Funny enough, the virtually-identical RX 550 only has x8 lanes enabled (still a x16 connector, though). Whether it's possible to contrive a scenario where PCIe 2.0 x8 would have a measurable impact, I agree that it would be GPU-bottlenecked in real world use cases.
However, one small consideration would be that increasing bus bandwidth will typically reduce latency, even if slightly. This might help with VR (on a more powerful card, that can actually *do* VR), where a couple miliseconds' stutter can potentially have nauseating consequences.
I think the point of the **no ECC** is that this is a 'high-yield' entry-level 50w pro card on a mature process. It is effectively Polaris gravy with nearly 600 dies on a 300mm wafer that includes 2 dozen+ Lexa variants.
They don't have any other silicon for a card like this. Vega wasn't scaled down this far and Navi hasn't even released yet. I think this is Polaris "12" aka Lexa, it's like 100mm2. it's adorable.
Polaris hasn't been replaced in this Performance tier yet.
That's not a 16x GPU if the image above is accurate. That probably is a non-issue from a performance standpoint given the overall compute power of the card, but it is noteworthy.
For actual specs rather than a seemingly broken source link, here's the URL to AMD's site:
This is not a games card - it is a low end CAD/CAM card - the important thing for the potential buyers is the full professional CAD/CAM support. 16 lane PCIe and PCIe 4.0 are completely unimportant for this card. Much higher performance CAD/CAM cards are available but they cost far more than $199 - this card is expected to compete with the Nvidia P1000 not the RTX 8000!!
So the doubling of the price (the 4GB variant of Radeon RX 550 this card is based on normally sells for $95 to $105) is due to CAD/CAM certification and professional drivers. Well, this is similar to the "professional price overhead" for much costlier cards..
Doubling the price due to high-quality electronic components that gonna last forever, rather than bottom-of-the-bucket noname discounted stuff. Professional GPUs have always been built like that, and it’s quiet expensive to do especially at their low volumes.
No no, you're not a 'real; gamer unless your card has configurable RGB lighting and a graphics of some sort of "battle-y" image pasted all over it. Can't be a 'real' gamer if all you are concerned about is the performance and not the fancy looks.
Is the performance really enough for anyone who needs a professional card though?
AMD might have been a little to eager here and priced it too low.
Both AMD and Nvidia have been able to make a fortune over the years by taking their regular 300$ chips, adding some ECC RAM and selling them for thousands.
A lot of professional use cases don't require any special GPU. The only reason even to use a dGPU is because Intel Xeon W and Ryzen non-G series lack an iGPU.
Nvidia's Quadro range goes right the way down to a GTX 1030-equivalent - the P400. 256 CUDA cores, 2 GB (thankfully GDDR5, unlike my Quadro K620), 0.64 fp32 TFLOPS, 30 W.
I should add that OEMs like Dell don't even offer non-workstation cards in their workstation chassis. So, if your reason for buying one of those machines doesn't involve any interactive rendering or GPU-compute, then you select a low-end professional card.
Not all Nvidia Quadros offer ECC RAM. It's only mentioned on the datasheets of the RTX 5000, 6000, and 8000 and the GV100. I'm pretty sure that's no oversight, for the others.
From the article: The AMD Radeon Pro WX 3200 is based on the company's Polaris architecture GPU featuring 640 stream processors that offers up to 1.66 TFLOPS of single precision compute performance.
From AMD's site: Peak Single Precision (FP32) Performance 3.3 TFLOPs
It's a mess-up on AMD's end, given that at the top of the same page, it reads "Up to 1.66 TFLOPS of peak SPFP to speed up professional applications for high performance.".
Also, given that the RX 590 with 3.6x the number of shaders (2304) and 20% higher boost clocks (1560MHz) achieves about 7.12 TFLOPS, there's no way this thing would pull 3.3 TFLOPS. It would need 1280 shaders for that.
Games need fp32 performance, so there's no way they'd cripple that.
And the fp64 simply isn't in the silicon, to begin with. This was primarily made as a gaming chip. Anyone needing much fp64 will have to buy Vega VII or Radeon Instinct MI50 or MI60
Huh, didn't even see the paragraph at the top of the AMD product page with the "... 1.66 TFLOPS ..." listing. I always just jump straight to the specs tables.
Looking at the specs page for the WX 3100, it's 1.25 TFLOPS / 78 GFLOPS, so makes sense for the WX 3200 to be 1.66 / 104 (a slight increase, not a 3-fold increase).
