Kaby Lake-G's Vega Credentials Questioned: Rapid Packed Math Not Working

[PaulAlcorn]

Now Kaby Lake-G is back in the news with more speculation that its graphics are more akin to Polaris than Vega, so we've conducted testing to see if it supports a new Vega feature.

Kaby Lake-G's Vega Credentials Questioned: Rapid Packed Math Not Working : Read more

 

[DotNetMaster777]

Good article about Kaby Lake G !

 

[drkrffxx]

2+2 = 4-1 = 3 Quick Maffs

 

[justin.m.beauvais]

I think that what we are seeing is more of a custom design than a generation specific architecture. So, if we are trying to throw labels around, then the question does become "which Vega features define Vega and which Polaris features define Polaris?" because classically AMD's products have been more evolutionary than revolutionary. Vega is touted as a new architecture, but it is still GCN based, making it an evolution.

So, if RPM isn't enabled yet, which is a big feature of Vega, then it is going to act like Polaris, once it is enabled (if it is there) then we will see performance more characteristic of Vega. I think that there isn't as much of a difference from Polaris to Vega as AMD lets on. Vega itself seems like Polaris 2.0, which isn't a bad thing. Polaris is a good architecture, and making it better with Vega enhancements has proved to match higher end NVidia products, it just can't reach all the way up NVidia's product stack to tag the 1080Ti.

 

[redgarl]

Is it another attempt to throw dirt at AMD? It is a custom chip... like the PS4 and the XBox One X.

Could AMD use their Vega process for their custom chip to third parties? Of course! With Vega 11 introduced, I don't see why they would use a Polaris Process that is not currently applied in iGPU while Vega 11 is.

By comparing gaming performances with the 2400G 11 CU, you can see the scaling seems accurate with the Vega M 20 CU.

 

[ATInsider]

Vega, the king of the GPU.

 

[Lutfij]

:/ I'm a little speechless but then again I've seen this happen before with rebranding/rebadgings.

 

[Giroro]

What is even going on in Vega World where going back to the last-gen architecture is even an option? shouldn't AMD have released an entire line of cards by now instead of one mid-high range card and a slightly chopped-down version of the same card?

Getting an APU out is nice, but seriously where is Vega?

 

[ET3D]

Any way to test tiled rendering? That's another feature which should be enabled for Vega but doesn't exist in Polaris.

 

[justin.m.beauvais]

What is even going on in Vega World where going back to the last-gen architecture is even an option? shouldn't AMD have released an entire line of cards by now instead of one mid-high range card and a slightly chopped-down version of the same card?

Getting an APU out is nice, but seriously where is Vega?

If AMD released a Vega 18, Vega 36, and Vega 48, do you think we would get our hands on them before Navi launches? Aside from the doom and gloom, Polaris is more than capable at its current position (crypto mining prices aside). Vega doesn't have many features that would terribly benefit something with an RX 580 level of performance. Don't think of Vega as the replacement architecture for Polaris. Think of it as the top end offerings like the R9 Fury was. Vega is the Fury spiritual successor after all.

 

[salgado18]

Is it another attempt to throw dirt at AMD? It is a custom chip... like the PS4 and the XBox One X.

Could AMD use their Vega process for their custom chip to third parties? Of course! With Vega 11 introduced, I don't see why they would use a Polaris Process that is not currently applied in iGPU while Vega 11 is.

By comparing gaming performances with the 2400G 11 CU, you can see the scaling seems accurate with the Vega M 20 CU.

I believe this throws dirt at Intel. They wouldn't be fooled by AMD on such a deal, if AMD actually sold the wrong gpu, we'd hear it as a lawsuit, not as an investigation. Do you really think that dozens of engineers at Intel wouldn't find that out, and report to their superiors?

But Intel could have ordered a custom semi-Vega chip to use the branding, even knowing it's not a full Vega core. Maybe cut costs or compatibility or faster production, but it all goes on Intel.

 

[shahnewaz]

It's a semi-custom chip made for Intel, and Intel dictates what features go onto the chip and what it should be called. Just because they called it "Radeon Vega" doesn't mean it has to mirror AMD's own GPU chips.

 

[bit_user]

One of the most recent AAA titles, Far Cry 5, supports this functionality for example.

...checks benchmarks..
Wow! Vega 56 actually beats a GTX 1080.

The Vega NCU's ability to pack two 16-bit operations into a 32-bit register is new, allowing 64 shaders to perform up to 256 16-bit ops per clock (or even 512 eight-bit ops per clock).

Here's what 16-bit operations it actually supports. You only get 256 per cycle per CU, if it's a MAD or FMA and you're counting both the multiply and add (though it's a common metric to quote, when talking about peak compute throughput):

V_PK_MAD_I16
V_PK_MUL_LO_U16
V_PK_ADD_I16
V_PK_SUB_I16
V_PK_LSHLREV_B16
V_PK_LSHRREV_B16
V_PK_ASHRREV_I16
V_PK_MAX_I16
V_PK_MIN_I16
V_PK_MAD_U16
V_PK_ADD_U16
V_PK_SUB_U16
V_PK_MAX_U16
V_PK_MIN_U16
V_PK_FMA_F16
V_PK_ADD_F16
V_PK_MUL_F16
V_PK_MIN_F16
V_PK_MAX_F16

As for 8-bit, these are unsigned integers and it only supports unweighted average and SAD (Sum of Absolute Differences), which is useful for compression and not much else. Unfortunately for anyone doing inferencing with convolutional neural networks, it lacks Pascal's int8 dot product.

