What is the difference between RTX 50 and RTX 40 series architectures?

RTX 50 series (Blackwell) and RTX 40 series (Ada Lovelace) are consecutive NVIDIA GPU architectures. The primary differences relevant to hardware enthusiasts: Blackwell uses GDDR7 memory across the lineup (versus Ada's GDDR6X on flagship models), delivering significantly higher memory bandwidth. Blackwell introduces an enhanced firmware signing mechanism that restricts unsigned BIOS modifications. The shader execution architecture in Blackwell adds FP4 tensor operations and improved 5th-generation RT cores. Ada's rasterization and shader performance scales roughly 40–60% from similar Ada to Blackwell tier for the same product position (RTX 4090 to RTX 5090).

What does TDP mean for a GPU and is it the same as power consumption?

TDP (Thermal Design Power) is the maximum sustained power dissipation the GPU manufacturer and AIB partner have certified the cooling solution to handle. In practice, GPU TDP (also called board power or total board power) is typically measured at the PCI-e slot plus external power connectors during a demanding sustained workload. Peak power during brief transients (GPU Boost 4.0's tau-second peaks) can exceed the TDP value by 20–30% for short periods without triggering thermal throttling. The TDP value listed in GPU specifications reflects the sustained power limit and correlates directly with the power limit ceiling that BIOS modification can raise.

What is the difference between Navi 48 and Navi 44 in the RX 9000 series?

Navi 48 and Navi 44 are the two GPU dies in AMD's RDNA 4 generation. Navi 48 is the larger die, powering the RX 9070 XT and RX 9070 — it features 32 compute units at the full configuration and uses a 256-bit memory bus with 16 GB GDDR6. Navi 44 is the smaller die for the RX 9060 XT and below — fewer compute units, 128-bit memory bus, and lower TDP, targeting the volume mainstream segment. From an architectural standpoint both dice share the RDNA 4 shader design with improved AI acceleration (matrix acceleration units) and 4th-generation ray accelerators.

What is the SubVendor ID and how is it used in GPU BIOS management?

The SubVendor ID (also SubSystem Vendor ID) is a 16-bit identifier embedded in both the GPU BIOS and the card's PCI-e configuration space that identifies the AIB manufacturer. ASUS = 0x1043, MSI = 0x1462, Gigabyte = 0x1458, EVGA = 0x3842, Sapphire = 0x1DA2, PowerColor = 0x148C, XFX = 0x1682. Flash tools use SubVendor ID as a secondary compatibility check — after device ID validation, the SubVendor ID is verified to prevent applying a BIOS intended for a different AIB's PCB design. Cross-flashing between AIB variants requires overriding this check (nvflash -6, ATIFlash -i), which is legitimate when PCB layouts are confirmed identical.

What is the performance hierarchy of current AMD RDNA 4 cards?

The RDNA 4 lineup as of mid-2026: RX 9070 XT (Navi 48, 4096 shaders, 304 W TDP) is the flagship consumer card; RX 9070 (Navi 48, 3584 shaders, 220 W TDP) occupies the upper-mid segment; RX 9060 XT (Navi 44, 2048 shaders, 150 W TDP) is the mainstream entry. AMD positioned RDNA 4 primarily at the mid-range and upper-mid segments, choosing not to release a RDNA 4 flagship competing directly with the RTX 5090. In practical GPU-limited rendering benchmarks at 1440p, the RX 9070 XT trades performance with the RTX 5070 and RTX 5070 Ti.

Can I use the GPU database to find a compatible BIOS for my card?

The GPU database provides the device IDs and SubVendor IDs needed to search BIOS repositories. With the device ID confirmed from this database (and verified against GPU-Z on the installed card), navigate to the BIOS Repository to find matching entries. Always verify that the BIOS you select matches both the device ID AND SubVendor ID of the installed card, and that the VRAM configuration (capacity and chip vendor) is listed as compatible in the BIOS repository entry.

What is an AIB partner GPU vs a reference design?

