AMD Introduces The Radeon Pro WX Series

[L_A_G]

chuckula I don't think the Radeon Pro Duo is just an Rx480, or even 2 of them glued together.... using specifications (you know, facts - not beliefs) - it's 128 compute units (the rx 480 has 36, for reference) - using (since it's polaris) the rx 480 (5.3 TFlops), that is in the neighborhood of up to 19Teraflops. It also has 8GB of HBM 4k memory... which is 1 terabyte/second memory bandwidth (rx 480 uses gddr5 @ 256 GB/s). For reference, the Titan X and Quadro P6000 has <= 12 Teraflop/s with <= 460 GB/sec memory bandwidth.

Nvidia's fastest GPU ever got 1 up'd the same day it was announced.

http://www.amd.com/en-us/products/gr...radeon-pro-duo
http://www.anandtech.com/show/10140/...radeon-pro-duo

Wake up sheeple!

The W7100 is pretty much a RX 480 except with 32 rather than 36 CUs. The 128 figure you're quoting is the number of texture units which is also down from the RX 480's 144.

Seriously, the Radeon Pro Duo you're referring to in those links was announced a year ago but didn't come out until a few months ago and even then it's essentially two Fury X's glued together.

[schmidtbag]

I wouldn't exactly call an Rx 480 with a higher price tag (which is what the WX 7100 is) a "beast" unless we are talking about power draw numbers.

I don't think you quite understand how workstation GPUs work; Nvidia is no different. Look at the average Quadro and you'll find it's basically some mid-range GeForce part with a few extra CUDA cores, a fancy new BIOS, and costs an extra $200+. Workstation graphics are designed for reliability and application-specific performance, where the companies team up to get the best possible results.

I don't know if this is still the case, but many (not all) of AMD's FirePro cards were literally just re-flashed Radeon chips. With a BIOS uploader, you could usually flash the FirePro BIOS to the Radeon equivalent and it would work fine.

[libv]

I don't think you quite understand how workstation GPUs work; Nvidia is no different. Look at the average Quadro and you'll find it's basically some mid-range GeForce part with a few extra CUDA cores, a fancy new BIOS, and costs an extra $200+. Workstation graphics are designed for reliability and application-specific performance, where the companies team up to get the best possible results.

I don't know if this is still the case, but many (not all) of AMD's FirePro cards were literally just re-flashed Radeon chips. With a BIOS uploader, you could usually flash the FirePro BIOS to the Radeon equivalent and it would work fine.

This was definitely true when we were doing RadeonHD. Downclocked, with a new pci-id, and several fixes to atomBIOS. Whether there were any chip level changes apart from burning in a new pci-id i do not know, i never noticed any but that does not mean that they were not there (or that they were there, known unknowns and such ;p). At RadeonHD, using real code, we fixed a modesetting issue or two by comparing the atomBIOS of the equivalent FireGL card with the consumer card, and updating our code (where we usually believed atomBIOS over the register documentation if we did not see a difference in our testing). FireGL, and the trickery in atomBIOS (supposed "scripts" which you couldn't upgrade yourself) is part of the reason as to why ATI was against RadeonHD and open source software. Our display driver could theoretically approach the display fidelity of the FireGL cards, and fix issues over many many generations.

[starshipeleven]

bridgman will the SSD versions do have build in Firmware on the flash for Debian style distros?

[bridgman]

For clarity, I said it was stupid in an intellectual sense, that it didn't actually improve anything and actually made microcode updates riskier and more difficult. Putting microcode into what is essentially an OEM-managed flash doesn't really fly unless OEMs are also onside with supporting updates specifically for Linux customers.

It's probably not stupid in a business sense though because some people observe demarcation between "microcode in a file" and "microcode in flash" despite the fact that the line was obviously drawn in the wrong place... the challenge is finding a business segment that is willing to spend more on making cards in exchange for greater sales into a market that most of them don't understand or even acknowledge.

Remember that the cards ship without flash, and that the M.2 slots are populated by end users. I have started looking into whether there is anything we could do with AMD-branded SSDs though.

[Taisserroots]

Why are you people talking about the Pro Duo? This is about the new Pro WX series, which IS polaris based. There's even a link to a radeon site about it in Michael's article

Some sites say fiji. Which is why most readers are unsure.

[starshipeleven]

Bridgman always pointed out hat "flash" and firmware stored on this flash on a graphic card is a stupid idea...

now AMD cards do have "flash" drive and they are still unable to store the firmware on a 1TB FLASH drive?

if there is 1TB of flash and no firmware stored in it this would be the most stupid think on earth.

And placing this firmware in the internal flash instead that in your hard drive changes things in what way?

It's still a closed firmware/microcode, flashing it on the device's onboard storage does not change that.

[duby229]

in theory it makes no difference. but in practice it is a huge different. you can use a vanilla Debian without the need of loading unfree by hand. it is just the second best after opensource firmware.

I don't understand at all. On gentoo I just emerge radeon-ucode. Done. That's literally it. Genkernel gives some pretty cool options on what to do with firmware that wouldn't exist if it had to be flashed on card. I think the current situation is far better than what you want.

[starshipeleven]

in theory it makes no difference. but in practice it is a huge different. you can use a vanilla Debian without the need of loading unfree by hand. it is just the second best after opensource firmware.

I'm still not seeing any kind of difference.

"not forcing me to add 1 line to Debian's sources.list, do an apt-get install linux-firmware-nonfree and lose 5 seconds of my precious time" is not "huge different".

If you want you can download (only the firmwares you need) manually from the links said here http://linuxfromscratch.org/blfs/vie.../firmware.html
just paste the blobs you need in the relevant folder. That way you will stil have as little as possible blobs.

I already had to do so for Debian as the firmware packages they ship is (how unexpectedly) outdated.

[bridgman]

Q is right on this one... most people using one of the "libre" distros (and some users of other distros) are going to flatly refuse to enable any non-free repo or install any non-free files/packages to get microcode, but will cheerfully use hardware which has a ton of microcode deployed through other means, eg ROM'ed on chip or loaded from external flash/ROM.

This will obviously change over time once it becomes clear that new guidelines are needed in order to run *any* current hardware, but we're not quite there yet.