon (normally all 3) and 1 on the iGPU, seemed to work fine. No real visible lag either, from what I could tell with limited testing.
Works on desktop just fine. For example Pentium G3258 integrated GPU forwarding fullhd over VGA rendered on ancient Nvidia 7900GS (old, but still 2-4x faster than integrated intel). Overhead was ~10-30% compared to direct monitor connection to nvidia card.
With a liitle more polishing over this combination, I can see the future: thinnest ultrabook with desktop grade performance and experience; packing a monstrous 4C8T top tier CPU and a high end graphics chipset in an ultrabook is definitely great for everyone. Would have been better if a Ryzen CPU is on the run for me.
This is going to be the future, no, It should have had been!
Agreed, I must have mistaken ultrabooks and notebooks. But rumor has it this hybrid chip will set its hold on 2018 MBPs. We might see how well it does first hand from Apple.
That's a Vega 8-compute units that's fully integrated with the CPU (and both the iGPU and CPU share a single memory controller to DDR4 RAM). The one talked about with the article is a Vega 24-compute units with extremely high-speed dedicated HBM memory. The performance difference would be night and day.
-------------
With that said, I purchased that laptop you're talking about. A brief overview:
1. Terrible screen: something like 60% sRGB coverage (maybe worse), and it shows with dithering and banding artifacts. Its also noticeably dimmer than other laptops.
2. Horrible HP Store -- I bought the laptop during Black-Friday, and HP shipped me a bricked laptop. Then they wouldn't replace the laptop because I got a "custom" SSD drive or some bullS!@#!@ excuse. My choices were to repurchase at a "non-sales" price, or get a refund. I opted for the refund and had to wait until the next sale to get a decent price. I suggest you buy from a reputable retailer who has a decent return policy, just in case the computer is broken or some other issue comes up. You do NOT want to be dealing with HP's online store customer service.
3. Minor default driver issues -- Fullscreen Youtube has occasional glitches and I've had the screen shutoff at least once randomly. Based on other owners: this seems to go away once you reinstall Windows and do some of HP's legwork for them. You should expect to reinstall Windows anyway because of the large number of bloatware (McAffee, Candy Crush, etc. etc.) that comes with the laptop.
4. AMD Driver Issues -- There's no Raven Ridge version of the latest "AMD Adrenaline" driver set. AMD's driver rollout for this laptop seems to be slow. I figure that as Raven Ridge APUs become more widespread, AMD's driver situation will improve.
5. Defaults to a Hard Drive -- Why HP? Just... why? But it has an M.2 slot, so its super easy for me to just stick a Samsung 960 Evo in there and fix that issue.
-----------
Aside from the poor quality screen (which stands out), its a good laptop overall. I think I'll be fine with it. Its got a numpad, the "360" 2-in-1 form factor definitely works, and the Stylus support is great.
So you can upgrade the laptop to your hearts content. You are going to be stuck with a dim, dithered, banding screen however. Its sufficient for office applications but holy crap the color irregularities are quite obvious. So this laptop's screen definitely can't be used for any serious art (which is a shame, since the Stylus support is so good).
Its definitely a $700ish laptop, with all of the issues that is typical in the $700 price range. (ie: Poor Screens are seriously common at this price range). You'll get very good price/performance and reasonable gaming on the go, since you get 4c / 8threads, a decent iGPU (its not as good as the MX150, but it can play the newest Doom at 720p and Rocket League at 57FPS / 720p).
I guess... the x360 as purchased isn't a very good laptop. But its actually very easy to turn it into a good laptop (aside from the screen). Reload Windows, fix up the drivers a bit, stick an M.2 SSD in there and its actually great.
Thanks a million! I've also had a mediocre experience with a previous generation x360. It is frustratingly close to being a very good machine in my opinion if just a few details were of higher quality. But I suppose it couldn't be quite as cheap that way.