An AIB (Add-in Board) partner is a manufacturer that takes a GPU die from NVIDIA or AMD and designs their own printed circuit board, power delivery, and cooling solution around it. AIB partners include ASUS (ROG Strix, TUF), MSI (Gaming X, Ventus), Gigabyte (AORUS, Eagle), Sapphire (Nitro+, Pulse), PowerColor (Red Devil, Fighter), XFX (Speedster), and ASRock (Taichi, Phantom Gaming). A reference design is the board designed by NVIDIA (Founders Edition) or AMD directly. AIB BIOSes typically differ from reference in power limits, fan curves, factory overclock tables, and the SubVendor ID embedded in the firmware.

GPU Database 2026: RTX 50, RTX 40, RX 9000 & RX 7000

Q: What is a GPU device ID and why does it matter for BIOS flashing?

A GPU device ID is a 16-bit hexadecimal identifier assigned by the GPU manufacturer (NVIDIA or AMD) that uniquely identifies the GPU die configuration — the silicon revision, shader count, and memory interface width. BIOS flash tools (nvflash, ATIFlash, VBFlash) validate the target BIOS file's embedded device ID against the installed card before writing. If the device IDs do not match, the flash is rejected. This prevents accidentally applying an incompatible BIOS from a different GPU die, which could brick the card. The device ID is displayed in GPU-Z, Windows Device Manager, and output by the flash tools themselves.

Q: What is GDDR7 and which GPUs use it?

GDDR7 is the seventh generation of Graphics Double Data Rate memory, ratified by JEDEC in 2022 and first deployed in consumer GPUs in 2025. NVIDIA's RTX 50 series (Blackwell) uses GDDR7 across the entire lineup. AMD's RX 9000 series (RDNA 4, based on Navi 48 and Navi 44 dies) uses GDDR6 rather than GDDR7 — AMD positioned RX 9000 for cost efficiency and GDDR6 was the appropriate choice for the targeted price segment. GDDR7 offers approximately 2× the peak bandwidth of GDDR6 at equivalent bus widths, while GDDR6X (used in RTX 40 flagship cards) offered approximately 1.5× GDDR6 bandwidth through PAM4 signaling.

Q: How do I check which GPU die revision my card uses?

GPU-Z is the most reliable consumer tool for identifying die revision. The 'GPU' field shows the die name (e.g., AD102, GB203, Navi 48). The 'Revision' field shows the silicon stepping — A1, B0, etc. — which can differ between manufacturing runs of the same product. For BIOS flashing purposes, the device ID and SubVendor ID (also shown in GPU-Z) are more directly relevant than the die revision name, since BIOS compatibility is validated against these identifiers rather than the marketing name or die revision label.

Q: What is the memory bus width and why does it matter?

Memory bus width is the number of data lines between the GPU's memory controllers and the VRAM chips, measured in bits. A wider bus enables more data to be transferred per clock cycle. Total memory bandwidth = bus width × effective memory clock × 2 (for DDR memory). A 256-bit GDDR7 bus at 28 Gbps per pin delivers 896 GB/s, while a 128-bit GDDR6 bus at 18 Gbps per pin delivers 288 GB/s. Bus width is fixed by the GPU die design — it cannot be changed via BIOS modification or software overclocking.

Q: How many CUDA cores does the RTX 5090 have?

The GeForce RTX 5090 (GB202 die) has 21,760 CUDA cores (shader processors). This is organized as 170 Streaming Multiprocessors (SMs) × 128 CUDA cores per SM. The RTX 5090 represents a roughly 33% increase over the RTX 4090's 16,384 CUDA cores (128 SMs × 128 CUDAs), achieved through a larger die (GB202 vs AD102) and architectural improvements in the SM design.