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
43 Comments
Back to Article
Mikewind Dale - Tuesday, July 2, 2019 - link
Fiddlesticks, no ECC.wolrah - Tuesday, July 2, 2019 - link
Is your "Source: AMD" link supposed to be going to a Youtube video if some Russian TV show?GreenReaper - Tuesday, July 2, 2019 - link
They're interviewing an AMD representative. Can't you tell by the red mugs?ballsystemlord - Tuesday, July 2, 2019 - link
That is really strange, maybe we should check their links for them more often.Or maybe it's another attack on the American voting system. :)
Slash3 - Wednesday, July 3, 2019 - link
We're on to you, Anton!Dragonstongue - Tuesday, July 2, 2019 - link
GDDR5 = automatic ECC if the AIB will use to "max" potential is their call (likely not for the price point)as for other knucklehad below no ecc no pci-e 4 "pass"
tool.
even pcie 2.0 at x8 can feed something like this perfectly fine.
mode_13h - Thursday, July 4, 2019 - link
Not sure what you mean by "automatic ECC".Funny enough, the virtually-identical RX 550 only has x8 lanes enabled (still a x16 connector, though). Whether it's possible to contrive a scenario where PCIe 2.0 x8 would have a measurable impact, I agree that it would be GPU-bottlenecked in real world use cases.
However, one small consideration would be that increasing bus bandwidth will typically reduce latency, even if slightly. This might help with VR (on a more powerful card, that can actually *do* VR), where a couple miliseconds' stutter can potentially have nauseating consequences.
Smell This - Thursday, July 4, 2019 - link
I think the point of the **no ECC** is that this is a 'high-yield' entry-level 50w pro card on a mature process. It is effectively Polaris gravy with nearly 600 dies on a 300mm wafer that includes 2 dozen+ Lexa variants.Phynaz - Tuesday, July 2, 2019 - link
Jeez, AMD must need to harvest every single die they can.AshlayW - Wednesday, July 3, 2019 - link
They don't have any other silicon for a card like this. Vega wasn't scaled down this far and Navi hasn't even released yet. I think this is Polaris "12" aka Lexa, it's like 100mm2. it's adorable.Polaris hasn't been replaced in this Performance tier yet.
PeachNCream - Tuesday, July 2, 2019 - link
That's not a 16x GPU if the image above is accurate. That probably is a non-issue from a performance standpoint given the overall compute power of the card, but it is noteworthy.For actual specs rather than a seemingly broken source link, here's the URL to AMD's site:
https://www.amd.com/en/products/professional-graph...
timecop1818 - Tuesday, July 2, 2019 - link
Bus TypePCIe 3.0 x16 (x8 electrical)
You are right. It's 8x. More than enough for what it's going to be doing, anyway.
mode_13h - Thursday, July 4, 2019 - link
Same as the RX 550, on which it's apparently based.PeachNCream - Saturday, July 6, 2019 - link
So true. It's just the professional variant of that same GPU.Soulkeeper - Tuesday, July 2, 2019 - link
Looks like a waste of money.No ecc, not pcie 16x, not pcie 4.0, overpriced, 2 generations old, etc.
Duncan Macdonald - Tuesday, July 2, 2019 - link
This is not a games card - it is a low end CAD/CAM card - the important thing for the potential buyers is the full professional CAD/CAM support. 16 lane PCIe and PCIe 4.0 are completely unimportant for this card. Much higher performance CAD/CAM cards are available but they cost far more than $199 - this card is expected to compete with the Nvidia P1000 not the RTX 8000!!Santoval - Wednesday, July 3, 2019 - link
So the doubling of the price (the 4GB variant of Radeon RX 550 this card is based on normally sells for $95 to $105) is due to CAD/CAM certification and professional drivers. Well, this is similar to the "professional price overhead" for much costlier cards..AshlayW - Wednesday, July 3, 2019 - link
Also the fact that it comes with an adorable little blue cooler!not_anton - Wednesday, July 3, 2019 - link
Doubling the price due to high-quality electronic components that gonna last forever, rather than bottom-of-the-bucket noname discounted stuff. Professional GPUs have always been built like that, and it’s quiet expensive to do especially at their low volumes.DanaGoyette - Tuesday, July 2, 2019 - link
I'm curious: how does the WX 3200 differ from the WX 3100 and the WX 4100?I just recently bought a WX 4100.
