AMD 選擇在農曆年期間將 Radeon VII 推出市場販售,讓我們有些壓力在。
相較於第一代 Vega 架構的 Radeon Vega 64 顯示卡,採用第二代 Vega 架構 GPU 的 AMD Radeon VII 除了 GPU 提升至 7nm 製程外,同時在記憶體部分也獲得非常顯著的提升;Radeon Vega 64 為 8GB HBM2,而 Radeon VII 是 16GB HBM2 記憶體配置。
托 TSMC 7nm 製程關係,AMD Radeon VII 不單單在效能方面獲得提升,同時也讓 GPU Die 面積從 495mm² 縮減至 331mm²。因為如此,AMD 才有機會在與 AMD Radeon RX Vega 64 相同大小的情況下,額外為 AMD Radeon VII 加入多 2 顆 HBM2 記憶體顆粒,讓記憶體頻寬提升至 1TB/s。
記憶體容量增加主要是因應遊戲對於顯示卡需求的提升,這也是因為使用者的螢幕解析度一直在進化。從過去的 720p、1080p、1440p 到頂級玩家追求的 3840 x 2160 解析度或是多螢幕同時輸出,這除了需要強而有力的顯示卡外,顯示卡本身所搭載的記憶體大小也有一定要求。
AMD 以 2010 年推出的 Battlefield Bad Company 2 512MB 顯示卡記憶體需求到 2018 年的 Battlefield V 需要 6GB / 8GB 顯示卡記憶體需求為實際例子說明顯示卡記憶體容量增加這件事;遊戲以外,Adobe Premiere 這類型的剪輯軟體對於顯示卡記憶體也有一定要求在。
Radeon VII | Radeon RX Vega 64 | Radeon RX Vega 56 | |
---|---|---|---|
Architecture | Vega 20 | Vega 10 | Vega 10 |
Process | 7nm | 14nm | 14nm |
Transistor Count | 13.2 billion | 12.5 billion | 12.5 billion |
Die Size | 331mm² | 495mm² | 495mm² |
Compute Units | 60 | 64 | 56 |
Stream Processors | 3840 | 4096 | 3584 |
Base GPU Clock | 1400MHz | 1274MHz | 1156MHz |
Boost GPU Clock | 1750MHz | 1546MHz | 1471MHz |
Engine Clock(Peak) | 1800MHz | 1630MHz | 1590MHz |
SP Performance(Peak) | Up to 14.2 TFLOPs | Up to 12.7 TFLOPs | Up to 10.5 TFLOPs |
Half Precision Performance(Peak) | Up to 28.1 TFLOPs | Up to 25.3 TFLOPs | Up to 21.0 TFLOPs |
Texture Fill-Rate(Peak) | 432.24 GT/s | Up to 395.8 GT/s | Up to 330.0 GT/s |
Pixel Fill-Rate(Peak) | 115.26 GP/s | Up to 98.9 GP/s | Up to 94.0 GP/s |
ROPs | 64 | 64 | 64 |
HBM2 | 16GB | 8GB | 8GB |
Memory Bandwidth | 1TB/s | 483.8GB/s | 410GB/s |
Memory Interface | 4096 bit | 2048 bit | 2048 bit |
Board Power | 300W | 295W | 210W |
作為 2 代 Vega 架構,與 Vega 10 架構的 Radeon RX Vega 64 與 Radeon RX56 相比,AMD Radeon VII 在頻率和延遲部分有所提升。
規格表中清楚看到,相較於 Vega 10 的 12.5 billion 電晶體數量,Die Size 較小的 Vega 20 電晶體數量來到了 13.2 billion,同時在 GPU 時脈都明顯變高;然而,Vega 20 的 Radeon VII 減少了 Compute Units 和 Stream Processors 的數量,但這樣的減少並未影響到 Radeon VII 的效能表現。
此外,Radeon VII 的溫度感測點也較 Radeon RX Vega 64 多了 2 倍。
這次對手我們找了 GeForce RTX 2080 作為比較對手,而我們選擇 EVGA GeForce RTX 2080 FTW3 ULTRA GAMING、EVGA GeForce RTX 2080 FTW3 ULTRA GAMING 與 AMD Radeon VII 公版卡進行測試。