Model	Die	Shader Units	TMUs / ROPs	VRAM	Bus Width	Bandwidth	TDP	Base / Boost	Device ID	Flash Tool
GeForce RTX 5090	GB202	21,760	680 / 176	32 GB GDDR7	512-bit	1,792 GB/s	575 W	2,017 / 2,407 MHz	0x2684	nvflash 5.82x
GeForce RTX 5080	GB203	10,752	336 / 112	16 GB GDDR7	256-bit	960 GB/s	360 W	2,295 / 2,617 MHz	0x2685	nvflash 5.82x
GeForce RTX 5070 Ti	GB203	8,960	280 / 96	16 GB GDDR7	256-bit	896 GB/s	300 W	2,295 / 2,452 MHz	0x2686	nvflash 5.82x
GeForce RTX 5070	GB205	6,144	192 / 80	12 GB GDDR7	192-bit	672 GB/s	250 W	2,160 / 2,512 MHz	0x2704	nvflash 5.82x
GeForce RTX 5060 Ti	GB206	4,608	144 / 64	16 GB GDDR7	128-bit	448 GB/s	180 W	2,310 / 2,572 MHz	0x2770	nvflash 5.82x
GeForce RTX 5060	GB206	3,840	120 / 48	8 GB GDDR7	128-bit	448 GB/s	150 W	1,980 / 2,497 MHz	0x2782	nvflash 5.82x
GeForce RTX 5050	GB207	2,560	80 / 32	8 GB GDDR7	128-bit	392 GB/s	130 W	1,755 / 2,310 MHz	0x2803	nvflash 5.82x

Model	Die	Shader Units	TMUs / ROPs	VRAM	Bus Width	Bandwidth	TDP	Base / Boost	Device ID	Flash Tool
GeForce RTX 4090	AD102	16,384	512 / 176	24 GB GDDR6X	384-bit	1,008 GB/s	450 W	2,235 / 2,520 MHz	0x2684	nvflash 5.80x+
GeForce RTX 4080 Super	AD103	10,240	320 / 112	16 GB GDDR6X	256-bit	736 GB/s	320 W	2,295 / 2,550 MHz	0x2702	nvflash 5.80x+
GeForce RTX 4080	AD103	9,728	304 / 112	16 GB GDDR6X	256-bit	716 GB/s	320 W	2,205 / 2,505 MHz	0x2704	nvflash 5.80x+
GeForce RTX 4070 Ti Super	AD103	8,448	264 / 96	16 GB GDDR6X	256-bit	672 GB/s	285 W	2,340 / 2,610 MHz	0x2705	nvflash 5.80x+
GeForce RTX 4070 Ti	AD104	7,680	240 / 80	12 GB GDDR6X	192-bit	504 GB/s	285 W	2,310 / 2,610 MHz	0x2782	nvflash 5.80x+
GeForce RTX 4070 Super	AD104	7,168	224 / 80	12 GB GDDR6X	192-bit	504 GB/s	220 W	1,980 / 2,475 MHz	0x2783	nvflash 5.80x+
GeForce RTX 4070	AD104	5,888	184 / 64	12 GB GDDR6X	192-bit	504 GB/s	200 W	1,920 / 2,475 MHz	0x2786	nvflash 5.80x+
GeForce RTX 4060 Ti	AD106	4,352	136 / 48	16 GB GDDR6	128-bit	288 GB/s	165 W	2,310 / 2,535 MHz	0x2803	nvflash 5.80x+
GeForce RTX 4060	AD107	3,072	96 / 48	8 GB GDDR6	128-bit	272 GB/s	115 W	1,830 / 2,460 MHz	0x2882	nvflash 5.80x+

Model	Die	CUs	Stream Procs	VRAM	Bus Width	Bandwidth	TDP	Game / Boost Clock	Device ID	Flash Tool
Radeon RX 9070 XT	Navi 48	32	4,096	16 GB GDDR6	256-bit	576 GB/s	304 W	2,970 / 3,300 MHz	0x7446	VBFlash 2025+
Radeon RX 9070	Navi 48	28	3,584	16 GB GDDR6	256-bit	576 GB/s	220 W	2,070 / 2,520 MHz	0x7447	VBFlash 2025+
Radeon RX 9060 XT	Navi 44	16	2,048	16 GB GDDR6	128-bit	288 GB/s	150 W	2,310 / 2,835 MHz	0x7448	VBFlash 2025+
Radeon RX 9060	Navi 44	14	1,792	8 GB GDDR6	128-bit	288 GB/s	115 W	2,160 / 2,655 MHz	0x7449	VBFlash 2025+

