Is the Nvidia RTX A4000 ADA suitable for Machine Learning?

server one
Rent GPU servers with instant deployment or a server with a custom configuration with professional-grade NVIDIA RTX 5500 / 5000 / A4000 cards. VPS with dedicated GPU cards are also available . The GPU card is dedicated to the VM and cannot be used by other clients. GPU performance in virtual machines matches GPU performance in dedicated servers.

In April, NVIDIA launched a new product, the RTX A4000 ADA, a small form factor GPU designed for workstation applications. This processor replaces the A2000 and can be used for complex tasks, including scientific research and engineering calculations and data visualization.

The RTX A4000 ADA features 6,144 CUDA cores, 192 Tensor and 48 RT cores, and 20GB GDDR6 ECC VRAM. One of the key benefits of the new GPU is its power efficiency: the RTX A4000 ADA consumes only 70W, which lowers both power costs and system heat. The GPU also allows you to drive multiple displays thanks to its 4x Mini-DisplayPort 1.4a connectivity.

When comparing the RTX 4000 SFF ADA GPUs to other devices in the same class, it should be noted that when running in single precision mode, it shows a performance similar to the latest generation RTX A4000 GPU, which consumes twice as much power (140W vs. 70W).

The ADA RTX 4000 SFF is built on the ADA Lovelace architecture and 5nm process technology. This enables next-generation Tensor Core and ray tracing cores, which significantly improve performance by providing faster and more efficient ray tracing and Tensor cores than the RTX A4000. In addition, ADA's RTX 4000 SFF comes in a small package - the card is 168mm long and as thick as two expansion slots.

Improved ray tracing kernels allows for efficient performance in environments where the technology is used, such as in 3D design and rendering. Furthermore, the new GPU's 20GB memory capacity enables it to handle large environments.

According to the manufacturer, fourth-generation Tensor cores deliver high AI computational performance - a twofold increase in performance over the previous generation. The new Tensor cores support FP8 acceleration. This innovative feature may work well for those developing and deploying AI models in environments such as genomics and computer vision.

It's also of note that the increase in encoding and decoding mechanisms makes the RTX 4000 SFF ADA a good solution for multimedia workloads such as video among others.

Technical specifications of NVIDIA RTX A4000 and RTX A5000 graphics cards, RTX 3090

ArchitectureAda LovelaceAmpereAmpereAmpere
Tech Process5 nm8 nm8 nm8 nm
Number of transistors (millions)35,80017,40028,30028,300
Memory bandwidth (Gb/s)280.0448768936.2
Video memory capacity (bits)160256384384
GPU memory (GB)20162424
CUDA cores6,1446 144819210496
Tensor cores192192256328
RT cores48486482
SP perf (teraflops)19.219,227,835,6
RT core performance (teraflops)44.337,454,269,5
Tensor performance (teraflops)306.8153,4222,2285
Maximum power (Watts)70140230350
InterfacePCIe 4.0 x 16PCI-E 4.0 x16PCI-E 4.0 x16PCIe 4.0 x16
Connectors4x Mini DisplayPort 1.4aDP 1.4 (4)DP 1.4 (4)DP 1.4 (4)
Form Factor2 slots1 slot2 slots2-3 slots
The vGPU softwarenonoYes, unlimitedYes. with limitations
Nvlinknono2x RTX A5000yes
CUDA support11.
VULKAN support1.3yesyesyes, 1.2
Price (USD)1,250100025001400

Description of the test environment

RTX A4000 ADARTX A4000
CPUAMD Ryzen 9 5950X 3.4GHz (16 cores)OctaCore Intel Xeon E-2288G, 3,5 GHz
RAM4x 32 Gb DDR4 ECC SO-DIMM2x 32 GB DDR4-3200 ECC DDR4 SDRAM 1600 MHz
Drive1Tb NVMe SSDSamsung SSD 980 PRO 1TB
MotherboardASRock X570D4I-2TAsus P11C-I Series
Operating SystemMicrosoft Windows 10Microsoft Windows 10

Test results

V-Ray 5 Benchmark

V-Ray GPU CUDA and RTX tests measure relative GPU rendering performance. The RTX A4000 GPU is slightly behind the RTX A4000 ADA (4% and 11%, respectively).