採用 Turing 架構的 NVIDIA GeForce RTX 2080 與 NVIDIA GeForce 2070 都是 8GB GDDR6 記憶體,並且支援 Real Time Ray Tracing(即時光線追蹤)技術;目前支援 Ray Tracing 技術的遊戲多達 11 款,其中最有名氣的應該是 Battlefield V,但這款遊戲在 EA 稍早公佈的財報中,販售數量低於官方預期…
以下為支援 Ray Tracing 技術的遊戲:
。《Assetto Corsa Competizione》
。《Atomic Heart》
。《Battlefield V》
。《Control》
。《Enlisted》
。《Justice》
。《JX3》
。《MechWarrior 5: Mercenaries》
。《Metro Exodus》
。《ProjectDH》
。《Shadow of the Tomb Raider》
在這次測試中,我們選擇 Intel Core i9-9900KF 與 ASUS ROG Maximus XI Hero(WI-FI)作為測試平台。
Graphics Device | EVGA GeForce RTX 2070 FTW3 ULTRA GAMING | AMD RADEON VEGA VII | EVGA GeForce RTX 2080 FTW3 ULTRA GAMING | ||
---|---|---|---|---|---|
Graphics Driver | 417.71 | PRESS Edition | 417.71 | ||
Processor Model | Intel Core i9-9900KF | Intel Core i9-9900KF | Intel Core i9-9900KF | ||
Processor Family | Coffee Lake Refresh | Coffee Lake Refresh | Coffee Lake Refresh | ||
Base Frequency (GHz) | 3.6 | 3.6 | 3.6 | ||
Boost Frequency (GHz) | 5 | 5 | 5 | ||
Physical Cores | 8 | 8 | 8 | ||
Logical Processors | 16 | 16 | 16 | ||
TDP (W) | 95 | 95 | 95 | ||
Cooling Solution | Noctua NH-U12S | Noctua NH-U12S | Noctua NH-U12S | ||
Motherboard Model | ASUS | ASUS | ASUS | ||
Motherboard Vendor | ROG MAXIMUS XI HERO (WI-FI) | ROG MAXIMUS XI HERO (WI-FI) | ROG MAXIMUS XI HERO (WI-FI) | ||
Chipset | Intel Z390 | Intel Z390 | Intel Z390 | ||
Chipset Driver | 25.20.15015.2003 | 25.20.15015.2003 | 25.20.15015.2003 | ||
BIOS Version | 0702 | 0702 | 0702 | ||
BIOS Settings | By default | By default | By default | ||
Memory Quantity | 32GB ([4x] 16GB) | 32GB ([4x] 16GB) | 32GB ([4x] 16GB) | ||
Memory Configuration | Dual Channel | Dual Channel | Dual Channel | ||
Memory Speed | DDR4-2133 | DDR4-2133 | DDR4-2133 | ||
Memory Vendor | Corsair | Corsair | Corsair | ||
Memory Model | CMR64GX4M4C3466C16 | CMR64GX4M4C3466C16 | CMR64GX4M4C3466C16 | ||
Storage Device | Intel SSD 760p 256GB | Intel SSD 760p 256GB | Intel SSD 760p 256GB | ||
Storage Device 2 (Used in Second Disk IO benchmark Only) | OCZ TR150 480GB | OCZ TR150 480GB | OCZ TR150 480GB | ||
Display | PHILIPS 276E8VJSB/96 | PHILIPS 276E8VJSB/96 | PHILIPS 276E8VJSB/96 | ||
Resolution | 2560 x 1440 | 2560 x 1440 | 2560 x 1440 | ||
3DMark 11 Build 1.0.132 | |||||
Entry | |||||
Graphics Score | 56313 | -1.90% | 55244 | 21.83% | 67305 |