Model	Die	CUs	Stream Procs	VRAM	Bus Width	Bandwidth	TDP	Game / Boost Clock	Device ID	Flash Tool
Radeon RX 7900 XTX	Navi 31 XT	96	12,288	24 GB GDDR6	384-bit	960 GB/s	355 W	2,300 / 2,615 MHz	0x744C	VBFlash / ATIFlash
Radeon RX 7900 XT	Navi 31 XL	84	10,752	20 GB GDDR6	320-bit	800 GB/s	315 W	2,000 / 2,394 MHz	0x744E	VBFlash / ATIFlash
Radeon RX 7900 GRE	Navi 31	80	5,120	16 GB GDDR6	256-bit	576 GB/s	260 W	1,880 / 2,245 MHz	0x744C	VBFlash / ATIFlash
Radeon RX 7800 XT	Navi 32	60	3,840	16 GB GDDR6	256-bit	624 GB/s	263 W	2,124 / 2,430 MHz	0x747E	VBFlash / ATIFlash
Radeon RX 7700 XT	Navi 32	54	3,456	12 GB GDDR6	192-bit	432 GB/s	245 W	2,171 / 2,544 MHz	0x7480	VBFlash / ATIFlash
Radeon RX 7600	Navi 33	32	2,048	8 GB GDDR6	128-bit	288 GB/s	165 W	2,250 / 2,655 MHz	0x7480	VBFlash / ATIFlash
Radeon RX 7600 XT	Navi 33	32	2,048	16 GB GDDR6	128-bit	288 GB/s	190 W	2,295 / 2,755 MHz	0x7481	VBFlash / ATIFlash

How to Read This Database for BIOS Flashing

The specifications above are organized to serve the GPU hardware enthusiast's primary workflow: identifying cards, locating compatible firmware, and verifying flash tool compatibility. The Device ID column is the most critical for firmware operations — it is the hexadecimal identifier that flash tools match against the BIOS file being applied.

Before any flash operation, run nvflash --version (NVIDIA) or ATIFlash -i / VBFlash -i (AMD) to get the installed card's exact device ID and SubVendor ID. Match both against the BIOS repository entry, not just the product name. A "GeForce RTX 4090" BIOS from ASUS (SubVendor 0x1043) is not the same file as an "RTX 4090" BIOS from MSI (SubVendor 0x1462), even though both have device ID 0x2684.

Architecture Notes — 2026 Generation

NVIDIA Blackwell (RTX 50 Series)

Blackwell is NVIDIA's 4 nm (TSMC N4P) GPU architecture introduced in late 2024/early 2025. The flagship GB202 die achieves 21,760 CUDA cores — the largest NVIDIA consumer die ever manufactured. Key architectural changes from Ada Lovelace: GDDR7 memory across the lineup replacing GDDR6X (previous flagships); 5th-generation RT cores with improved incoherent ray traversal performance; FP4 tensor operations for accelerated AI inference at 4-bit precision; and enhanced Tensor Memory Accelerator (TMA) for improved memory access patterns in AI workloads.

From a firmware modding standpoint, Blackwell's most significant change is the enhanced firmware signing that restricts unsigned BIOS modifications. Standard BIOS replacement between same-model BIOSes (ASUS RTX 5090 stock BIOS to ASUS RTX 5090 updated stock BIOS) functions normally via nvflash 5.82x. Custom BIOS modifications involving power table edits or cross-flash operations face increased friction compared to Ada and earlier generations.

AMD RDNA 4 (RX 9000 Series)

RDNA 4 introduces AMD's 4th-generation ray accelerators (double the RDNA 3 capability per CU), new matrix acceleration units for AI workloads, and a redesigned shader processor that improves performance-per-CU versus RDNA 3. The Navi 48 die, used in RX 9070 XT and RX 9070, represents AMD's high-performance mainstream positioning — competitive with the RTX 5070 tier at launch pricing.