Machine Learning

"Dogs vs. Cats"

To compare the performance of GPUs for neural networks, we used the "Dogs vs. Cats" dataset - the test analyzes the content of a photo and distinguishes whether the photo shows a cat or a dog. All the necessary raw data can be found here. We ran this test on different GPUs and cloud services and got the following results:

In this test, the RTX A4000 ADA slightly outperformed the RTX A4000 by 9%, but keep in mind the small size and low power consumption of the new GPU.


AI-Benchmark allows you to measure the performance of the device during an AI model output task. The unit of measurement may vary according to the test, but usually it is the number of operations per second (OPS) or the number of frames per second (FPS).

RTX A4000RTX A4000 ADA
1/19. MobileNet-V21.1 — inference | batch=50, size=224x224: 38.5 ± 2.4 ms1.2 — training | batch=50, size=224x224: 109 ± 4 ms1.1 — inference | batch=50, size=224x224: 53.5 ± 0.7 ms1.2 — training | batch=50, size=224x224: 130.1 ± 0.6 ms
2/19. Inception-V32.1 — inference | batch=20, size=346x346: 36.1 ± 1.8 ms2.2 — training | batch=20, size=346x346: 137.4 ± 0.6 ms2.1 — inference | batch=20, size=346x346: 36.8 ± 1.1 ms2.2 — training | batch=20, size=346x346: 147.5 ± 0.8 ms
3/19. Inception-V43.1 — inference | batch=10, size=346x346: 34.0 ± 0.9 ms3.2 — training | batch=10, size=346x346: 139.4 ± 1.0 ms3.1 — inference | batch=10, size=346x346: 33.0 ± 0.8 ms3.2 — training | batch=10, size=346x346: 135.7 ± 0.9 ms
4/19. Inception-ResNet-V24.1 — inference | batch=10, size=346x346: 45.7 ± 0.6 ms4.2 — training | batch=8, size=346x346: 153.4 ± 0.8 ms4.1 — inference batch=10, size=346x346: 33.6 ± 0.7 ms4.2 — training batch=8, size=346x346: 132 ± 1 ms
5/19. ResNet-V2-505.1 — inference | batch=10, size=346x346: 25.3 ± 0.5 ms5.2 — training | batch=10, size=346x346: 91.1 ± 0.8 ms5.1 — inference | batch=10, size=346x346: 26.1 ± 0.5 ms5.2 — training | batch=10, size=346x346: 92.3 ± 0.6 ms
6/19. ResNet-V2-1526.1 — inference | batch=10, size=256x256: 32.4 ± 0.5 ms6.2 — training | batch=10, size=256x256: 131.4 ± 0.7 ms6.1 — inference | batch=10, size=256x256: 23.7 ± 0.6 ms6.2 — training | batch=10, size=256x256: 107.1 ± 0.9 ms
7/19. VGG-167.1 — inference | batch=20, size=224x224: 54.9 ± 0.9 ms7.2 — training | batch=2, size=224x224: 83.6 ± 0.7 ms7.1 — inference | batch=20, size=224x224: 66.3 ± 0.9 ms7.2 — training | batch=2, size=224x224: 109.3 ± 0.8 ms
8/19. SRCNN 9-5-58.1 — inference | batch=10, size=512x512: 51.5 ± 0.9 ms8.2 — inference | batch=1, size=1536x1536: 45.7 ± 0.9 ms8.3 — training | batch=10, size=512x512: 183 ± 1 ms8.1 — inference | batch=10, size=512x512: 59.9 ± 1.6 ms8.2 — inference | batch=1, size=1536x1536: 53.1 ± 0.7 ms8.3 — training | batch=10, size=512x512: 176 ± 2 ms
9/19. VGG-19 Super-Res9.1 — inference | batch=10, size=256x256: 99.5 ± 0.8 ms9.2 — inference | batch=1, size=1024x1024: 162 ± 1 ms9.3 — training | batch=10, size=224x224: 204 ± 2 ms
