AnandTech: The Intel NUC8i7HVK (Hades Canyon) Review: Kaby Lake-G Benchmarked

[raghu78]

I have never seen you work so hard to tilt something in favor of Intel.

But you are misrepresenting the content of those links to make your case.

You keep saying it is faster than desktop GTX 1050 Ti. But the link (notebookcheck) you include says:

"Gaming Performance - More like a GTX 1050 Ti"

Also they are using a laptop based GTX 1050 Ti.

For an actual Desktop GTX 1050 Ti, see the link I showed further up the page, which uses an actual desktop GTX 1050 Ti and they are clearly the same performance, and PCPer says:

"Compared to our desktop GPUs, the RX Vega M GH graphics fall in line with the performance of a GTX 1050 Ti."

I'm not sure what you are having such a hard time accepting that Intel Kaby-G has GPU performance equivalent to a desktop GTX 1050 Ti.

I am also not sure what you are having a problem accepting that Vega M GH is in between 1050ti desktop and 1060 Max Q. Read the second review from digital trends where they compare actual gaming notebooks with 1060 MaxQ against Rx Vega M GH. Very close.

 

[IntelUser2000]

I am also not sure what you are having a problem accepting that Vega M GH is in between 1050ti desktop and 1060 Max Q. Read the second review from digital trends where they compare actual gaming notebooks with 1060 MaxQ against Rx Vega M GH. Very close.

Drivers can probably fix that. Even though its based on Vega, it might need specific optimizations. Plus, they can work the power sharing feature to work better.

 

[PeterScott]

I am also not sure what you are having a problem accepting that Vega M GH is in between 1050ti desktop and 1060 Max Q. Read the second review from digital trends where they compare actual gaming notebooks with 1060 MaxQ against Rx Vega M GH. Very close.

Because the reviews that actually test a Desktop 1050 Ti say and show it's essentially identical. That is the reality.

For some reason you keep bringing the laptops into Desktop argument when they don't apply. GPUs in Laptops perform worse, and they vary a lot because the cooling varies a lot.

Find some actual reviews where Hades Canyon is outperforming a Desktop 1050 Ti, to back your unsubstantiated claim.

So far, my claim is backed by actual testing, yours is some weird assumption based on laptops.

 

[french toast]

$600 is too low. Even if you assume $300 for CPU and $300 for GPU, you still don't count for the rest of the system.

No I don't think it is, you can't look at it like that (even though intel stupidly does).
It is intel who manufacturers the APU (and pays AMD a small fee) as well as intel who designs and produces the hardware, so you can figure that all in to the BOM as the manufacturing costs...instead of buying them from separate companies like other competitors do...intel had a massive opportunity to crush the market and have a hit mass market device.
This has no expensive ram, no high end SSD, No microsoft payments for OS, it has decent CPU performance (probably 45w 4/8 kabylake mobile performance) and desktop 1050TI GPU performance on the same package, fancy cooling and led system wouldn't cost that much when designed and produced internally...neither the mobo which intel probably make?..250w PSU? Not much.
5-600$ seems about reasonable for top SKU (4-500 for the lower).
Now granted the competition in this mini barebones pc market is also way overpriced even if they offer better value (worst design imo)...but they have to buy from intel at their prices!..intel pays internal manufacturing costs.
This 600$ price point would allow for hundreds of dollars of profit as well as a large amount of sales, essentially an Uber powerful, far more useful Nvidia shield box/tiny gaming PC = huge success.

Instead as always they got greedy so this will sell only in small quantities, extremely high margins yes but low number of units sold = missed opportunity imo.

Yeah, definitely overpriced. Another Anandtech story:
https://www.anandtech.com/show/12594/xiaomi-announces-mi-gaming-laptop

A whole laptop, with CPU/GTX 1060, RAM, SSD, Display, HDD for under $1000.

We need Xiaomi to build a cool SFF with GTX 1060.

I have never seen you work so hard to tilt something in favor of Intel.

But you are misrepresenting the content of those links to make your case.

You keep saying it is faster than desktop GTX 1050 Ti. But the link (notebookcheck) you include says:

"Gaming Performance - More like a GTX 1050 Ti"

Also they are using a laptop based GTX 1050 Ti.

For an actual Desktop GTX 1050 Ti, see the link I showed further up the page, which uses an actual desktop GTX 1050 Ti and they are clearly the same performance, and PCPer says:

"Compared to our desktop GPUs, the RX Vega M GH graphics fall in line with the performance of a GTX 1050 Ti."

I'm not sure what you are having such a hard time accepting that Intel Kaby-G has GPU performance equivalent to a desktop GTX 1050 Ti.

