Radeon RX 5700 XT in AMD-Benchmarks: At least as fast as Nvidia's Geforce RTX 2070

ByAdeline Belluz 2019-06-11 594

AMD organized a Techday around Matisse (Ryzen 3000) and Navi (Radeon RX 5700) in the run-up to E3. In the aftermath of the Next Horizon gaming event, the trade press is allowed to report on it. Included: First architecture details of Navi compared to GCN as well as new benchmarks, which see the Radeon RX 5700 XT at least at eye level of Nvidia's Geforce RTX 2070.

The first two graphics cards from AMD's Navi generation are named Radeon RX 5700 XT and Radeon RX 5700. Both use the Navi-10 GPU, which covers the new upper class, but is positioned below Vega 20, which sits on the Radeon VII. First of all: Navi 10 is exactly 251 mm² in size and comprises 10.3 billion transistors. Remember: Vega 10 uses 12.5 billion transistors on 486 mm² in Globalfoundries' 14LPP process, Vega 20 13.2 billion on 331 mm². The latter GPU scales worse with TSMC's 7nm process because the fast PCI Express 4.0 and Infinity fabric controllers shrink poorly.

Radeon RX 5700 XT with full Navi 10 configuration

The fastest model, the Radeon RX 5700 XT, comes with a fully activated Navi-10 GPU. This means: 2,560 shader units organized within 40 compute units and a 256 bit wide memory interface designed for GDDR6 RAM. AMD uses 8 GiByte with 14 Gbps (7.000 MHz), which results in a bandwidth of 448 GB/s. On the GPU side, the chip manufacturer now speaks of a basic, game and boost clock. The explanation behind this is simple: The base clock is guaranteed under load and "normal" operating conditions, the boost value corresponds to the end of the flagpole. The game clock should be a realistic practical indication.

In the case of the Radeon RX 5700 XT, 9.75 TFLOPS FP32 computing power at up to 1,905 MHz GPU frequency is mentioned. The so-called Total Board Power (TBP) amounts to 225 watts. The Radeon RX 5700 uses a slightly reduced Navi 10 graphics chip with 36 compute units or 2,304 shader units. On the memory side it remains at 8 GiByte 14-Gbps-GDDR6. With a lower GPU clock rate of 1.465 to 1.725 MHz, AMD limits the TBP to 180 watts.

Modell

RX 5700 XT

RX 5700

Radeon VII

RX Vega 64

RX Vega 56

Codename/Confection

Navi 10 XT

Navi 10 XL

Vega 20 XT

Vega 10 XT

Vega 10 XL

Chip size (pure die)

251 mm²

251 mm²

331 mm²

486 mm²

486 mm²

Transistors Graphics chip (million)

10.300

10.300

13.200

12.500

12.500

manufacturing processes

7 nm FF

7 nm FF

7 nm FF

14 nm LPP

14 nm LPP

ALUs/SIMDs/TMUs/ROPs

2.560/40/160/64

2.304/36/144/64

3.840/60/240/64

4.096/64/256/64

3.584/64/224 /64

GPU base clock (MHz)

1.605

1.465

1.400

1.247

1.156

GPU boost clock (MHz)

1.755

1.625

1.750

1.546

1.471

FP32/FP64 performance (TFLOPS)

8,98/?

7,49/?

13,44/3,36

12,66/0,79

10.544/659

Size of the level 2 cache (KiB)

4.096

4.096

4.096

4.096

4.096

Memory connection (bit)

256

256

4.096

2.048

2.048

Speed RAM (GT/s)

14

14

2

1,89

1,6

storage type

GDDR6

GDDR6

HBM gen2

HBM gen2

HBM gen2

Memory transfer (GB/s)

448

448

1024

483,8

409,6

Storage capacity (MiB)

8.192

8.192

16.384

8.192

8.192

PCI-Express Power Connectors

Je 1 × 6- + 8-polig

2 × 6-polig

2 × 8-polig

2 × 8-polig

2 × 8-polig

Typical power consumption

225 Watt

180 Watt

300 Watt

295 Watt

210 Watt

RDNA to bring raw material better onto the road

Although Navi 10 has significantly less raw performance, graphics cards with the GPU should in practice fail significantly faster than their GCN-based predecessors, including the Radeon RX Vega 64 with 4,096 shaders. This is due to the RDNA architecture, which sees AMD as a fundamental innovation compared to the Graphics Core Next. At the Techday, the chip manufacturer discussed the details of the GCN successor in more detail.

Per Compute Unit it remains superficial with 64 FP32 shaders, which are however reorganized. Instead of four SIMD16 blocks, each CU contains two SIMD32 blocks. By doubling the number of schedulers, they can process either one Wave64 or two Wave32 instructions per clock, which can quadruple the workload by up to four times. GCN to Vega requires Wave64 formats that can be distributed over all four bars. In the Worstcase, one instruction feeds only one SIMD16 unit, while three quarters of a CU have no work - this is where GCN-specific optimizations by developers come in to tickle more performance from Vega and Co. Navi 10, on the other hand, still handles a double FP16 rate. Two compute units can be combined to form a workgroup processor, which can increase performance if they want to access the same data. AMD didn't comment on FP64 support, we wouldn't expect a high rate in view of the gaming focus.

 You may also want to read:
 Sony A7S III: release date, features
  Huawei P40 and P40 Pro are on the way
  Xiaomi Mi 9T and Mi 9T Pro release date and price

 

Gearbest Clearance sale

 

Gearbest online shopping on GearBest

Extensive Product Selection

● Over 300,000 products

● 20 different categories

● 15 local warehosues

● Multiple top brands

Convenient Payment

● Global payment options: Visa, MasterCard, American Express

● PayPal, Western Union and bank transfer are accepted

● Boleto Bancario via Ebanx (for Brazil)

Prompt Shipping

● Unregistered air mail

● Registered air mail

● Priority line

● Expedited shipping

Dedicated After-sales Service

● 45 day money back guarantee

● 365 day free repair warranty

● 7 day Dead on Arrival guarantee (DOA)

You might also like

HOT TOPICS

Related Products