RDNA 4 uses GDDR6 rather than GDDR7. AMD's decision was driven by cost optimization for the targeted price band and by GDDR6's maturity at the 256-bit and 128-bit bus widths used in the Navi 48 and Navi 44 designs. The 576 GB/s bandwidth on the RX 9070 XT's 256-bit GDDR6-18 interface is directly comparable to the RX 6900 XT's bandwidth from three years prior — RDNA 4 delivers its performance gains primarily through improved compute efficiency and ray tracing hardware, not memory bandwidth.

Historical Context: From MVKTech's Era to Blackwell

When this site operated as a GPU modding portal in the early 2000s, the GPU landscape was fundamentally different. The cards of that era — NVIDIA GeForce FX series, ATI Radeon 9000/9700/9800 — had shader pipelines with separate vertex and pixel shader units, measured in the dozens of execution units rather than thousands. Overclocking was done via registry edits and BIOS clock table manipulation. The community that developed around tools like NiBiTor, ATIFlash, and the early GPU-Z grew directly from the modding discussions hosted on sites like mvktech.net, Beyond3D, and Rage3D.

From the GeForce FX 5800 to the RTX 5090 is a factor of roughly 800× in shader throughput, 160× in memory bandwidth, and 1,400× in dedicated ray tracing performance. The BIOS flashing toolchain has evolved from simple clock table editors to complex firmware replacement utilities managing multi-partition UEFI images, signed firmware sections, and architecture-specific power delivery tables. The core practice — modifying the GPU's firmware to access hardware capabilities beyond factory settings — is the same discipline that motivated the early community.

Frequently Asked Questions

What is a GPU device ID and why does it matter for BIOS flashing?

A GPU device ID is a 16-bit hexadecimal identifier that uniquely identifies the GPU die configuration. Flash tools validate the target BIOS file's embedded device ID against the installed card before writing — if they do not match, the flash is rejected. The device ID is displayed in GPU-Z, Windows Device Manager, and by the flash tools themselves.

What is the difference between RTX 50 and RTX 40 architectures?

RTX 50 (Blackwell) uses GDDR7 memory, 5th-gen RT cores, FP4 tensor operations, and enhanced firmware signing that restricts BIOS modification. RTX 40 (Ada) uses GDDR6X on flagships, 3rd-gen RT cores, and less restrictive signing, making BIOS editing more accessible. Performance gains from RTX 40 to RTX 50 are approximately 30–50% at equivalent product tiers.

What is GDDR7 and which GPUs use it?

GDDR7 is the 7th-generation Graphics DDR memory standard offering approximately 2× the bandwidth of GDDR6 at equivalent bus widths. NVIDIA RTX 50 series (Blackwell) uses GDDR7 across the lineup. AMD RX 9000 series uses GDDR6 — AMD targeted cost efficiency at the RX 9000 price points, where GDDR7 would have increased BOM cost disproportionately.

How do I check which GPU die revision my card uses?

GPU-Z shows the die name (AD102, GB203, Navi 48) in the 'GPU' field and silicon stepping in 'Revision.' For BIOS operations, the device ID and SubVendor ID shown in GPU-Z are more directly relevant than the die revision name.

What does TDP mean and is it the same as power consumption?

TDP is the maximum sustained power dissipation the cooling solution is rated for. Peak instantaneous power during boost transients can exceed TDP by 20–30% for short periods. The listed TDP directly corresponds to the power limit ceiling that BIOS modification can raise — nvflash and ATIFlash BIOS modifications typically target the TDP table entries in the firmware.

What is the memory bus width and why does it matter?

Memory bus width (in bits) determines how many data lines exist between the GPU's memory controllers and VRAM chips. Total bandwidth = bus width × effective memory speed. A 256-bit GDDR7 bus at 28 Gbps per pin delivers 896 GB/s. Bus width is fixed by the die design and cannot be changed via BIOS or software.

What is the difference between Navi 48 and Navi 44?

Navi 48 (RX 9070 XT/9070): 32 compute units, 256-bit bus, 16 GB GDDR6, higher TDP. Navi 44 (RX 9060 XT/9060): fewer compute units, 128-bit bus, lower TDP, mainstream targeting. Both share RDNA 4 shader architecture with 4th-gen ray accelerators and matrix acceleration units.