Exactly, what are you getting here? 45w kabylake mobile CPU, desktop 1050ti, fancy cooling,design and Led, some clever fitting and airflow..which does demand a design premium, but look at the bare components which are all internal intel components = Cheap (compared to OEM).. intel had a great opportunity missed.
600$ for an outside company to manufacture and sell? Doable, for intel itself?? Very profitable.

 

[IntelUser2000]

Now granted the competition in this mini barebones pc market is also way overpriced even if they offer better value (worst design imo)...but they have to buy from intel at their prices!..intel pays internal manufacturing costs.

Yes, that's what you do as a business. Try to make a company that prices a product solely based on raw materials and tell me you won't get bankrupt in 5 years. Pricing is based on external factors, like what the market is willing to accept.

Intel does not sell directly either. You have to buy it from places like Newegg.

There are lots of people buying PCs that can't ever be bothered to buy higher end parts, and they'll go for the cheapest. The rest? Manufacturers have to jump at every opportunity to try to upsell and create a checkbox for those that are inclined. $400 Laptops? Sure, heck a lot of people buy them, but I doubt anyone is making any desirable amounts of money from them. You use higher end products, and upselling to do so, which is why configurations show high cost adder for higher tier components, more than MSRP difference of the said component.

 

[BigDaveX]

Respectable enough performance-wise. There might be better options in terms of price-performance, but this thing probably has its niche.

As an aside, comparing Hades Canyon to the old Skull Canyon box, it's clear what a monumental waste of resources the more powerful IGPs that Intel were putting in their high end mobile chips between Haswell and Skylake were. All it took was AMD putting Ryzen cores in their APUs to render Intel's entire IGP line-up as laughably slow and irrelevant as it was in the pre-Bulldozer era.

 

[epsilon84]

A bit below what I expected graphics wise as this seems closer to a 1050 Ti than a 1060, whereas all the hype was that it would provide 1060 levels of performance - another product where the hype exceeds the reality.

Then again, it performs about as well as it should considering the 24CUs. A GTX 1060 is only about 30% slower than a Vega 56 so you would need a hypotethical 'Vega 40' or thereabouts to perform as well as a GTX 1060.

 

[Bouowmx]

you would need a hypotethical 'Vega 40' or thereabouts to perform as well as a GTX 1060.

Or has Vega performance per FLOPS gone down from Polaris 10 (36 CUs/2304 cores)?

 

[PeterScott]

A bit below what I expected graphics wise as this seems closer to a 1050 Ti than a 1060, whereas all the hype was that it would provide 1060 levels of performance - another product where the hype exceeds the reality.

Then again, it performs about as well as it should considering the 24CUs. A GTX 1060 is only about 30% slower than a Vega 56 so you would need a hypotethical 'Vega 40' or thereabouts to perform as well as a GTX 1060.

36 CUs like in the RX 480/580 which is what actually competes with the GTX 1060.

The marketing was based on 3Dmark numbers, which are pretty much pointless, and Max Q 1060, which is slower than the stander laptop 1060, which is slower itself than desktop 1060.

In actual games, it's closer to a desktop 1050 ti. Which would be expected if you ignored the marketing (AKA exaggeration).

 

[PeterScott]

It really seems there is room here for AMD to release their rumored 28 CU Mobile part, paired with a Ryzen CPU that outperforms Hades Canyon, undercuts it in price, while still earning a nice profit margin.

 

[raghu78]

A bit below what I expected graphics wise as this seems closer to a 1050 Ti than a 1060, whereas all the hype was that it would provide 1060 levels of performance - another product where the hype exceeds the reality.

Then again, it performs about as well as it should considering the 24CUs. A GTX 1060 is only about 30% slower than a Vega 56 so you would need a hypotethical 'Vega 40' or thereabouts to perform as well as a GTX 1060.

http://www.legitreviews.com/intel-hades-canyon-nuc8i7hvk-nuc-review-radeon-rx-vega-m-gpu_204024

The Hades Canyon NUC measures in at 221mm x 142mm x 39mm and has a volume of 1.2L.

Try cramming a GTX 1060 desktop level perf into a 1.2 litre chassis. Its impressive that AMD got the highly inefficient Vega to deliver this level of perf in a NUC. The perf is just slightly behind GTX 1060 Max Q and I would say thats an achievement for AMD given the form factor and portability benefits of the NUC. For people who want to compare with a GTX 1060 desktop its an irrelevant comparison as the GTX 1060 desktop is not going to be available in a form factor as compact as Hades Canyon. GTX 1060 Max Q is >=30% slower than GTX 1060 desktop and just 10-20% faster than the Rx Vega M GH. The GTX 1060 notebook lies mid way in between GTX 1060 and 1060 MaxQ.

