Monday, February 10th 2020

AMD Radeon Instinct MI100 "Arcturus" Hits the Radar, We Have its BIOS

AMD's upcoming large post-Navi graphics chip, codenamed "Arcturus," will debut as "Radeon Instinct MI100", which is an AI-ML accelerator under the Radeon Instinct brand, which AMD calls "Server Accelerators." TechPowerUp accessed its BIOS, which is now up on our VGA BIOS database. The card goes with the device ID "0x1002 0x738C," which confirms "AMD" and "Arcturus,". The BIOS also confirms that memory size is at a massive 32 GB HBM2, clocked at 1000 MHz real (possibly 1 TB/s bandwidth, if memory bus width is 4096-bit).

Both Samsung (KHA884901X) and Hynix memory (H5VR64ESA8H) is supported, which is an important capability for AMD's supply chain. From the ID string "MI100 D34303 A1 XL 200W 32GB 1000m" we can derive that the TDP limit is set to a surprisingly low 200 W, especially considering this is a 128 CU / 8,192 shader count design. Vega 64 and Radeon Instinct MI60 for comparison have around 300 W power budget with 4,096 shaders, 5700 XT has 225 W with 2560 shaders, so either AMD achieved some monumental efficiency improvements with Arcturus or the whole design is intentionally running constrained, so that AMD doesn't reveal their hand to these partners, doing early testing of the card.

-- images removed --

Looking through the BIOS I also found what looks like several clock tables that top out at 1334 MHz, 1091 MHz, 1000 MHz. AMD's engineers typically list clocks in the following order: GPU clock, SOC clock, memory clock. This suggests that the GPU will tick at up to 1334 MHz, way lower than what Navi and Vega were able to achieve — maybe they do that to operate the chip in a more power-efficient way. The memory clock at 1000 MHz, matches the BIOS id string's "1000m", and falls in range with the 2.0 - 2.4 Gbps that Samsung is speccing their HBM2 memory chips at.

Arcturus' debut as a Radeon Instinct product follows the pattern of AMD debuting new big GPUs as low-volume/high-margin AI-ML accelerators first, followed by Radeon Pro and finally Radeon client graphics products. Arcturus is not "big Navi," rather it seems to be much closer to Vega than to Navi, which makes perfect sense given its target market. AMD's Linux sources mention "It's because Arcturus has not 3D engine", which could hint at what AMD did with this chip: take Vega and remove all 3D raster graphics ability, which shaves a few billion transistors off the silicon, freeing up space for more CUs. For gamers, AMD is planning a new line of Navi 20-series chips leveraging 7 nm EUV for launch throughout 2020. Various higher-ups at AMD, including its CEO, publicly hinted that a big client-segment GPU is in the works, and that the company is very much interested at taking another swing at premium 4K UHD gaming. Sources: Arcturus Linux Patches, Arcturus Linux Patches
Add your own comment

73 Comments on AMD Radeon Instinct MI100 "Arcturus" Hits the Radar, We Have its BIOS

#26
1idd0kun
notb
In real life, Nvidia dominated GPU computing before they added Tensors - all thanks to a better ecosystem and support. And nothing changed here.
Even if Mi100 temporarily pulls ahead in performance, it won't be enough.
Well, I think it's a matter of whether their software stack get some interest from developers. Having key super computers (like Frontier) using their hardware will help in that department.

Also, AMD already has a MI200 in the works. MI100 looks like it's gonna be pretty powerful, but MI200 will be even better, and is probably what they're gonna deploy in Frontier.
Posted on Reply
#27
Midland Dog
only amd can make a gpu so bad at being a gpu that it has no graphics processing capability
Posted on Reply
#28
notb
Midland Dog
only amd can make a gpu so bad at being a gpu that it has no graphics processing capability
This is not a graphics card.
Posted on Reply
#29
HenrySomeone
Hehe, they seem to follow the motto: "Bad product is still better than no product"
Posted on Reply
#30
Vya Domus
john_
I looked at NVIDIA Quadro RTX 8000 specs and it says over 200 TOPS at INT8. Is there really so huge difference between Radeon and RTX, or does Nvidia counts differently?
They use different ratios of execution units. Everything is a trade-off, MI60 has a lot of FP64 units, Turing doesn't, Volta does but it doesn't have any RT cores. The thing is though INT8/INT4/FP16 aren't that critical.

Out of all of those, FP64 units have become indispensable. It used to be that they were very expensive power and size wise and that's why GPUs of the past skimped on that but any real compute accelerator nowadays needs to have strong FP64 performance. 64 bit floating point is usually the de facto precision for simulations and that sort of stuff, you can do without tensor cores or INT8/INT4/FP16 but not without FP64 in a data center environment. That's why there have been no large Turing based Tesla's, because no one would have wanted them due to their poor FP64 performance.
Posted on Reply
#31
notb
HenrySomeone
Hehe, they seem to follow the motto: "Bad product is still better than no product"
They've signed a few large contracts and they have to deliver a GPGPU accelerator. It doesn't have to be the best available. It only has to match the agreed specification.
Posted on Reply
#32
ARF
T1beriu
That's not AMD's spec page. It's Techpowerup's speculated spec list.
When the user with the Arcturus card run GPU-Z and submitted the BIOS to the database, well, I don't think so.
I think there is a way for the software to read the specifications and put them into the database.