10/19. ResNet-SRGAN10.1 — inference | batch=10, size=512x512: 85.8 ± 0.6 ms10.2 — inference | batch=1, size=1536x1536: 82.4 ± 1.9 ms10.3 — training | batch=5, size=512x512: 133 ± 1 ms10.1 — inference | batch=10, size=512x512: 98.9 ± 0.8 ms10.2 — inference | batch=1, size=1536x1536: 86.1 ± 0.6 ms10.3 — training | batch=5, size=512x512: 130.9 ± 0.6 ms
11/19. ResNet-DPED11.1 — inference | batch=10, size=256x256: 114.9 ± 0.6 ms11.2 — inference | batch=1, size=1024x1024: 182 ± 2 ms11.3 — training | batch=15, size=128x128: 178.1 ± 0.8 ms11.1 — inference | batch=10, size=256x256: 146.4 ± 0.5 ms11.2 — inference | batch=1, size=1024x1024: 234.3 ± 0.5 ms11.3 — training | batch=15, size=128x128: 234.7 ± 0.6 ms
12/19. U-Net12.1 — inference | batch=4, size=512x512: 180.8 ± 0.7 ms12.2 — inference | batch=1, size=1024x1024: 177.0 ± 0.4 ms12.3 — training | batch=4, size=256x256: 198.6 ± 0.5 ms12.1 — inference | batch=4, size=512x512: 222.9 ± 0.5 ms12.2 — inference | batch=1, size=1024x1024: 220.4 ± 0.6 ms12.3 — training | batch=4, size=256x256: 229.1 ± 0.7 ms
13/19. Nvidia-SPADE13.1 — inference | batch=5, size=128x128: 54.5 ± 0.5 ms13.2 — training | batch=1, size=128x128: 103.6 ± 0.6 ms13.1 — inference | batch=5, size=128x128: 59.6 ± 0.6 ms13.2 — training | batch=1, size=128x128: 94.6 ± 0.6 ms
14/19. ICNet14.1 — inference | batch=5, size=1024x1536: 126.3 ± 0.8 ms14.2 — training | batch=10, size=1024x1536: 426 ± 9 ms14.1 — inference | batch=5, size=1024x1536: 144 ± 4 ms14.2 — training | batch=10, size=1024x1536: 475 ± 17 ms
15/19. PSPNet15.1 — inference | batch=5, size=720x720: 249 ± 12 ms15.2 — training | batch=1, size=512x512: 104.6 ± 0.6 ms15.1 — inference | batch=5, size=720x720: 291.4 ± 0.5 ms15.2 — training | batch=1, size=512x512: 99.8 ± 0.9 ms
16/19. DeepLab16.1 — inference | batch=2, size=512x512: 71.7 ± 0.6 ms16.2 — training | batch=1, size=384x384: 84.9 ± 0.5 ms16.1 — inference | batch=2, size=512x512: 71.5 ± 0.7 ms16.2 — training | batch=1, size=384x384: 69.4 ± 0.6 ms
17/19. Pixel-RNN17.1 — inference | batch=50, size=64x64: 299 ± 14 ms17.2 — training | batch=10, size=64x64: 1258 ± 64 ms17.1 — inference | batch=50, size=64x64: 321 ± 30 ms17.2 — training | batch=10, size=64x64: 1278 ± 74 ms
18/19. LSTM-Sentiment18.1 — inference | batch=100, size=1024x300: 395 ± 11 ms18.2 — training | batch=10, size=1024x300: 676 ± 15 ms18.1 — inference | batch=100, size=1024x300: 345 ± 10 ms18.2 — training | batch=10, size=1024x300: 774 ± 17 ms
19/19. GNMT-Translation19.1 — inference | batch=1, size=1x20: 119 ± 2 ms19.1 — inference | batch=1, size=1x20: 156 ± 1 ms

The results of this test show that the performance of the RTX A4000 is 6% higher than RTX A4000 ADA, however, with the caveat that the test results may vary depending on the specific task and operating conditions employed.


