Officially announced – at long last! – in a livestreamed event last week, AMD's RDNA 4 architecture is here with the RX 9070 XT leading the charge. Landing with a $599 MSRP, this GPU is firmly targeting Nvidia's RTX 5070 Ti, with AMD promising compelling rasterisation performance alongside huge improvements to ray tracing. FSR 4 also touts significant image quality improvements thanks to its new ML-powered algorithm, so on paper AMD looks to have ticked all the boxes. But how does that translate into the real world experience? We find out today…

Timestamps:

00:00 Intro
00:41 Recap and spec overview
01:48 Test setup
03:02 Alan Wake 2
03:41 Black Myth: Wukong
04:11 Cyberpunk 2077
04:14 Final Fantasy XVI
05:14 Forza Horizon 5
05:47 Ghost of Tsushima
06:21 Horizon Forbidden West
06:54 The Last of Us Part 1
07:20 Plague Tale: Requiem
07:42 Senua’s Saga: Hellblade 2
08:09 Starfield
08:36 Total War: Warhammer III
08:55 12-game average results
10:57 Cost per frame analysis
11:55 RT Alan Wake II
12:28 RT Black Myth: Wukong
12:57 RT Cyberpunk 2077
13:19 RT F1 24
13:40 RT Ratchet & Clank
14:00 RT Returnal
14:12 RT Shadow of the Tomb Raider
14:23 RT Star Wars Outlaws
14:43 RT 8-game average
15:18 FSR 4 analysis
18:56 Meet the Sapphire Pulse & ASRock Taichi
20:19 Thermals and acoustics
21:24 Power draw and efficiency
23:18 Closing thoughts

In lieu of an official AMD reference (or MBA) card, today we are using the ASRock RX 9070 XT Taichi for our testing. This card does come factory overclocked when using the default Gaming BIOS, but I did all my testing with the Quiet BIOS that runs at reference specifications.

It's also worth noting that while this is a standalone review for the RX 9070 XT, we do have a day one review for the RX 9070 (non-XT), and you can find that HERE.

	RX 9070 XT	RX 9070	RX 7900 GRE	RX 7800 XT	RX 7700 XT
Architecture	RDNA 4	RDNA 4	RDNA 3	RDNA 3	RDNA 3
Manufacturing Process	TSMC N4	TSMC N4	TSMC N5 GCD + N6 MCD	TSMC N5 GCD + N6 MCD	TSMC N5 GCD + N6 MCD
Transistor Count	53.9 billion	53.9 billion	57.7 billion	28.1 billion	28.1 billion
Die Size	357 mm²	357 mm²	300 mm² GCD 220 mm² MCD	200 mm² GCD 150 mm² MCD	200 mm² GCD 150 mm² MCD
Compute Units	64	56	80	60	54
Ray Accelerators	64	56	80	60	54
Stream Processors	4096	3584	5120	3840	3456
Game GPU Clock	2400 MHz	2070 MHz	1880 MHz	2124 MHz	2171 MHz
Boost GPU Clock	Up to 2970 MHz	Up to 2520 MHz	Up to 2245 MHz	Up to 2430 MHz	Up to 2544 MHz
ROPs	128	128	192	96	96
AMD Infinity Cache	64MB	64MB	64MB	64MB	48MB
Memory	16GB GDDR6	16GB GDDR6	16GB GDDR6	16GB GDDR6	12GB GDDR6
Memory Data Rate	20 Gbps	20 Gbps	18 Gbps	19.5 Gbps	18 Gbps
Memory Bandwidth	645 GB/s	645 GB/s	576 GB/s	624 GB/s	432 GB/s
Memory Interface	256-bit	256-bit	256-bit	256-bit	192-bit
Board Power	304W	220W	260 W	263W	245W

First, let's take a quick look at the specs. The RX 9070 XT is built on the new Navi 48 die, measuring 357mm², and it's worth pointing out this is a monolithic chip, so AMD is not using a chiplet-based design as per RDNA 3. In total, Navi 48 silicon packs in 53.9 billion transistors.

As a full implementation of Navi 48, the RX 9070 XT packs in 64 Compute Units, and each CU houses 64 Steam Processors, for a total of 4096 shaders. There's also 64 Ray Accelerators – one per CU – and 128 ROPs.

As for clock speed, the RX 9070 XT runs notably faster than its sibling, the RX 9070, given it sports a rated game clock of 2400MHz and a boost clock of up to 2970MHz.