HwGeek
I smell $1000 Gaming GPU's this year by AMD...
No way, I smell much lower prices, after all, AMD has to regain some mindshare and lost positions.

ppn
ASML
"...new maximum die size of 429 mm². Say goodbye to the massive dies we got used to from Intel and Nvidia. ...",

Vega 20 Transistors 13,230 million Density 40.0M / mm² 7 nm Die Size 331 mm²

MI100 Transistors29,000 million Density 41.4M / mm² 7 nm+ Die Size 700 mm²

Same Density as first gen 7nm. The real EUV should be 18% denser or 505 mm² NAVi21 gets shrinked to exactly 429mm²
Probably fake limit of only 429 mm². That would mean bye-bye enthusiasts videocards.
Posted on Reply
#33
T4C Fantasy
CPU & GPU DB Maintainer
ppn
ASML
"...new maximum die size of 429 mm². Say goodbye to the massive dies we got used to from Intel and Nvidia. ...",

Vega 20 Transistors 13,230 million Density 40.0M / mm² 7 nm Die Size 331 mm²

MI100 Transistors29,000 million Density 41.4M / mm² 7 nm+ Die Size 700 mm²

Same Density as first gen 7nm. The real EUV should be 18% denser or 505 mm² NAVi21 gets shrinked to exactly 429mm²
So this will be first gen 7nm to reach 500+mm2? Or for it to have 8192 cores it is made possible by losing the raster engines?

ARF
When the user with the Arcturus card run GPU-Z and submitted the BIOS to the database, well, I don't think so.
I think there is a way for the software to read the specifications and put them into the database.



No way, I smell much lower prices, after all, AMD has to regain some mindshare and lost positions.



Probably fake limit of only 429 mm². That would mean bye-bye enthusiasts videocards.
The bios doesn't show any die information, all of this is speculation by me and will be updated when new info comes out.
Posted on Reply
#34
1idd0kun
HenrySomeone
Hehe, they seem to follow the motto: "Bad product is still better than no product"
No idea why anyone would think this is a "bad product" when we don't even know the specs yet.
Posted on Reply
#35
IceShroom
john_
I looked at NVIDIA Quadro RTX 8000 specs and it says over 200 TOPS at INT8. Is there really so huge difference between Radeon and RTX, or does Nvidia counts differently?
TOPS vs TFLOPS, not same.
Posted on Reply
#36
evolucian911
Theoretically, and based on V7 pro performance. A version of this card for Pros would be insane. To call it a monster would be an understatement. On top of that, since its Vega they can add the ability like with last Vega based card to use m.2 as extra ram. I'm certain this card is coming. They just need to focus and market it for that segment and not gaming like before. Call it the FirePro X. This specific version though is obviously just for compute and even still it is beyond anything on the market by Far.
Posted on Reply
#37
Khonjel
Let me set this straight. It's like taking out passenger/driver cabin and trunk space off a Tesla and fill it with battery and more powerful motor (delete raster capability, basically the little doo-dah that shows stuff on your monitor and fill the space with CUs). Even steering system because this Tesla will just go straight in highly specialized environment (only for AI-ML acceleration unlike GPUs that are usually jack of all trades).
Except extra battery and motor wouldn't be used for faster 0-60 and top speed rather hauling more stuff so a Tesla semi or pickup (not for gaming but workstation only).

Some things you get after reading this thread:
It's not a gaming card at all. It's workstation card through and through. But people will like a gaming card with this spec.
People don't understand that with raster stuff added back in this thing will become MASSIVE.
AMD also has brain-dead haters for their products who don't even understand what they're hating.

My personal take is that this is a good development. I've long prescribed that AMD should have different architectures for different markets instead of "jack of all trades, master of none". Now that CPU business is making them money, hope they spend clever money on R&D for RTG.
Posted on Reply
#38
evolucian911
Khonjel
Let me set this straight. It's like taking out passenger/driver cabin and trunk space off a Tesla and fill it with battery and more powerful motor (delete raster capability, basically the little doo-dah that shows stuff on your monitor and fill the space with CUs). Even steering system because this Tesla will just go straight in highly specialized environment (only for AI-ML acceleration unlike GPUs that are usually jack of all trades).
Except extra battery and motor wouldn't be used for faster 0-60 and top speed rather hauling more stuff so a Tesla semi or pickup (not for gaming but workstation only).

Some things you get after reading this thread:
It's not a gaming card at all. It's workstation card through and through. But people will like a gaming card with this spec.
People don't understand that with raster stuff added back in this thing will become MASSIVE.
AMD also has brain-dead haters for their products who don't even understand what they're hating.