RTX A 4000

BenchmarkingModel average train time (ms)
Training double precision type mnasnet0_562.995805740356445
Training double precision type mnasnet0_7598.39066505432129
Training double precision type mnasnet1_0126.60405158996582
Training double precision type mnasnet1_3186.89460277557373
Training double precision type resnet18428.08079719543457
Training double precision type resnet34883.5790348052979
Training double precision type resnet501016.3950300216675
Training double precision type resnet1011927.2308254241943
Training double precision type resnet1522815.663013458252
Training double precision type resnext50_32x4d1075.4373741149902
Training double precision type resnext101_32x8d4050.0641918182373
Training double precision type wide_resnet50_22615.9953451156616
Training double precision type wide_resnet101_25218.524832725525
Training double precision type densenet121751.9759511947632
Training double precision type densenet169910.3225564956665
Training double precision type densenet2011163.036551475525
Training double precision type densenet1612141.505298614502
Training double precision type squeezenet1_0203.14435005187988
Training double precision type squeezenet1_198.04857730865479
Training double precision type vgg111697.710485458374
Training double precision type vgg11_bn1729.2972660064697
Training double precision type vgg132491.615080833435
Training double precision type vgg13_bn2545.1631927490234
Training double precision type vgg163371.1953449249268
Training double precision type vgg16_bn3423.8639068603516
Training double precision type vgg19_bn4314.5153522491455
Training double precision type vgg194249.422650337219
Training double precision type mobilenet_v3_large105.54619789123535
Training double precision type mobilenet_v3_small37.6680850982666
Training double precision type shufflenet_v2_x0_526.51611328125
Training double precision type shufflenet_v2_x1_061.260504722595215
Training double precision type shufflenet_v2_x1_5105.30067920684814
Training double precision type shufflenet_v2_x2_0181.03694438934326
Inference double precision type mnasnet0_517.397074699401855
Inference double precision type mnasnet0_7528.902697563171387
Inference double precision type mnasnet1_038.387718200683594
Inference double precision type mnasnet1_358.228821754455566
Inference double precision type resnet18147.95727252960205
Inference double precision type resnet34293.519492149353
Inference double precision type resnet50336.44991874694824
Inference double precision type resnet101637.9982376098633
Inference double precision type resnet152948.9351654052734
Inference double precision type resnext50_32x4d372.80876636505127
Inference double precision type resnext101_32x8d1385.1624917984009
Inference double precision type wide_resnet50_2873.048791885376
Inference double precision type wide_resnet101_21729.2765426635742
Inference double precision type densenet121270.13323307037354
Inference double precision type densenet169327.1932888031006
Inference double precision type densenet201414.733362197876
Inference double precision type densenet161766.3542318344116
Inference double precision type squeezenet1_074.86292839050293
Inference double precision type squeezenet1_134.04905319213867
Inference double precision type vgg11576.3767147064209
Inference double precision type vgg11_bn580.5839586257935
Inference double precision type vgg13853.4365510940552
Inference double precision type vgg13_bn860.3136301040649
Inference double precision type vgg161145.091052055359
Inference double precision type vgg16_bn1152.8028392791748
Inference double precision type vgg19_bn1444.9562692642212
Inference double precision type vgg191437.0987701416016
Inference double precision type mobilenet_v3_large30.876317024230957
Inference double precision type mobilenet_v3_small11.234536170959473
Inference double precision type shufflenet_v2_x0_57.425284385681152
Inference double precision type shufflenet_v2_x1_018.25782299041748
Inference double precision type shufflenet_v2_x1_533.34946632385254
Inference double precision type shufflenet_v2_x2_057.84676551818848