The memory configuration is the same between both GPUs though, with 16GB of GDDR6 memory clocked at 20Gbps, operating over a 256-bit memory interface, for total memory bandwidth of 645 GB/s. 64MB of Infinity Cache is also present.

Power draw for the RX 9070 XT rated at 304W Total Board Power (TBP), but we are using our updated GPU power testing methodology in this review, so read on for our most detailed power and efficiency testing yet.

The post AMD RX 9070 XT Review ft. ASRock first appeared on KitGuru.

Launching alongside the RX 9070 XT, today we also have a full review of the slightly cheaper RX 9070 (non-XT). This cut-down Navi 48 GPU targets a $549 MSRP and, as such, is going head-to-head with Nvidia's RTX 5070 which we reviewed only yesterday. But with just a $50 gap between the RX 9070 and the XT model, does AMD's own product positioning make sense? All that and more is explored in this analysis.

Timestamps:

00:00 Intro
00:41 Recap and spec overview
01:48 Test setup
03:02 Alan Wake 2
03:41 Black Myth: Wukong
04:11 Cyberpunk 2077
04:14 Final Fantasy XVI
05:14 Forza Horizon 5
05:47 Ghost of Tsushima
06:21 Horizon Forbidden West
06:54 The Last of Us Part 1
07:20 Plague Tale: Requiem
07:42 Senua’s Saga: Hellblade 2
08:09 Starfield
08:36 Total War: Warhammer III
08:55 12-game average results
10:57 Cost per frame analysis
11:55 RT Alan Wake II
12:28 RT Black Myth: Wukong
12:57 RT Cyberpunk 2077
13:19 RT F1 24
13:40 RT Ratchet & Clank
14:00 RT Returnal
14:12 RT Shadow of the Tomb Raider
14:23 RT Star Wars Outlaws
14:43 RT 8-game average
15:18 FSR 4 analysis
18:56 Meet the Sapphire Pulse & ASRock Taichi
20:19 Thermals and acoustics
21:24 Power draw and efficiency
23:18 Closing thoughts

In lieu of an official AMD reference (or MBA) card, today we are using the Sapphire RX 9070 Pulse for our testing. This card still comes at reference specifications, however, while there's a range of factory overclocked models set to launch at retail tomorrow.

It's also worth noting that while this is a standalone review for the RX 9070, we do have a day one review for the RX 9070 XT, and you can find that HERE.

	RX 9070 XT	RX 9070	RX 7900 GRE	RX 7800 XT	RX 7700 XT
Architecture	RDNA 4	RDNA 4	RDNA 3	RDNA 3	RDNA 3
Manufacturing Process	TSMC N4	TSMC N4	TSMC N5 GCD + N6 MCD	TSMC N5 GCD + N6 MCD	TSMC N5 GCD + N6 MCD
Transistor Count	53.9 billion	53.9 billion	57.7 billion	28.1 billion	28.1 billion
Die Size	357 mm²	357 mm²	300 mm² GCD 220 mm² MCD	200 mm² GCD 150 mm² MCD	200 mm² GCD 150 mm² MCD
Compute Units	64	56	80	60	54
Ray Accelerators	64	56	80	60	54
Stream Processors	4096	3584	5120	3840	3456
Game GPU Clock	2400 MHz	2070 MHz	1880 MHz	2124 MHz	2171 MHz
Boost GPU Clock	Up to 2970 MHz	Up to 2520 MHz	Up to 2245 MHz	Up to 2430 MHz	Up to 2544 MHz
ROPs	128	128	192	96	96
AMD Infinity Cache	64MB	64MB	64MB	64MB	48MB
Memory	16GB GDDR6	16GB GDDR6	16GB GDDR6	16GB GDDR6	12GB GDDR6
Memory Data Rate	20 Gbps	20 Gbps	18 Gbps	19.5 Gbps	18 Gbps
Memory Bandwidth	645 GB/s	645 GB/s	576 GB/s	624 GB/s	432 GB/s
Memory Interface	256-bit	256-bit	256-bit	256-bit	192-bit
Board Power	304W	220W	260 W	263W	245W

First, let's take a quick look at the specs. The RX 9070 is built on the new Navi 48 die, measuring 357mm², and it's worth pointing out this is a monolithic chip, so AMD is not using a chiplet-based design as per RDNA 3. In total, Navi 48 silicon packs in 53.9 billion transistors.