Physics Score | 17911 | -1.04% | 17724 | 0.50% | 17813 |
Combined Score | 17075 | -3.54% | 16470 | 4.83% | 17266 |
GT1 | 264.81 | -8.78% | 241.55 | 20.06% | 290 |
GT2 | 270.56 | 9.10% | 295.17 | 8.58% | 320.49 |
GT3 | 352.34 | 0.90% | 355.51 | 18.82% | 422.43 |
GT4 | 165.93 | -4.53% | 158.42 | 33.36% | 211.27 |
PT | 56.86 | -1.04% | 56.27 | 0.50% | 56.55 |
CT | 79.42 | -3.54% | 76.61 | 4.83% | 80.31 |
Performance | |||||
Graphics Score | 32177 | 20.41% | 38744 | 3.87% | 40242 |
Physics Score | 17857 | -1.36% | 17614 | 1.23% | 17831 |
Combined Score | 17379 | -3.08% | 16844 | 2.24% | 17221 |
GT1 | 160.71 | 5.88% | 170.16 | 12.45% | 191.34 |
GT2 | 162.05 | 14.50% | 185.55 | 4.81% | 194.48 |
GT3 | 198.2 | 22.66% | 243.12 | 1.90% | 247.74 |
GT4 | 89.67 | 33.31% | 119.54 | -1.08% | 118.25 |
PT | 56.69 | -1.36% | 55.92 | 1.23% | 56.61 |
CT | 80.84 | -3.08% | 78.35 | 2.23% | 80.1 |
Extreme | |||||
Graphics Score | 10548 | -3.61% | 10167 | 28.64% | 13079 |
Physics Score | 17895 | -1.57% | 17614 | 1.46% | 17871 |
Combined Score | 12706 | -6.53% | 11876 | 23.91% | 14715 |
GT1 | 58.24 | 1.20% | 58.94 | 18.00% | 69.55 |
GT2 | 54.7 | 0.73% | 55.1 | 18.40% | 65.24 |
GT3 | 55.01 | -12.56% | 48.1 | 41.21% | 67.92 |
GT4 | 29.77 | -2.86% | 28.92 | 33.44% | 38.59 |
PT | 56.81 | -1.57% | 55.92 | 1.45% | 56.73 |
CT | 59.1 | -6.53% | 55.24 | 23.90% | 68.44 |
3DMark Build 2.8.6427 | |||||
Fire Strike | |||||
Grpahics score | 23604 | 16.51% | 27500 | 0.25% | 27569 |
Graphics test 1 | 111.98 | 21.08% | 135.58 | -1.35% | 133.75 |
Graphics test 2 | 94.72 | 12.90% | 106.94 | 1.55% | 108.6 |
Physics score | 25595 | -0.14% | 25560 | -0.40% | 25458 |
Physics test | 81.26 | -0.14% | 81.15 | -0.41% | 80.82 |
Combined score | 11362 | -7.51% | 10509 | 9.27% | 11483 |
Combined test | 52.85 | -7.51% | 48.88 | 9.27% | 53.41 |
Fire Strike Extreme | |||||
Grpahics score | 11262 | 19.45% | 13452 | -1.69% | 13224 |
Graphics test 1 | 58.3 | 21.13% | 70.62 | -0.50% | 70.27 |
Graphics test 2 | 42.21 | 18.24% | 49.91 | -2.52% | 48.65 |
Physics score | 25701 | -1.30% | 25368 | 1.66% | 25788 |
Physics test | 81.59 | -1.30% | 80.53 | 1.66% | 81.87 |
Combined score | 5398 | -4.17% | 5173 | 25.32% | 6483 |
Combined test | 25.11 | -4.18% | 24.06 | 25.35% | 30.16 |