BenchmarkingModel average train time
Training half precision type mnasnet0_520.266618728637695
Training half precision type mnasnet0_7521.445374488830566
Training half precision type mnasnet1_026.714019775390625
Training half precision type mnasnet1_326.5126371383667
Training half precision type resnet1819.624991416931152
Training half precision type resnet3432.46446132659912
Training half precision type resnet5057.17473030090332
Training half precision type resnet10198.20127010345459
Training half precision type resnet152138.18389415740967
Training half precision type resnext50_32x4d75.56005001068115
Training half precision type resnext101_32x8d228.8706636428833
Training half precision type wide_resnet50_2113.76442432403564
Training half precision type wide_resnet101_2204.17311191558838
Training half precision type densenet12168.97401332855225
Training half precision type densenet16985.16453742980957
Training half precision type densenet201103.299241065979
Training half precision type densenet161137.54578113555908
Training half precision type squeezenet1_016.71830177307129
Training half precision type squeezenet1_112.906527519226074
Training half precision type vgg1151.7004919052124
Training half precision type vgg11_bn57.63327598571777
Training half precision type vgg1386.10869407653809
Training half precision type vgg13_bn95.86676120758057
Training half precision type vgg16102.91589260101318
Training half precision type vgg16_bn113.74778270721436
Training half precision type vgg19_bn131.56734943389893
Training half precision type vgg19119.70191955566406
Training half precision type mobilenet_v3_large31.30636692047119
Training half precision type mobilenet_v3_small19.44464683532715
Training half precision type shufflenet_v2_x0_513.710575103759766
Training half precision type shufflenet_v2_x1_023.608479499816895
Training half precision type shufflenet_v2_x1_526.793746948242188
Training half precision type shufflenet_v2_x2_024.550962448120117
Inference half precision type mnasnet0_54.418272972106934
Inference half precision type mnasnet0_754.021778106689453
Inference half precision type mnasnet1_04.42598819732666
Inference half precision type mnasnet1_34.618926048278809
Inference half precision type resnet185.803341865539551
Inference half precision type resnet349.756693840026855
Inference half precision type resnet5015.873079299926758
Inference half precision type resnet10128.268003463745117
Inference half precision type resnet15240.04594326019287
Inference half precision type resnext50_32x4d19.53421115875244
Inference half precision type resnext101_32x8d62.44826316833496
Inference half precision type wide_resnet50_233.533992767333984
Inference half precision type wide_resnet101_259.60897445678711
Inference half precision type densenet12118.052735328674316
Inference half precision type densenet16921.956982612609863
Inference half precision type densenet20127.85182476043701
Inference half precision type densenet16137.41891860961914
Inference half precision type squeezenet1_04.391803741455078
Inference half precision type squeezenet1_12.4281740188598633
Inference half precision type vgg1117.11493968963623
Inference half precision type vgg11_bn18.40585231781006
Inference half precision type vgg1328.438148498535156
Inference half precision type vgg13_bn30.672597885131836
Inference half precision type vgg1634.43562984466553
Inference half precision type vgg16_bn36.92122936248779
Inference half precision type vgg19_bn43.144264221191406
Inference half precision type vgg1940.5385684967041
Inference half precision type mobilenet_v3_large5.350713729858398
Inference half precision type mobilenet_v3_small4.016985893249512
Inference half precision type shufflenet_v2_x0_55.079126358032227
Inference half precision type shufflenet_v2_x1_05.593156814575195
Inference half precision type shufflenet_v2_x1_55.649552345275879
Inference half precision type shufflenet_v2_x2_05.355663299560547
Training double precision type mnasnet0_550.2386999130249
Training double precision type mnasnet0_7580.66896915435791
Training double precision type mnasnet1_0103.32422733306885
Training double precision type mnasnet1_3154.6230697631836
Training double precision type resnet18337.94031620025635
Training double precision type resnet34677.7706575393677
Training double precision type resnet50789.9243211746216
Training double precision type resnet1011484.3351316452026
Training double precision type resnet1522170.570478439331
Training double precision type resnext50_32x4d877.3719882965088
Training double precision type resnext101_32x8d3652.4944639205933
Training double precision type wide_resnet50_22154.612874984741
Training double precision type wide_resnet101_24176.522083282471
Training double precision type densenet121607.8699731826782
Training double precision type densenet169744.6409797668457
Training double precision type densenet201962.677731513977
Training double precision type densenet1611759.772515296936
Training double precision type squeezenet1_0164.3690824508667
Training double precision type squeezenet1_178.70647430419922
Training double precision type vgg111362.6095294952393
Training double precision type vgg11_bn1387.2539138793945
Training double precision type vgg132006.0230445861816
Training double precision type vgg13_bn2047.526364326477
Training double precision type vgg162702.2086429595947
Training double precision type vgg16_bn2747.241234779358
Training double precision type vgg19_bn3447.1724700927734
Training double precision type vgg193397.990345954895
Training double precision type mobilenet_v3_large84.65698719024658
Training double precision type mobilenet_v3_small29.816465377807617
Training double precision type shufflenet_v2_x0_527.401342391967773
Training double precision type shufflenet_v2_x1_048.322744369506836
Training double precision type shufflenet_v2_x1_582.22103118896484
Training double precision type shufflenet_v2_x2_0141.7021369934082
Inference double precision type mnasnet0_512.988653182983398
Inference double precision type mnasnet0_7522.422199249267578
Inference double precision type mnasnet1_030.056486129760742
Inference double precision type mnasnet1_346.953935623168945
Inference double precision type resnet18118.04479122161865
Inference double precision type resnet34231.52336597442627
Inference double precision type resnet50268.63497734069824
Inference double precision type resnet101495.2010440826416
Inference double precision type resnet152726.4922094345093
Inference double precision type resnext50_32x4d291.47679328918457
Inference double precision type resnext101_32x8d1055.10901927948
Inference double precision type wide_resnet50_2690.6917667388916
Inference double precision type wide_resnet101_21347.5529861450195
Inference double precision type densenet121224.35829639434814
Inference double precision type densenet169268.9145278930664
Inference double precision type densenet201343.1972026824951
Inference double precision type densenet161635.866231918335
Inference double precision type squeezenet1_061.92759037017822
Inference double precision type squeezenet1_127.009410858154297
Inference double precision type vgg11462.3375129699707
Inference double precision type vgg11_bn468.4495782852173
Inference double precision type vgg13692.8219032287598
Inference double precision type vgg13_bn703.3538103103638
Inference double precision type vgg16924.4353818893433
Inference double precision type vgg16_bn936.5075063705444
Inference double precision type vgg19_bn1169.098300933838
Inference double precision type vgg191156.3771772384644
Inference double precision type mobilenet_v3_large24.2356014251709
Inference double precision type mobilenet_v3_small8.85490894317627
Inference double precision type shufflenet_v2_x0_56.360034942626953
Inference double precision type shufflenet_v2_x1_014.301743507385254
Inference double precision type shufflenet_v2_x1_524.863481521606445
Inference double precision type shufflenet_v2_x2_043.8505744934082