Gomez Addams - Tuesday, July 2, 2019 - link
Performance. The WX4100 is about 50% faster - 2.4 TF vs 1.66 TF.BOBBYdotTV - Tuesday, July 2, 2019 - link
8 cu's on the 3100 vs 10 cu's on the 3200.Santoval - Tuesday, July 2, 2019 - link
1.66 FP32 TFLOPs? Really?mode_13h - Thursday, July 4, 2019 - link
The S should be upper-case. TFLOPS is a unit, whereas a TFLOP is nonsense.Trillion Floating-point Operations Per Second.
phoenix_rizzen - Friday, July 5, 2019 - link
AMD's site shows it's twice that:Peak Single Precision (FP32) Performance
3.3 TFLOPs
Peak Double Precision (FP64) Performance
104 GFLOPs
mode_13h - Tuesday, July 9, 2019 - link
That's incorrect, as noted below.tipoo - Wednesday, July 3, 2019 - link
Man I love the design and color of their Pro shrouds. The gamer ones should be this minimal.rrinker - Wednesday, July 3, 2019 - link
No no, you're not a 'real; gamer unless your card has configurable RGB lighting and a graphics of some sort of "battle-y" image pasted all over it. Can't be a 'real' gamer if all you are concerned about is the performance and not the fancy looks.mode_13h - Thursday, July 4, 2019 - link
For gaming, you don't want single-slot. In a card this weak, it's unlikely to matter, but dual-slot is a win for noise and longevity.I agree that the aluminum shroud is classy.
Ro_Ja - Wednesday, July 3, 2019 - link
I honestly wanna see a single slot GPU like this from the gaming cards for both Nvidia and AMD, gaming video cards are looking more funny now.mode_13h - Thursday, July 4, 2019 - link
You might want to see them, but I don't want to hear them. And gamers won't want to OC them.I think there are good reasons gamer cards are >= 2 slots.
Also, consider that the pro cards typically have lower clocks.
V900 - Wednesday, July 3, 2019 - link
Is the performance really enough for anyone who needs a professional card though?AMD might have been a little to eager here and priced it too low.
Both AMD and Nvidia have been able to make a fortune over the years by taking their regular 300$ chips, adding some ECC RAM and selling them for thousands.
mode_13h - Thursday, July 4, 2019 - link
A lot of professional use cases don't require any special GPU. The only reason even to use a dGPU is because Intel Xeon W and Ryzen non-G series lack an iGPU.Nvidia's Quadro range goes right the way down to a GTX 1030-equivalent - the P400. 256 CUDA cores, 2 GB (thankfully GDDR5, unlike my Quadro K620), 0.64 fp32 TFLOPS, 30 W.
mode_13h - Thursday, July 4, 2019 - link
I should add that OEMs like Dell don't even offer non-workstation cards in their workstation chassis. So, if your reason for buying one of those machines doesn't involve any interactive rendering or GPU-compute, then you select a low-end professional card.mode_13h - Thursday, July 4, 2019 - link
Not all Nvidia Quadros offer ECC RAM. It's only mentioned on the datasheets of the RTX 5000, 6000, and 8000 and the GV100. I'm pretty sure that's no oversight, for the others.LarsBars - Thursday, July 4, 2019 - link
The WX 3100's display controller can support up to five displays. While this article says four, I would be surprised if it weren't 5.phoenix_rizzen - Thursday, July 4, 2019 - link
According to the specs from AMD's site it's only 4:Display Outputs
DisplayPort: 4x Mini-DisplayPort 1.4
HDMI™: None
DVI: None
VGA: None
phoenix_rizzen - Friday, July 5, 2019 - link
From the article:The AMD Radeon Pro WX 3200 is based on the company's Polaris architecture GPU featuring 640 stream processors that offers up to 1.66 TFLOPS of single precision compute performance.
From AMD's site:
Peak Single Precision (FP32) Performance
3.3 TFLOPs
Peak Double Precision (FP64) Performance
104 GFLOPs
silverblue - Friday, July 5, 2019 - link
It's a mess-up on AMD's end, given that at the top of the same page, it reads "Up to 1.66 TFLOPS of peak SPFP to speed up professional applications for high performance.".Also, given that the RX 590 with 3.6x the number of shaders (2304) and 20% higher boost clocks (1560MHz) achieves about 7.12 TFLOPS, there's no way this thing would pull 3.3 TFLOPS. It would need 1280 shaders for that.
Daeros - Friday, July 5, 2019 - link
You're forgetting that the gaming drivers & BIOS usually have higher clocks but crippled FP32 performance, especially double-precision calculations.mode_13h - Tuesday, July 9, 2019 - link
Games need fp32 performance, so there's no way they'd cripple that.And the fp64 simply isn't in the silicon, to begin with. This was primarily made as a gaming chip. Anyone needing much fp64 will have to buy Vega VII or Radeon Instinct MI50 or MI60
phoenix_rizzen - Monday, July 8, 2019 - link
Huh, didn't even see the paragraph at the top of the AMD product page with the "... 1.66 TFLOPS ..." listing. I always just jump straight to the specs tables.Looking at the specs page for the WX 3100, it's 1.25 TFLOPS / 78 GFLOPS, so makes sense for the WX 3200 to be 1.66 / 104 (a slight increase, not a 3-fold increase).
Nice catch. Bad AMD marketing folks! :)