My personal take is that this is a good development. I've long prescribed that AMD should have different architectures for different markets instead of "jack of all trades, master of none". Now that CPU business is making them money, hope they spend clever money on R&D for RTG.
This is why I said a "version " of this card for pro. Which would obviously mean rasters in amongst other things and inevitably higher tdp.
Posted on Reply
#39
Proedros
I hope W1zz had nothing to do with it. :mad:
Posted on Reply
#40
Steevo
HenrySomeone
Hehe, they seem to follow the motto: "Bad product is still better than no product"
Your poor attempt at trolling and fanboyism is honestly the coolest thing ever, ohhh baby you make me so excited for sucking Nvidia off with you......

700mm die, at 1Ghz plus, and an interposer to hold it and HBM? I imagine yield loss is huge, but maybe they have perfected it so it's actually profitable. Now when do we get to see actual performance numbers?
Posted on Reply
#41
notb
Steevo
700mm die, at 1Ghz plus, and an interposer to hold it and HBM? I imagine yield loss is huge, but maybe they have perfected it so it's actually profitable. Now when do we get to see actual performance numbers?
Or they have no choice - they need to make such a chip (PoC).
Posted on Reply
#42
HD64G
A shock for many that is has 128CUs as they couldn't get that big Navi will have even 80CUs. But AMD is on rails now with only marketing being inferior but their products are on top level. Even their latest GPU drivers for Navi arch are improved (the biggest customer problem for the last 5-6 months). Impressive feat nevertheless, especially for 200W. Big Navi now can easily become the fastest GPU with some distance while using under 300W. It seems that 7nm+ helps muchly in efficiency.
Posted on Reply
#43
TheGuruStud
Steevo
Your poor attempt at trolling and fanboyism is honestly the coolest thing ever, ohhh baby you make me so excited for sucking Nvidia off with you......

700mm die, at 1Ghz plus, and an interposer to hold it and HBM? I imagine yield loss is huge, but maybe they have perfected it so it's actually profitable. Now when do we get to see actual performance numbers?
No doubt cut down dies will be sold, so I'm sure there's plenty of profit even if it's 50% fully functioning yield.
Posted on Reply
#44
HwGeek
HD64G
A shock for many that is has 128CUs as they couldn't get that big Navi will have even 80CUs. But AMD is on rails now with only marketing being inferior but their products are on top level. Even their latest GPU drivers for Navi arch are improved (the biggest customer problem for the last 5-6 months). Impressive feat nevertheless, especially for 200W. Big Navi now can easily become the fastest GPU with some distance while using under 300W. It seems that 7nm+ helps muchly in efficiency.
Maybe this is related to what they have done with Ryzen 4000 Vega cores efficiency.
Posted on Reply
#45
TheGreatPascalian
The coments here, jesus. This GPU can't even render graphics. Do people even realize that ? Its purely AI GPU which explains the low wattage. Its similar to 75W Tesla T4 which is purely AI as well while having RTX 2070S spec.
Posted on Reply
#46
ARF
ppn
The estimation changed, not a 700 die any loner.
Radeon VII is already 331 mm² , it is physically impossible to fit 2X the shaders with only 6 billion transistors more, and in 420 mm² .

420 mm² is impossible with these specs.

Have you got any source that they can manufacture larger than 429 mm² dies on N7+ and have you got any source that this particular chip is on N7+, and not in N7 ?
Posted on Reply
#47
notb
TheGreatPascalian
The coments here, jesus. This GPU can't even render graphics. Do people even realize that ? Its purely AI GPU which explains the low wattage. Its similar to 75W Tesla T4 which is purely AI as well while having RTX 2070S spec.
Absolutely NOT.

Obviously, this card can render. It's built around a normal GPU. It just can't provide a video signal - there are no outputs and no logic dedicated for this task.
It can be used in any scenario that can utilize GPGPU (including AI, obviously).

This is NOT similar to Tesla T4.
In green camp you have the V100, which is an all-mighty, all-round, dual-slot accelerator. Mi100 (like Mi60 now) will compete in this segment.
Nvidia also makes the Tesla T4, which has half of V100 Tensor (AI) potential, but just 5% of it's double-precision performance. T4 is single-slot, 1/4th of V100 price and uses 75W (V100 is up to 300W).

Which means that if you need V100's double-precision (all-round) performance, you buy a V100. You can't go wrong with this card.
But if you don't need it, you buy 2xTesla T4's - you get pretty much the same performance in stuff like Deep Learning (e.g. image recognition) for half the money and half the power.
Posted on Reply
#48
kapone32
HenrySomeone
That would be another massive failure just like the Crap-eon 7, but then again, AMD just doesn't seem to tire of failures... :D
Why was the Radeon 7 a failure in your eyes?
Posted on Reply
#49
ppn
ARF
Radeon VII is already 331 mm² , it is physically impossible to fit 2X the shaders with only 6 billion transistors more, and in 420 mm² .
In Mi60 -4096 shading units = 160 mm2. double that, the memory controllers remain unchanged. SO therefore 50% bigger die 2X shaders. Shrink to 7nm+ with 18% better density. looks like perfect prediction by DB maintainer.
Posted on Reply
#50
TheGuruStud
kapone32
Why was the Radeon 7 a failure in your eyes?
B/c it wasn't 3 hundred dollars and they couldn't afford it...and ignoring that it wasn't actually a gaming card.
Posted on Reply
Add your own comment