Kaby-G Nuc Review

[wahdangun]

NUC mining? LOL!

Seriously though at $999 it isn't that attractive. Also, unless the HBM2 timings can be adjusted with modified BIOS, it'll hash only little better than RX 560 cards. This is up to Intel whether they'll allow modded BIOS to work.

If they made a quad-socket motherboard and got a socketable Core i7 8809G chips out, maybe a different story.

with lower TDP than RX 560 but with the performance of RX 570, then it will be Mining friendly, especially with the GPU price right now, so with lower electricity plus the cpu can be used for mining, the ROI will be better than RX 570 with current price.

 

[NTMBK]

with lower TDP than RX 560 but with the performance of RX 570, then it will be Mining friendly, especially with the GPU price right now, so with lower electricity plus the cpu can be used for mining, the ROI will be better than RX 570 with current price.

Mining rigs try to strap as many GPUs to a single low-end CPU as possible. A single midrange GPU permanently attached to a high end CPU like this is not a great balance for mining.

However, it does bode well for the discrete Vega Mobile chip coming soon. If AMD put that on a PCIe card I bet they can sell plenty to miners.

 

[The Stilt]

Dunno, shouldn't Vega have increased performance per clock from Fiji, with DSBR and L2 cache?

Yes, according to AMD it should have.
For some reason it never really materialized: https://www.gamersnexus.net/guides/2977-vega-fe-vs-fury-x-at-same-clocks-ipc

 

[itsmydamnation]

Yes, according to AMD it should have.
For some reason it never really materialized: https://www.gamersnexus.net/guides/2977-vega-fe-vs-fury-x-at-same-clocks-ipc

Thats pretty old, a year of driver development probably has helped Vega.

 

[The Stilt]

Thats pretty old, a year of driver development probably has helped Vega.

Here is slightly newer: https://www.hardocp.com/article/2017/09/12/radeon_rx_vega_64_vs_r9_fury_x_clock_for/1

Well below 5% difference on average.

 

[itsmydamnation]

Here is slightly newer: https://www.hardocp.com/article/2017/09/12/radeon_rx_vega_64_vs_r9_fury_x_clock_for/1

Well below 5% difference on average.

Still 7 months old.

If we assume that lots of games are ROP limited for AMD, then the difference is likely the binning rastizer all things being equal.

Vega is much better on memory bandwidth compared to fury ( needed for the extra clocks) and had one of its major advancements fail ( auto Geometry fast path). So while its "IPC" is the same, if you could clock Fury like you can Vega, Vega's IPC would be much higher.

 

[wahdangun]

Mining rigs try to strap as many GPUs to a single low-end CPU as possible. A single midrange GPU permanently attached to a high end CPU like this is not a great balance for mining.

However, it does bode well for the discrete Vega Mobile chip coming soon. If AMD put that on a PCIe card I bet they can sell plenty to miners.

But with gpu shortage, it's better to have one immediately rather than waiting.

Yeah dgpu edition of this will be great, and make me wonder how can Vega be this efficient

 

[Olikan]

Yes, according to AMD it should have.
For some reason it never really materialized: https://www.gamersnexus.net/guides/2977-vega-fe-vs-fury-x-at-same-clocks-ipc

I guess its the supposed geometry bottleneck... making vega's ROPs improvement almost irrelevant

 

[TempAcc99]

Vega in Raven Ridge APU can be Memory bound. Polaris - cannot. It has 256 Bit memory bus with GDDR5 memory interface.

RX480 / 580 both do have a memory bottleneck in certain situation. The design, especially the RX580 is driven to the limits. Even with the RX480 memory OC only led to significant gains. Same can be said for vega 56. OCing it to vega 64 levels and they are almost identical meaning the extra shaders are pretty much useless for gaming.

Dunno, shouldn't Vega have increased performance per clock from Fiji, with DSBR and L2 cache?

It should have but it's ROP and memory bandwidth limited. I cant' find a link but I remember the theory was that GCN is limited that it can't have more than 64 ROPs which would have clearly helped vega 64 (see above, with same clocks vega 56 and 64 are almost identical). Also with vega56 a memory OC alone nets you a performance increase meaning it's also bandwidth limited. AFAIK hbm2 should have release with higher clocks so AMDs calculations were right but then the manufacturing turned out to be much harder and clocks could not be met and hence vega 56+64 end up with a memory bottleneck. Vega 56 was actually pretty good if you tune it (OC + undervolt). Was because mining craze...

 

[IntelUser2000]

Where does this estimate come from? In terms of bandwidth and compute power it's quite close to RX 570, with lower memory latency. Depending on mining algo it may end up trading blows with it.

They are similar stock, but RX 570 can be pushed much further by custom BIOS timings. We're talking 40-50%.

If you cannot adjust the timings on this part, it'll be equal to a souped up RX 560 part. Not to mention you need to buy a $999 device to do so.

 

[SirDinadan]

As now several comment is about how the RX 580 is bandwidth starved to some degree, it begs the question: why AMD went crazy with the gfx clocks on the 580 instead of opting for GDDR5X memory?

 

[TempAcc99]

As now several comment is about how the RX 580 is bandwidth starved to some degree, it begs the question: why AMD went crazy with the gfx clocks on the 580 instead of opting for GDDR5X memory?

Because GDDR5x is actually quiet different from GDDR5 meaning they would have needed to make a new memory controller instead of just changing default specs and the product number.