AMD Vega Hardware Reviews

[w0lfram]

I have been interested in AMD's "Infinity Fabric" for quite some time.

I truly think the media and press have not been covering this, or AMD is being reserved and coy. Because AMD's fabric is in Ryzen, Threadripper, EPYC, RavenRidge & Vega. And their stock keeps going up. I don't ever hear anything bad about about Infinity Fabric, just people taking stabs at what it can't do.

I think it is more advanced technology than many are willing to admit. I am going to let this presentation slide sink in:

That is here and now folks. What do you think AMD is going to do over the next 6 months with that ?
The protocols exists, it just needs to mature.

 

[xEx]

If the process in GF were the problem ryzen would have being a disaster and it actually manage to beat Intel...So there is no responsible for Vega performance but AMDs engineering team and thats it. Lack of money, talent or leadership it doesn't metter, the product is not inferior to its competition. AMD needs to try harder next time.

 

[3dilettante]

I truly think the media and press have not been covering this, or AMD is being reserved and coy.

It's been covered repeatedly and brought up repeatedly by AMD.

I don't ever hear anything bad about about Infinity Fabric, just people taking stabs at what it can't do.

As long as they're good stabs.

I think it is more advanced technology than many are willing to admit. I am going to let this presentation slide sink in:

That is here and now folks. What do you think AMD is going to do over the next 6 months with that ?
The protocols exists, it just needs to mature.

Could you give some detail as to what this means for Vega 10? It's not clear there's a problem with it at this point, and as long as it moves bits from A to B without issue, that's mostly what Vega asks of it.

 

[Deleted member 13524]

If the process in GF were the problem ryzen would have being a disaster and it actually manage to beat Intel...

Do you happen to know how Intel's 14FF compares to TSMC's 16FF+?

 

[Moloch]

LOL, I can relate to that.

I have found some clips on YouTube that my system (6core/12 thread @ 4Ghz) couldn't play back smoothly in a browser that didn't support hardware acceleration, but played just fine on one that could. I'll see if I can dig one up when I get home. 4K+HDR videos can make my system struggle as well since VP9 profile 2 isn't accelerated on the 1060.

Even when playback performance is OK, it's still nice to not have to spend that many CPU cycles on something that a dedicated decoder can do with a couple of watts and without breaking a sweat.

What's interesting is that while AMD spent zero engineering resources on raising their video decode support to be on par with Nvidia and Intel, they did spend some on enabling encode/decode virtualization support, something that is super important in the consumer space.

I browsed some more 8k YouTube clips and did manage to find some that wouldn't play smoothly, my cpu is a 6700k@4.6ghz.
Many of Amd's recent decisions make me scratch my head.

 

[w0lfram]

It's been covered repeatedly and brought up repeatedly by AMD.
As long as they're good stabs.

Could you give some detail as to what this means for Vega 10? It's not clear there's a problem with it at this point, and as long as it moves bits from A to B without issue, that's mostly what Vega asks of it.

I think AMD's direction with Vega, looks very clear to me.

I don't think it is any secret, AMD are going to go the route of Zen. And leverage the Vega die as many ways as possible, using their infinity fabric (ie: Threadripper/EPYC) . There is ZERO reason to assume otherwise, as everything we know leads directly to this.



As for hyperbolic speculation?
For all we know, AMD is getting dirt cheap prices for Vega wafers as a "thank you" for keeping Global Foundries afloat, and working tightly with AMD their yields are 80%, like Zen's. So AMD could possibly be making larger margin on Vega, & will eventually be able to sell off downclocked cut-down Vega 48's for $199 this holiday. (& still get a voucher for $200 off a monitor... lol)

 

[3dilettante]

If the process in GF were the problem ryzen would have being a disaster and it actually manage to beat Intel...So there is no responsible for Vega performance but AMDs engineering team and thats it. Lack of money, talent or leadership it doesn't metter, the product is not inferior to its competition. AMD needs to try harder next time.

GF's process likely has a primary role in why Zen lags in clock speed despite what appears to be a roughly equal number of pipeline stages to its competition. Other quirks like the size of the CCX and other implementation choices were likely influenced by the limitations on complexity and performance of the process they were implemented on.

That Zen managed to hit a decent optimization point as a highly clocked CPU doesn't mean a GPU would be as successful. AMD's description of process tweaks for its prior APUs is evidence that it could be the opposite. It would make sense to tilt the tables in favor of the highly-priced CPU, if AMD had to choose how its priorities would go.

That's not to say that the process is the deciding factor for Vega, just that it is possible for a CPU's situation to be significantly different on the same node.

 

[Geeforcer]

I think AMD's direction with Vega, looks very clear to me.

I don't think it is any secret, AMD are going to go the route of Zen. And leverage the Vega die as many ways as possible, using their infinity fabric (ie: Threadripper/EPYC) . There is ZERO reason to assume otherwise, as everything we know leads directly to this.