Source: https://developer.amd.com/wp-content/resources/Vega_Shader_ISA_28July2017.pdf

 

[alextheblue]

it's possible that Kaby Lake-G's Radeon graphics are a hybrid with enough Vega functionality to justify Vega marketing.

This is the most likely answer. KBL-G is likely "mostly-Vega".

The real question is... will Intel partner with AMD again for Navi, or will they hold off until their own solution is ready?

 

[bit_user]

it's possible that Kaby Lake-G's Radeon graphics are a hybrid with enough Vega functionality to justify Vega marketing.

This is the most likely answer. KBL-G is likely "mostly-Vega".

The HBM2 probably classifies it as Vega, in the minds of most consumers who are savvy enough to know the difference.

 

[bit_user]

Since Intel leaves its HD Graphics 630 block enabled on Kaby Lake-G, we also tested that. An outcome of 585.22 MPix/s with FP32 operations and 868.21 MPix/s with FP16 demonstrates Intel's own support for mixed data types.

They've had this capability since Broadwell (2014). I'm not aware of any CPUs with >= Gen8 HD Graphics on which it's disabled. So, yeah... if the HD Graphics block is working, you'll get double-rate fp16 from it. Check the fp64 ratio, while you're at it! Intel at least delivers the performance in one area...

https://software.intel.com/sites/default/files/Compute%20Architecture%20of%20Intel%20Processor%20Graphics%20Gen8.pdf

 

[bit_user]

In fact, application support remains limited. ... And underlying architectures may not mean much to gamers beyond their correlation to performance.

Exactly. fp16 support among games is quite rare. So, the HBM2 (which the market knows as a Vega-specific feature) is the main selling point.

That said, we do want to see companies stay consistent with their messaging so consumers can be sure of what they're getting. Transparency helps as well, and while we're not quite pointing fingers here, both AMD and Intel could have done a better job explaining just what is (and isn't) included in these intriguing new chips.

IMO, it was reasonable for them to call it Vega. You have to ask yourself would it have caused more or less market confusion to generate a new name or do something like call it Polaris+ with HBM2. In point of fact, they do consistently call it "RX Vega M".

However, I would agree that their communication with the press (and any datasheets) should've included a footnote saying it lacked RPM support.

 

[kareefh]

"Transparency helps as well, and while we're not quite pointing fingers here, both AMD and Intel could have done a better job explaining just what is (and isn't) included in these intriguing new chips."

I'm not quite pointing fingers here, but this would like squirrelly on Intel to be transparent here. This was outsourced to AMD for Intel with Intel as their customer. It would most likely be breach of contact for AMD to say anything about Intel's "order" publicly without their consent. It would be up to Intel what they want to make public about the design, similar to how sony and Microsoft would have all the say about their consoles despite AMD building the chip for them.

 

[bit_user]

I can't find anything on amd.com or radeon.com about "RX Vega M", but intel.com has these datasheets:

https://newsroom.intel.com/news/8th-gen-intel-core-radeon-rx-vega-m-graphics/
https://newsroom.intel.com/wp-content/uploads/sites/11/2018/01/8th-gen-radeon-rx-vega-m-product-overview.pdf

The second one quotes Peak SP performance. Nothing about half, fp16, or int16 performance.

IMO, it was a deliberate decision - probably based on trying to pack the most useful horsepower within a finite amount of die area. Also, if anyone did stress the 16-bit path, it might've tipped the balance of their power or thermal distribution to an extreme they didn't want to support.

 

[SoerenHedemand]

"Truthfully, it's all a bunch of branding buzzword bingo, anyway". Fact is, it's Vega because AMD says it's Vega; it's their brand.

 

[husker]

And here I thought what happens in Vega, stays in Vega.

 

[msroadkill612]

I could believe much of the business end of vega is polaris. The key to vega is architectural imo - introducing the gpu also to Infinity Fabric, which the intel G series dont use.

Strategically for amd, I a pleased they have not jeopardized key IP or or market advantage. Its just a custom mobile dgpu for which intel is the customer.

They are sick of nvidia behaving like primadonnas ("thats our role"), and ~uniquely AMD Hbm2, is a no brainer for mobile if you can afford it.

 

[msroadkill612]

It's vaguely vega.

 

[bit_user]

I could believe much of the business end of vega is polaris. The key to vega is architectural imo - introducing the gpu also to Infinity Fabric, which the intel G series dont use.

I would say there's a lot more to it than that:

https://radeon.com/_downloads/vega-whitepaper-11.6.17.pdf

Several of the features seem like they could hurt energy efficiency, however, perhaps to an extent not compensated for by its other efficiency improvements. That would lend credence to the idea that Intel might've preferred Polaris.

 

[zodiacfml]

Aside from the branding, Vega isn't much superior than Polaris. What matters are the numbers in the benchmarks.