Starting things off, we have a set of new video cards from NVIDIA– and a launch that seemingly shows that NVIDIA is preparing to upgrade its expert visualization branding. Being revealed today and set to deliver at the end of the year is the NVIDIA RTX A6000, NVIDIAs next-generation, Ampere-based expert visualization card. The successor to the Turing-based Quadro RTX 8000/6000, the A6000 will be NVIDIAs flagship professional graphics card, offering everything under the sun as far as NVIDIAs graphics functions go, and chart-topping performance to back it up. The A6000 will be a Quadro card in whatever however name; literally.
NVIDIAs second GTC of 2020 is occurring this week, and as has rapidly become a tradition, one of CO Jensen Huangs “kitchenside talks” begins the occasion. As the de facto replacement for GTC Europe, this fall virtual GTC is a bit of a lower-key event relative to the Spring edition, but its still one that is seeing some NVIDIA hardware presented to the world.
NVIDIA Professional Visualization CardSpecification Comparison
Half Precision.
?
?
32.6 TFLOPS.
29.6 TFLOPS.
NVIDIA has actually never listed GA102 as offering ECC on its internal pathways– this is generally limited to their big, datacenter-class chips– so this is nearly certainly partial support by means of “soft” ECC, which provides error correction against the DRAM and DRAM bus by setting aside some DRAM capacity and bandwidth to function as ECC. The A6000 also supports NVIDIAs standard frame lock and 3D Vision Pro features with their respective ports.
Tensor Performance.
?
?
130.5 TFLOPS.
118.5 TFLOPs( FP16).
TDP.
300W.
300W.
295W.
250W.
Memory Bus Width.
384-bit.
384-bit.
384-bit.
4096-bit.
For screen outputs, the A6000 ships with a quad-DisplayPort configuration, which is normal for NVIDIAs high-end expert visualization cards. Notably this generation, nevertheless, this implies the A6000 remains in a little bit of an odd spot given that DisplayPort 1.4 is slower than the HDMI 2.1 requirement likewise supported by the GA102 GPU. I would expect that its possible for the card to drive an HDMI 2.1 screen with a passive adapter, however this is going to be reliant on how NVIDIA has configured the card and if HDMI 2.1 signaling will tolerate such an adapter.
Boost Clock.
?
?
1770MHz.
~ 1450MHz.
Manufacturing Process.
Samsung 8nm.
Samsung 8nm.
TSMC 12nm FFN.
TSMC 12nm FFN.
The most noteworthy effect here is the addition of display outputs, something that was never ever on NVIDIAs calculate cards for obvious factors. The A40 consists of three DisplayPort outputs (one fewer than the A6000), offering the server-focused card the capability to straight drive a display. In describing the inclusion of screen I/O in a server part, NVIDIA said that theyve had demands from users in the media and broadcast industry, who have actually been utilizing servers in places like video trucks, however still need display screen outputs.
Total NVIDIA is no complete stranger to offering passively cooled cards; however its been a while because we last saw a passively cooled high-end Quadro card. Most just recently, NVIDIAs passive cards have been focused on the compute market, with parts like the Tesla T4 and P40. The A40, on the other hand, is a bit different and a bit more enthusiastic, and a reflection of the blurring lines between compute and graphics in at least a few of NVIDIAs markets.
NVIDIA A40– Passive ProViz.
CUDA Cores.
10752.
10752.
4608.
5120.
Architecture.
Ampere.
Ampere.
Turing.
Volta.
In regards to efficiency, NVIDIA is promoting the A6000 as offering almost two times the efficiency (or more) of the Quadro RTX 8000 in specific situations, especially jobs making the most of the considerable boost in FP32 CUDA cores or the comparable efficiency boost in RT core throughput. Unfortunately NVIDIA has either yet to lock down the specs for the card or is choosing against revealing them at this time, so we do not understand what the clockspeeds and resulting efficiency in FLOPS will be. Especially, the A6000 just has a TDP of 300W, 50W lower than the GeForce RTX 3090, so I would anticipate this card to be clocked lower than the 3090.
Otherwise, as we saw with the GeForce cards introduced last month, Ampere itself is not a significant technological overhaul to the previous Turing architecture. While newer and substantially more effective, there are not numerous brand-new marquee features to be found on the card. Together with the expanded number of information types supported in the tensor cores (especially BFloat16), the other changes probably to be seen by professional visualization users is decode support for the new AV1 codec, along with PCI-Express 4.0 assistance, which will offer the cards twice the bus bandwidth when utilized with AMDs current platforms.
Like the A6000, the A40 will be hitting the streets in the near future. Developed to be sold mainly through OEMs, NVIDIA expects it to start showing up in servers in early 2021.
By the numbers, the A40 is a comparable flagship-level graphics card, using a totally made it possible for GA102 GPU. Its not quite a twin to the A6000, however besides the cooling difference, the only other modification under the hood is the memory configuration. Whereas the A6000 utilizes 16 Gbps GDDR6, A40 clocks it to 14.5 Gbps. Otherwise NVIDIA has not divulged expected GPU clockspeeds, but with a 300W TDP, we d anticipate them to be comparable to the A6000.
Introduce Price.
?
?
$ 10,000.
$ 9,000.
GPU.
GA102.
GA102.
TU102.
GV100.