 

[raghu78]

It really seems there is room here for AMD to release their rumored 28 CU Mobile part, paired with a Ryzen CPU that outperforms Hades Canyon, undercuts it in price, while still earning a nice profit margin.

https://zhuanlan.zhihu.com/p/34510590

I think the AMD Fenghuang Raven part could be a single die APU built on GF 12LP . I remember Mark Papermaster saying that all future client products from AMD in 2018 will be built at 12LP. Raven Ridge was a late 2017 product and probably the last 14LPP product. If we are talking of a single die APU I think AMD could deliver GTX 1060 notebook or MaxQ performance depending on how 12LP voltage / freq curve is.

 

[PeterScott]

http://www.legitreviews.com/intel-hades-canyon-nuc8i7hvk-nuc-review-radeon-rx-vega-m-gpu_204024

The Hades Canyon NUC measures in at 221mm x 142mm x 39mm and has a volume of 1.2L.

Try cramming a GTX 1060 desktop level perf into a 1.2 litre chassis. Its impressive that AMD got the highly inefficient Vega to deliver this level of perf in a NUC. The perf is just slightly behind GTX 1060 Max Q and I would say thats an achievement for AMD given the form factor and portability benefits of the NUC. For people who want to compare with a GTX 1060 desktop its an irrelevant comparison as the GTX 1060 desktop is not going to be available in a form factor as compact as Hades Canyon. GTX 1060 Max Q is >=30% slower than GTX 1060 desktop and just 10-20% faster than the Rx Vega M GH. The GTX 1060 notebook lies mid way in between GTX 1060 and 1060 MaxQ.

It's Intel that did the impressive work here, getting a 200 watt cooling solution into that 1.2 liter chassis. Power usage is similar to discrete CPU + GTX 1050 Ti, that it performs like.

 

[Gikaseixas]

Impressive performance for such a small print device. Kudos to Intel and AMD

 

[IRobot23]

How did Anantech get 100W more than other?
https://www.anandtech.com/show/1257...yon-nuc8i7hvk-review-kaby-lakeg-benchmarked/8

First at least you can tell someone that under furmark/aida sst AMD GPu didn't throttle (frequency) which insanely good.
Stating 60W for GPU part is normal and probably 20W for HBM and 0.82 efficiency ~ 100W... hard to tell since they do so bad testing.

Example: this image showed me that probably HW info reporting is wrong and GPu power is around 110W (+20-25 for system and 0,82-0,85 for efficiency).

Red CPU die is under violet GPU?

Then this one : Which is how I got 110W under furmark.

Something just doesn't feel right in these testing. It's either frequency or something,... Anyway, its really hard to make conclusion out of it. From the graph (power at wall) you can say that it actually downclocked few times or something like it.

Gaming benchmark?
https://www.anandtech.com/show/1257...yon-nuc8i7hvk-review-kaby-lakeg-benchmarked/4
Are you guys even serious?
Why would you do GTA V benchmark, BF1 or The Witcher 3... no lets take Bad company 2 and GTA IV.... put fortnite in GTA V , BF1 and top ten on steam that are more GPU intensive. Who even plays sleeping dogs these days... and and The Talos of principle? Yeah, I heard they have over 7 million accounts...

Can do frequency in gaming @ thermal throttling, gaming power consumption...? Nah that is not important...

My conclusion? Well I need to take some more time to check some other review that will do testing that matters.

 

[Glo.]

Stating 60W for GPU part is normal and probably 20W for HBM and 0.82 efficiency ~ 100W... hard to tell since they do so bad testing.

HBM2 single stack draws 1.2W of power @1.2V/800 MHz. That is whole point of HBM2 in the first place. Drawing as low power as possible.

Full 256 GB/s, 1000 MHz HBM2 stack would draw max. 2W of power.

GDDR5 memory, single chip, uses 4W of power @1.5V/8000-9000 MHz.

Whole memory subsystem for RX 480 draws 32W of power under load.

Remember guys. GPUs overclock and declock themselves depending on the load, to save power.

 

[ElFenix]

https://www.anandtech.com/show/12572/the-intel-hades-canyon-nuc8i7hvk-review-kaby-lakeg-benchmarked

It's too bad AT doesn't create forum topics for its reviews like other sites to foster discussions, and instead has a kludgy front page only comment system.

used to back in the fusetalk days, but that changed over 10 years ago when comments moved to the front page. iirc there was some discussion when the board moved to vbulletin to integrate forum comments and front page comments, but i don't know if that even made it to an alpha system.

 

[Glo.]

Want to know what is strange? If all of those reviews are correct, RX Vega M is slightly slower, or on par with Radeon Pro WX5100.

Radeon Pro WX5100 - 1792 GCN4 cores, 8 GB GDDR5 6000 MHz/160 GB/s with 1086 MHz core clock.
Radeon RX Vega M 1536 GCN5 cores, 4 GB HBM2 204 GB/s with 1190 MHz core clock.