As for hyperbolic speculation?
For all we know, AMD is getting dirt cheap prices for Vega wafers as a "thank you" for keeping Global Foundries afloat, and working tightly with AMD their yields are 80%, like Zen's. So AMD could possibly be making larger margin on Vega, & will eventually be able to sell off downclocked cut-down Vega 48's for $199 this holiday. (& still get a voucher for $200 off a monitor... lol)

I am sorry, what???

 

[mrcorbo]

I browsed some more 8k YouTube clips and did manage to find some that wouldn't play smoothly, my cpu is a 6700k@4.6ghz.
Many of Amd's recent decisions make me scratch my head.

Vega was the result of AMD trying to create an architecture that would scale from consumer through Professional up to HPC. If you assume every time there was a decision to be made in development that could either benefit the latter markets or the former, they chose the latter every time some of those choices make more sense. They spent a lot of engineering resources and die space on things that aren't really impactful for a consumer GPU in 2017. Some of those features will be brought forward into Navi and beyond and at that time they may have a greater impact (HBCC, specifically, is going to be a necessity for Navi to work as concieved), but right now? They just make for bigger, more expensive chips that still have to be clocked past their peak power efficiency to be even somewhat competitive.

 

[3dilettante]

I think AMD's direction with Vega, looks very clear to me.

I don't think it is any secret, AMD are going to go the route of Zen. And leverage the Vega die as many ways as possible, using their infinity fabric (ie: Threadripper/EPYC) . There is ZERO reason to assume otherwise, as everything we know leads directly to this.

The wafer shots are not detailed enough to show, but does this mean you believe there are PHY for off-die links in Vega 10?

 

[swaaye]

Vega was the result of AMD trying to create an architecture that would scale from consumer through Professional up to HPC. If you assume every time there was a decision to be made in development that could either benefit the latter markets or the former, they chose the latter every time some of those choices make more sense. They spent a lot of engineering resources and die space on things that aren't really impactful for a consumer GPU in 2017. Some of those features will be brought forward into Navi and beyond and at that time they may have a greater impact (HBCC, specifically, is going to be a necessity for Navi to work as concieved), but right now? They just make for bigger, more expensive chips that still have to be clocked past their peak power efficiency to be even somewhat competitive.

Big Fermi, Big Maxwell and are the same story though but they didn't embarrass themselves.

We dont know what GP100 is like however. NV's HBM2 chip that we are unworthy of as consumers. Sounds a weee bit out of Vega's class from the spec sheet.

 

[Esrever]

Big Fermi, Big Maxwell and are the same story though but they don't embarrass themselves. I suppose we dont know what GP100 is like. NV's HBM2 chip that we are unworthy of as consumers. Heh.

Big Firmi definitely did embarrass itself.

 

[swaaye]

Big Firmi definitely did embarrass itself.

Oh really? It beat AMD's products across the board. Except on power consumption, I suppose. Solid HPC product too.

 

[Esrever]

Oh really? It beat AMD's products across the board. Except on power consumption, I suppose. Solid HPC product too.

So the memes are a compliment?

 

[swaaye]

So the memes are a compliment?

Thermi?

Interwebs are filled with idiots so I don't really read alot into it. 480 was hot and stuff but still beat 5870. 580 was out of 6970's league. And AMD was a joke in HPC.

 

[mrcorbo]

Big Fermi, Big Maxwell and are the same story though but they didn't embarrass themselves.

We dont know what GP100 is like however. NV's HBM2 chip that we are unworthy of as consumers. Sounds a weee bit out of Vega's class from the spec sheet.

AMD aren't Nvidia.

 

[Esrever]

Thermi?

Interwebs are filled with idiots so I don't really read alot into it. 480 was hot and stuff but still beat 5870. 580 was out of 6970's league. And AMD was a joke in HPC.

You can say that if you like but people making fun of Firmi for the last 10 years has gotta indicate it was pretty embarrassing.

 

[swaaye]

You can say that if you like but people making fun of Firmi for the last 10 years has gotta indicate it was pretty embarrassing.

No, internet sentiment doesn't necessarily indicate anything more than mob/bandwagon nonsense. And it's called "Fermi". Go look over the reviews and read about its Quadro and Tesla incarnations and judge for yourself.

 

[BRiT]

*AHEM* This is the AMD Vega Hardware Review thread. Please keep all other NOISE out of it.

 

[rcf]

We dont know what GP100 is like however. NV's HBM2 chip that we are unworthy of as consumers. Sounds a weee bit out of Vega's class from the spec sheet.

Yes, we need some nice "Quadro GP100 vs WX 9100" benchmarks. That would be very interesting (and expensive).
The few Quadro GP100 benchmarks that I've seen show GP100 having a similar advantage over GP102 as GP102 have over GP104.