The new graphics card has proven to be an effective solution for a number of work tasks. Thanks to its compact size, it is ideal for powerful SFF (Small Form Factor) computers. Also, it is notable that the 6,144 CUDA cores and 20GB of memory with a 160-bit bus makes this card one of the most productive on the market. Furthermore, a low TDP of 70W helps to reduce power consumption costs. Four Mini-DisplayPort ports allow the card to be used with multiple monitors or as a multi-channel graphics solution.

The RTX 4000 SFF ADA represents a significant advance over previous generations, delivering performance equivalent to a card with twice the power consumption. With no PCIe power connector, the RTX 4000 SFF ADA is easy to integrate into low-power workstations without sacrificing high performance.

Rent GPU servers with instant deployment or a server with a custom configuration with professional-grade NVIDIA RTX 5500 / 5000 / A4000 cards. VPS with dedicated GPU cards are also available . The GPU card is dedicated to the VM and cannot be used by other clients. GPU performance in virtual machines matches GPU performance in dedicated servers.

Other articles


Personal Shadowsocks/XRay (XTLS) VPN-server with 3X-UI control panel

Hostkey has prepared a new service for its clients - server rental with pre-installed VPN 3X-UI. The new service is available for order on the company's website.


Dell and Supermicro servers: authorization via LDAP

We describe how to integrate heterogeneous server platforms with centralized LDAP authentication.


What is Distributed Storage? Types and Examples

In this article, we discuss distributed storage, a technology that stores data across multiple servers for increased availability, reliability, and scalability.


Installing and configuring ESXi servers with ESXi-Foreman PXE Deploy

Let us explain how ESXi-Foreman PXE Deploy can simplify your virtualization efforts and save you time and effort when installing and configuring ESXi servers.


Top 10 graphics cards for machine learning

How to choose the right graphics card and maximize the efficiency of processing large amounts of data and performing parallel computing.

HOSTKEY Dedicated servers and cloud solutions Pre-configured and custom dedicated servers. AMD, Intel, GPU cards, Free DDoS protection amd 1Gbps unmetered port 30
4.3 67 67