The difference? RX Vega M has 64 ROPs, and WX5100 - 32.

The only place where Vega actually beats the Radeon Pro is in efficiency. 55W vs 75W.

 

[Olikan]

The only place where Vega actually beats the Radeon Pro is in efficiency. 55W vs 75W.

And mostly comes from HBM

 

[french toast]

Yes, that's what you do as a business. Try to make a company that prices a product solely based on raw materials and tell me you won't get bankrupt in 5 years. Pricing is based on external factors, like what the market is willing to accept.

Intel does not sell directly either. You have to buy it from places like Newegg.

There are lots of people buying PCs that can't ever be bothered to buy higher end parts, and they'll go for the cheapest. The rest? Manufacturers have to jump at every opportunity to try to upsell and create a checkbox for those that are inclined. $400 Laptops? Sure, heck a lot of people buy them, but I doubt anyone is making any desirable amounts of money from them. You use higher end products, and upselling to do so, which is why configurations show high cost adder for higher tier components, more than MSRP difference of the said component.

The idea of a business is to A) make money B) increase brand value C) Both of which should offer value to investors.

This bloated big company way of operating sure does work in most cases, but sometimes it's a bad fit.
The idea is to make a desirable product and return the most profit, there is two main ways of doing this...go ultra high end in either performance or design and sell high = fewer sales but high margins.
Or lower BOM, lower margin but if positioned right very high sales and very high profit.

Haydes canyon offers too little of the former to generate enough sales, the market is bloated, haydes canyon just offers more of the same, just some.nice design, it doesn't disrupt the market in the way it could.
They could make the box for probably 300$..that's being generous...that leaves plenty of room for retail markup and tonnes of profit if sold at 600$, people would grab these things off of the shelves like gold dust as it would disrupt the market.

That would offer more profit on the whole than the crazy high margins and low sales.
Of course you also have the factor of intel sells the other players the CPUs in that bloated market...if they out sold them they would lose revenue from other OEMs.
Mmm, you might be right in the end but not for the reasons you thought.

Edit; what's the point in designing an innovative APU if you are not going to take advantage of all of its beneficiary's? Namely space saving, fewer components and cost?...haydes canyon is too expensive to sell well, but I suppose as long as OEMs can buy the chip to make cheaper and smaller systems than they could with descrete parts then it could be a winner.
I have my doubts but we will see, as a technology concept that you can buy haydes canyon looks good.

 

[IRobot23]

HBM2 single stack draws 1.2W of power @1.2V/800 MHz. That is whole point of HBM2 in the first place. Drawing as low power as possible.

Full 256 GB/s, 1000 MHz HBM2 stack would draw max. 2W of power.

GDDR5 memory, single chip, uses 4W of power @1.5V/8000-9000 MHz.

Whole memory subsystem for RX 480 draws 32W of power under load.

Remember guys. GPUs overclock and declock themselves depending on the load, to save power.

Really only 2W? I am sure I saw(heard) different things. I thought it is like 40% less power for same bandwidth.

 

[jpiniero]

Edit; what's the point in designing an innovative APU if you are not going to take advantage of all of its beneficiary's? Namely space saving, fewer components and cost?...haydes canyon is too expensive to sell well, but I suppose as long as OEMs can buy the chip to make cheaper and smaller systems than they could with descrete parts then it could be a winner.
I have my doubts but we will see, as a technology concept that you can buy haydes canyon looks good.

The sales are going to come from the laptop models, not Hades Canyon. I still think it's more of an EMIB R&D project more than anything so we'll see if the laptop models get any traction.

 

[whm1974]

Sure is pricey for a bare bones system. I sure I can build a MiniITX better and cheaper then what a fully configured Hades Canyon will cost. GPU prices not withstanding.

 

[Qwertilot]

The sales are going to come from the laptop models, not Hades Canyon. I still think it's more of an EMIB R&D project more than anything so we'll see if the laptop models get any traction.

Not sure why there should be many of those? The power draw/efficiency is a non trivial problem for a notebook.

The other obvious issue is that the 1060 is nearly 2 years old and due for replacement quite soon.

 

[Glo.]

Really only 2W? I am sure I saw(heard) different things. I thought it is like 40% less power for same bandwidth.

All of memory chips use very low amounts of power. Difference in power consumption between 16 GB kit of DDR4 2400 MHz and 3200 MHz is 10W, and all is due to higher clocks, and higher voltage.

If single HBM2 memory chip would consume 20W - that would be completely ridiculous number. What is the point of it, if 4 stacks of it would use 80W of power, while delivering 1 TB/s bandwidth?

Whole point of HBM was massive power consumption saving.