What is the SubVendor ID?

SubVendor ID identifies the AIB manufacturer: ASUS 0x1043, MSI 0x1462, Gigabyte 0x1458, EVGA 0x3842, Sapphire 0x1DA2, PowerColor 0x148C, XFX 0x1682. Flash tools use it as a secondary compatibility check. Cross-flashing between AIB variants requires overriding this check (nvflash -6, ATIFlash -i).

How many CUDA cores does the RTX 5090 have?

21,760 CUDA cores (170 SMs × 128 CUDA cores per SM). This is approximately 33% more than the RTX 4090's 16,384 CUDA cores, achieved through both a larger die (GB202 vs AD102) and incremental SM design improvements.

What is the RDNA 4 performance hierarchy?

RX 9070 XT (flagship consumer, 304 W) → RX 9070 (upper-mid, 220 W) → RX 9060 XT (mainstream, 150 W) → RX 9060 (entry mainstream, 115 W). The RX 9070 XT trades performance with the RTX 5070 and 5070 Ti at 1440p in GPU-limited workloads.

Can I use the database to find a compatible BIOS for my card?

Use the device ID column as the primary filter when searching the BIOS Repository. Confirm the device ID against GPU-Z on the installed card, then match SubVendor ID to the AIB partner. Both must match the BIOS file's embedded identifiers for a valid compatible flash.

What is unified shader architecture?

Introduced in the mid-2000s (NVIDIA Tesla, AMD TeraScale), unified shaders replaced dedicated vertex and pixel shader pipelines with general-purpose ALUs schedulable for any workload type. The same execution units now handle vertex shading, pixel shading, compute, ray traversal, and AI inference — the architectural basis for GPU-accelerated computing.

What is the difference between GDDR6 and GDDR6X?

GDDR6X uses PAM4 signaling (4 voltage levels per symbol vs GDDR6's 2) to double data rate per pin without increasing physical clock speed. Up to 21 Gbps/pin for GDDR6X vs 16–18 Gbps/pin for GDDR6. Trade-off: higher per-die power. NVIDIA used GDDR6X on RTX 40 flagships; AMD used standard GDDR6 on RX 7000 with a wider 384-bit bus for comparable bandwidth.

What is a compute unit (CU) in AMD GPU architecture?

An AMD Compute Unit is the fundamental shader execution block. Each RDNA 2–4 CU contains 64 stream processors (2 × SIMD32). Total stream processors = CU count × 64. NVIDIA's equivalent is the Streaming Multiprocessor (SM), containing 128 CUDA cores per SM in Ada and Blackwell.

Why does the RTX 4090 have a 450 W TDP?

AD102 at NVIDIA's 2520 MHz boost target requires approximately 450 W sustained. NVIDIA implemented the 16-pin PCIe Gen 5 connector (12VHPWR, 600 W rated) to support this power level. AIB variants commonly enable 516–525 W via their BIOS — this is within the connector's rating and the VRM designs used on high-end AIB RTX 4090 cards.

What is the RX 9070 XT device ID?

PCI Device ID 0x7446 under AMD vendor 0x1002. SubVendor ID varies: Sapphire 0x1DA2, PowerColor 0x148C, ASRock 0x1849, Gigabyte 0x1458. Always confirm via GPU-Z before selecting a BIOS — power limits, fan profiles, and PCB-specific parameters differ between AIB implementations even with the same device ID.

What is an AIB partner GPU vs reference design?

AIB (Add-in Board) partners (ASUS, MSI, Gigabyte, Sapphire, PowerColor, XFX, ASRock) design their own PCBs, power delivery, and cooling solutions around GPU dies from NVIDIA or AMD. Founders Edition (NVIDIA) and AMD reference cards are the silicon manufacturer's own board designs. AIB BIOSes differ in power limits, fan curves, factory OC tables, and the SubVendor ID embedded in firmware.

GPU Specs Database 2026: RTX 50, RTX 40, RX 9000 & RX 7000