Fire Strike Ultra | |||||
Grpahics score | 5534 | 24.32% | 6880 | -6.74% | 6416 |
Graphics test 1 | 30.15 | 20.27% | 36.26 | 1.08% | 36.65 |
Graphics test 2 | 20.02 | 27.17% | 25.46 | -11.55% | 22.52 |
Physics score | 25736 | -0.96% | 25490 | 0.62% | 25648 |
Physics test | 81.7 | -0.95% | 80.92 | 0.62% | 81.42 |
Combined score | 3001 | 2.37% | 3072 | 15.33% | 3543 |
Combined test | 13.96 | 2.36% | 14.29 | 15.33% | 16.48 |
Time Spy | |||||
Graphics score | 9353 | -5.60% | 8829 | 25.33% | 11065 |
Graphics test 1 | 60.17 | 2.53% | 61.69 | 14.69% | 70.75 |
Graphics test 2 | 54.25 | -11.89% | 47.8 | 35.02% | 64.54 |
CPU score | 10746 | -0.17% | 10728 | 1.09% | 10845 |
CPU test | 36.1 | -0.14% | 36.05 | 1.08% | 36.44 |
Time Spy Extreme | |||||
Graphics score | 4343 | -1.89% | 4261 | 18.80% | 5062 |
Graphics test 1 | 27.53 | 13.33% | 31.2 | 3.27% | 32.22 |
Graphics test 2 | 25.54 | -12.76% | 22.28 | 33.12% | 29.66 |
CPU score | 5120 | -0.08% | 5116 | -0.02% | 5115 |
Average simulation time per frame | 68.4 | 0.00% | 68.4 | 0.00% | 68.4 |
HITMAN, DX12 | |||||
1440P, Preset = Ultra, SMAA, Anisotropic 16X | |||||
Minimum FPS | 1.66 | 1.08 | 1.63 | ||
Average FPS | 41.99 | 13.74% | 47.76 | 10.39% | 52.72 |
Maximum FPS | 5208333.5 | 2554.15 | 5208333.5 | ||
Metro 2033 Redux, DX11 | |||||
1440P, Preset = Very High, SSAA, AF 16X, Tesselation = Very High, Physx OFF | |||||
Minimum FPS | 23.94 | 25.39 | 10.62 | ||
Average FPS | 55.33 | 10.25% | 61 | 14.21% | 69.67 |
Maximum FPS | 205.25 | 130.82 | 245.4 | ||
Metro:Last Light Redux, DX11 | |||||
1440P, Preset = Very High, SSAA, AF 16X, Tesselation = Very High, Physx OFF | |||||
Minimum FPS | 23.83 | 21.77 | 25.2 | ||
Average FPS | 61 | 11.48% | 68 | 9.31% | 74.33 |
Maximum FPS | 129.89 | 146.39 | 148.7 | ||
Shadow of the Tomb Raider, DX12 | |||||
1440P, Preset = Highest, SMAA | |||||
Minimum FPS | 58 | 64 | 71 | ||
Average FPS | 72 | 15.28% | 83 | 7.23% | 89 |
Maximum FPS | 107 | 128 | 130 | ||
95% FPS | 61 | 13.11% | 69 | 10.14% | 76 |
Sid Meier's Civilization VI, DX12 | |||||
1440P, Preset = Very High | |||||
Avg Frame Time | 9.649 | -4.28% | 10.062 | 19.04% | 8.146 |
99th Percentile | 13.238 | -0.74% | 13.336 | 18.63% | 10.851 |