Introduce Date.
12/2020.
Q1 2021.
Q4 2018.
March 2018.
The A6000 will be the first of todays video cards to deliver. According to NVIDIA, the card will be available in the channel as an add-in card beginning in mid-December– in the nick of time to make a 2020 launch. The card will then start showing up in OEM systems in early 2021.
The first professional visualization card to be introduced based on NVIDIAs brand-new Ampere architecture, the A6000 will have NVIDIA hitting the market with its finest foot forward. The card uses a fully-enabled GA102 GPU– the very same chip utilized in the GeForce RTX 3080 & & 3090– and with 48GB of memory, is loaded with as much memory as NVIDIA can put on a single GA102 card today.
Tensor Cores.
336.
336.
576.
640.
Signing up with the new A6000 is a very comparable card designed for passive cooling, the NVIDIA A40. Based upon the same GA102 GPU as the A6000, the A40 offers virtually all of the same functions as the active-cooled A6000, just in a purely passive type element suitable for use in high density servers.
Single Precision.
?
?
16.3 TFLOPS.
14.8 TFLOPS.
NVLink.
1x NVLink3112.5 GB/sec.
1x NVLink3112.5 GB/sec.
1x NVLInk250GB/sec.
2x NVLInk2100GB/sec.
Cooling.
Active.
Passive.
Active.
Active.
.
A6000.
A40.
RTX 8000.
GV100.
Ultimately, this serves as something of an extra feature differentiator between the A40 and NVIDIAs official PCIe compute card, the PCIe A100. And while its not particularly aimed at the edge calculate market, where the T4 still reigns supreme, make no error: the A40 is still capable of being used as a calculate card.
Memory Clock.
16Gbps GDDR6.
14.5 Gbps GDDR6.
14Gbps GDDR6.
1.7 Gbps HBM2.
ECC.
Partial( DRAM).
Partial( DRAM).
Partial( DRAM).
Complete.
VRAM.
48GB.
48GB.
48GB.
32GB.
Quadro No More?
Beginning things off, we have a pair of new video cards from NVIDIA– and a launch that apparently suggests that NVIDIA is getting ready to overhaul its expert visualization branding. Being announced today and set to ship at the end of the year is the NVIDIA RTX A6000, NVIDIAs next-generation, Ampere-based expert visualization card. The follower to the Turing-based Quadro RTX 8000/6000, the A6000 will be NVIDIAs flagship expert graphics card, using everything under the sun as far as NVIDIAs graphics features go, and chart-topping efficiency to back it up. The first expert visualization card to be launched based on NVIDIAs new Ampere architecture, the A6000 will have NVIDIA hitting the market with its finest foot forward. In all these cases, both Quadro and NVIDIAs former Tesla lineup have actually come to represent NVIDIAs “premium” offerings: parts that get access to the complete suite of NVIDIAs hardware and software application features, unlike the consumer GeForce items which have particular high-end functions kept.
Maybe more significant than that is the general state of NVIDIAs expert services. A crucial foundation of NVIDIAs graphics items, professional visualization is a relatively stable market– which is to state its not a major growth market in the method that video gaming and datacenter compute have been. As a result, professional visualization has been getting slowly subsumed by NVIDIAs calculate parts, specifically in the server area where many items can be provisioned for either compute or graphics requirements. In all these cases, both Quadro and NVIDIAs previous Tesla lineup have come to represent NVIDIAs “premium” offerings: parts that get access to the complete suite of NVIDIAs hardware and software application features, unlike the consumer GeForce products which have particular high-end features withheld.
At any rate, it will be intriguing to see where NVIDIA goes from here. Even with the overlap in audiences, branding division has its advantages at times. And with NVIDIA now producing GPUs that lack vital screen capabilities (GA100), it appears like making it clear what hardware can (and cant) be used for graphics is going to stay important moving forward.
With all of that stated, there are a couple of factors in play that might be driving NVIDIAs decision. Is that the company has actually already retired one of its other item brand names in the last couple of years: Tesla. Formerly used for NVIDIAs compute accelerators, Tesla was retired and never replaced, leaving us with the likes of the NVIDIA T4 and A100. Naturally, Tesla is something of a diplomatic immunity, as the name has actually progressively ended up being synonymous with the electric cars and truck business, regardless of in both cases being picked as a reference to the famous scientist. Quadro, by comparison, has relatively little (however not absolutely no) overlap with other organization entities.
Perhaps because of that last-minute change, NVIDIA hasnt released any official explanation for their decision. At stated value its certainly an odd one, as the Quadro brand name is among NVIDIAs longest-lived brand names, second just to GeForce itself. NVIDIA still manages the lions share of the professional visualization market also, so at face worth there appears to be little reason for NVIDIA to shake-up a very stable market.
For veteran observers, perhaps the most intriguing development from todays launch is whats not present: NVIDIAs Quadro branding. Regardless of being targeted at their traditional professional visualization market, the A6000 is not being branded as a Quadro card, a modification that was made at almost the last minute.
It may extremely well be that NVIDIA doesnt see a requirement for a specific Quadro brand name too much longer, since the market for Quadro (professional visualization) and Tesla (computing) are one in the very same. Though the two vary in their particular needs, they still utilize the exact same NVIDIA hardware, and frequently pay the same high NVIDIA costs.