38 comments
[ 2.5 ms ] story [ 82.4 ms ] threadLooking at a GTX 1030 this $70 card's performance to value improves greatly. It has the same performance as a AMD 480. https://www.newegg.com/Product/Product.aspx?Item=N82E1681413...
Are you confusing it with the AMD RX 460? The 460 is similar to a 1030.
An actual AMD RX 480/580 is far, far ahead of that. It's closer in performance to a GTX 1060.
Yeah I don’t think that’s true: http://gpu.userbenchmark.com/Compare/Nvidia-GT-1030-vs-AMD-R...
This opens up for integrated GPUs on laptops as well, which would be very interesting.
- The Quad-Core Mac Minis from 2011 & 2012 with 45W TDP. (e.g. i7-3720QM)
- The 2011 Mac Mini with dGPU with 35+26=61W TDP. (i7-2620M + Radeon HD 6630M)
Mind you, we can always dream that Apple will care about those of us who have wanted a Mac without a built-in display that fits between the Mac Mini and the Mac Pro.
GPUs use up tons of power, but 99% of the time they sit around and are "off". There's no need to turn on the GPU when browsing Facebook, and modern Win10 can keep the graphics card off.
Anyway, 100W TDP is perfectly acceptable for laptops, as long as this is a "worst case" TDP. Besides, gamers do want to game, and many are willing to accept bad battery life for the ability to run Witcher 3 or Tomb Raider on the go.
However if a new chassis is in the works and space is a premium it could lead to an iMac with chips similar to this but that would be disappointing as many are waiting for six core processors. For an Mini, wouldn't that TDP be too high?
Apple has an odd relationship with its desktop models. With the iMac Pro and "promise" of a new mini and Mac Pro there is a lot of hope for a resurgence however short it might be.
We all thought the whole reason of sticking a dGPU close to CPU was because of space, hence a Notebook type machine. But if anyone could enlighten me why I want a desktop with Integrated dGPU rather then CPU+GPU combo which is likely cheaper, and easier to cool down.
Not to mention you are stuck with KabyLake 4 Core instead of the newer Coffee Lake with 6 Core.
Routing HBM to the GPU using PCB traces is expensive, and in the low-to-mid end, saving money on PCB and cooling matters. If EMIB is going to be used anyway, why not have the GPU in there as well. With 1024-bit 4GB HBM on package, memory bandwidth won't be a problem, and the i7-8809G will likely be faster than a PS4, which means it'll be good enough for 1080p gaming.
KabyLake was probably chosen because this design was finalized before Coffee Lake. The next iteration will likely have more cores.
1. It will only make sense if Intel were actually manufacturing the GPU themselves as well. In this case they dont. And EMIB is an Intel specific technology, AMD doesn't use it when they package HBM with their dGPU.
2. AMD makes a profit before selling it to Intel, and Intel has to make a profits for integrating it with EMIB, i doubt the cost saving of using EMIB together with HBM, if that was true, out done the profits margin of both companies.
Can both be used at the same time?
Would make a nice use case for Vulkan-based games.
This means that you have to pay a penalty to use the Radeon to render, but send its framebuffer result back to the HD630 for displaying. This is also the case current Nvidia Optimus or AMD Enduro systems. Draw backs from this approach are frame tear / lack of proper V-Sync and added latency.
If this ends up being used by Apple however, it is likely that a hardware demux will be used to connect the display to both graphics chips, and switch it back and forth on demand. Other vendors don't do this, because it is more expensive, and passing the display from one GPU to another requires the pixel clock rate (among other details) to be shared between the different GPUs, which is not something that the vendor driver (Intel/Nvidia/AMD) stacks support.
Note that such a hardware demux is used in previous generations of MacBook Pros, and it is not publicly documented, which means very poor and/or very late support in Open Source operating systems such as Linux.
Some system builders (Lenovo and Asus have done this, probably others) wire the internal display of laptops to the Intel GPU, and the external HDMI/DP to the discrete graphics. In the past, this allowed you to power more displays (when Intel GPUs were restricted to 2 displays max), and it still allows you to avoid the Enduro/Optimus hassle for external displays, when you are likely to be powering the laptop with the A/C adapter anyway. It also meant that vendor/OEM specific drivers needed to be used in some cases, though.