Avg FPS | 103.6376827 | -4.10% | 99.38382031 | 23.52% | 122.7596366 |
99th Percentile FPS | 75.5401118 | -0.73% | 74.985003 | 22.90% | 92.15740485 |
Far Cry 5, DX11 | |||||
1440P, Preset = Very High | |||||
Minimum FPS | 78 | 86 | 94 | ||
Average FPS | 89 | 8.99% | 97 | 9.28% | 106 |
Maximum FPS | 102 | 117 | 123 | ||
For Honor, DX11 | |||||
1440P, Preset = Very High | |||||
Minimum FPS | 74.14 | 86.54 | 53.03 | ||
Average FPS | 101.24 | 11.99% | 113.38 | 6.76% | 121.04 |
Maximum FPS | 128.19 | 140.67 | 152.03 | ||
Tom Clancy's Ghost Recon Wildlands, DX11 | |||||
1440P, Preset = Very High | |||||
Minimum FPS | 46 | 47.38 | 54.62 | ||
Average FPS | 53.68 | 5.38% | 56.57 | 10.48% | 62.5 |
Maximum FPS | 60.34 | 65.67 | 70.51 | ||
Tom Clancy's Rainbow Six Siege, DX11 | |||||
1440P, Preset = Very High | |||||
Overall Min FPS | 50.5 | 92.6 | 29.8 | ||
Overall Average FPS | 185 | 10.16% | 203.8 | 8.34% | 220.8 |
Overall Maximum FPS | 225.8 | 306.7 | 348.4 | ||
Tom Clancy's The Division, DX12 | |||||
1440P, Preset = Very High | |||||
Avg FPS | 79.4 | 26.20% | 100.2 | -9.28% | 90.9 |
Typical FPS | 80 | 26.25% | 101 | -9.41% | 91.5 |
Summary | 4.44% | 9.57% |
在這次測試中,我們選擇 Intel Core i9-9900KF 與 ASUS ROG Maximus XI Hero(WI-FI)作為測試平台。
就目前 AMD 釋出的驅動,在我們測試條件(2560 x 1440)下,AMD Radeon VII 遊戲表現基本上是落後 EVGA GeForce RTX 2080 FTW3 ULTRA GAMING,但與 EVGA GeForce RTX 2070 FTW3 ULTRA GAMING 相比的話,在很多遊戲下是具有一定贏面。
3DMark 11 部分,EVGA GeForce RTX 2080 FTW3 ULTRA GAMING 最高可以領先 28%,但是在 Performance 模式下,GeForce RTX 2080 領先 AMD Radeon VII 僅僅只有 3%。
只是來到 3DMark,AMD Radeon VII 只有在 Time Spy 和 Time Spy Extreme 模式下大幅落後 EVGA GeForce RTX 2080 FTW3 ULTRA GAMING。
遊戲方面,AMD Radeon VII 除了在 DirectX 12 的 Sid Meier’s Civilization VI 落後 22.9% 外,其餘遊戲介於 7 ~ 14% 間。
對於 EVGA GeForce RTX 2070 FTW3 ULTRA GAMING 落後與 AMD Radeon VII 並沒有太大意外,因為當初 CES 的 AMD Keynote 上對手是建議售價為 799 美元的 NVIDIA GeForce RTX 2080 而非建議售價 599 美元的 NVIDIA GeForce RTX 2070 顯示卡。
從 4GB HBM 記憶體的 Fury X 到 8GB HBM2 記憶體的 Radeon RX Vega 64 我們存在著很多幻想,現在 7nm 製程與 16GB HBM2 記憶體的 AMD Radeon VII 登場,似乎讓我們見到紅軍準備迎戰綠軍的一些影子。
以 AMD Radeon VII 的 699 美元建議售價來說,它有著一定的吸引力在,畢竟它不單單是為了遊戲,更可以用在影片剪輯和 3D 渲染上。沒有 Real Time Ray Tracing 技術是否很重要,也許是,但不會在這個世代的產品中抱著遺憾。
未來 AMD 顯示卡驅動陸續釋出更新,也許 Vega 20 架構的 AMD Radeon VII 會有著更佳,或是超越 NVIDIA GeForce RTX 2080 的遊戲表現。