While a full Navi 48 GPU packs in 64 Compute Units, the 9070 is cut down to 56 CUs, and each CU houses 64 Steam Processors, for a total of 3584 shaders. There's also 56 Ray Accelerators – one per CU – and 128 ROPs.

As for clock speed, the RX 9070 runs notably slower than its sibling, the RX 9070 XT, and instead it sports a rated game clock of 2070MHz and a boost clock of up to 2520MHz.

The memory configuration is the same between both GPUs though, with 16GB of GDDR6 memory clocked at 20Gbps, operating over a 256-bit memory interface, for total memory bandwidth of 645 GB/s. 64MB of Infinity Cache is also present.

Power draw for the RX 9070 rated at 220W Total Board Power (TBP), but we are using our updated GPU power testing methodology in this review, so read on for our most detailed power and efficiency testing yet.

The post AMD RX 9070 Review ft. Sapphire first appeared on KitGuru.

Following on from the RTX 5070 Ti that hit the market last month, today we can present our review of Nvidia's new RTX 5070. The smallest Blackwell GPU yet, it's also the cheapest xx70-class SKU since the launch of the RTX 3070 back in October 2020, given it lands with a £539/$549 MSRP. On-paper specs don't suggest much of a leap forward compared to the RTX 4070 Super, though, so just how fast is this new graphics card? We look at rasterisation, ray tracing, DLSS 4 and more to find out, while also putting Nvidia's claim of ‘RTX 4090' performance to the test…

With the retail launch for the RTX 5070 set for tomorrow, March 5th, it will be the last of the four RTX 50 series GPUs that Nvidia announced back at CES 2025. RTX 5090 was first out in mid-January, while RTX 5080 followed a week later, and the aforementioned RTX 5070 Ti landed in February.

This review is all about the RTX 5070 though, and we have plenty of data and charts to look through before delivering our final verdict on this graphics card…

	RTX 5080	RTX 5070 Ti	RTX 5070	RTX 4070 Super	RTX 4070
Process	TSMC N4	TSMC N4	TSMC N4	TSMC N4	TSMC N4
SMs	84	70	48	56	46
CUDA Cores	10752	8960	6144	7168	5888
Tensor Cores	336	280	192	224	184
RT Cores	84	70	48	56	46
Texture Units	336	280	192	224	184
ROPs	112	96	80	80	64
GPU Boost Clock	2617 MHz	2452 MHz	2512 MHz	2475 MHz	2475 MHz
Memory Data Rate	30 Gbps	28 Gbps	28 Gbps	21 Gbps	21 Gbps
L2 Cache	65536 KB	49152 KB	49152 KB	49152 KB	36864 KB
Total Video Memory	16GB GDDR7	16GB GDDR7	12GB GDDR7	12GB GDDR6X	12GB GDDR6X
Memory Interface	256-bit	256-bit	192-bit	192-bit	192-bit
Memory Bandwidth	960 GB/Sec	896 GB/Sec	672 GB/Sec	504 GB/Sec	504 GB/Sec
TGP	360W	300W	250W	220W	200W

First, a quick spec recap. RTX 5070 marks the introduction of a new Blackwell die – GB205, where RTX 5070 is not quite full implementation but comes close. Comprised of five Graphics Processing Clusters (GPCs), each holds up to eight Texture Processing Clusters (TPCs), with a total of 24. Each TPC is home to two Streaming Multiprocessors (SMs), giving us 48, and each SM still holds 128 CUDA Cores, meaning the RTX 5070 has a total of 6144 shaders. We also find 48 RT cores, 192 Tensor cores, 192 Texture Units, and 80 ROPs.

This time around, however, there's no node-shrink, and GB205 remains fabricated on TSMC's N4 node, as per the RTX 40 series. As such, rated clock speed has not stepped forwards this generation, with the RTX 5070 rated at 2512MHz, within 40MHz of the RTX 4070 and RTX 4070 Super.

The memory configuration is similar to its predecessors, too. We still find 12GB of memory operating a 192-bit interface, the primary difference being the jump to 28Gbps GDDR7 modules, increasing memory bandwidth to 672 GB/s. L2 cache still comes in at 49MB.

Lastly, the RTX 5070 features a 259W TGP. This is something we focus on closely in this review, using our enhanced GPU power testing methodology, so read on for our most detailed power and efficiency testing yet.

The post Nvidia RTX 5070 Review: RTX 4090 Performance? first appeared on KitGuru.