Basically, it means Optimus style and hardware-muxer style setups are obsolete, and display framebuffer and render framebuffer are entirely disconnected naturally; and also does not require any overt cross-architecture support (ex: Nvidia Optimus support in Nvidia's driver requires explicit Intel support; dma-buf solutions don't).
Presumably, OSX also can leverage this.
Also, in Windows 10, the window manager's compositor (equivalent thereof) does all the event signaling fine; ie, you could easily have a situation where a game is rendered on one GPU, displayed on the other, and the display GPU + monitor has VESA Adaptive-sync enabled, and the game correctly triggers new sync windows as if it had been running on the same GPU.
Does this work in practice? Who knows, but its supposed to, especially with fixes that 1703 and 1709 added.
Multi-monitor setup with 2 on the Radeon (normally all 3) and 1 on the iGPU, seemed to work fine. No real visible lag either, from what I could tell with limited testing.
This is going to be the future, no, It should have had been!
I'm waiting for the reviews to come in before making any decisions, but it looks like quite a nice machine.
-------------
With that said, I purchased that laptop you're talking about. A brief overview:
1. Terrible screen: something like 60% sRGB coverage (maybe worse), and it shows with dithering and banding artifacts. Its also noticeably dimmer than other laptops.
2. Horrible HP Store -- I bought the laptop during Black-Friday, and HP shipped me a bricked laptop. Then they wouldn't replace the laptop because I got a "custom" SSD drive or some bullS!@#!@ excuse. My choices were to repurchase at a "non-sales" price, or get a refund. I opted for the refund and had to wait until the next sale to get a decent price. I suggest you buy from a reputable retailer who has a decent return policy, just in case the computer is broken or some other issue comes up. You do NOT want to be dealing with HP's online store customer service.
3. Minor default driver issues -- Fullscreen Youtube has occasional glitches and I've had the screen shutoff at least once randomly. Based on other owners: this seems to go away once you reinstall Windows and do some of HP's legwork for them. You should expect to reinstall Windows anyway because of the large number of bloatware (McAffee, Candy Crush, etc. etc.) that comes with the laptop.
4. AMD Driver Issues -- There's no Raven Ridge version of the latest "AMD Adrenaline" driver set. AMD's driver rollout for this laptop seems to be slow. I figure that as Raven Ridge APUs become more widespread, AMD's driver situation will improve.
5. Defaults to a Hard Drive -- Why HP? Just... why? But it has an M.2 slot, so its super easy for me to just stick a Samsung 960 Evo in there and fix that issue.
-----------
Aside from the poor quality screen (which stands out), its a good laptop overall. I think I'll be fine with it. Its got a numpad, the "360" 2-in-1 form factor definitely works, and the Stylus support is great.
The laptop is also fully expandable, with an M.2 slot, SATA slot, and easily accessible DDR4 DIMMs: http://h10032.www1.hp.com/ctg/Manual/c05819570
So you can upgrade the laptop to your hearts content. You are going to be stuck with a dim, dithered, banding screen however. Its sufficient for office applications but holy crap the color irregularities are quite obvious. So this laptop's screen definitely can't be used for any serious art (which is a shame, since the Stylus support is so good).
Its definitely a $700ish laptop, with all of the issues that is typical in the $700 price range. (ie: Poor Screens are seriously common at this price range). You'll get very good price/performance and reasonable gaming on the go, since you get 4c / 8threads, a decent iGPU (its not as good as the MX150, but it can play the newest Doom at 720p and Rocket League at 57FPS / 720p).
I guess... the x360 as purchased isn't a very good laptop. But its actually very easy to turn it into a good laptop (aside from the screen). Reload Windows, fix up the drivers a bit, stick an M.2 SSD in there and its actually great.