Guru3D.com
  • HOME
  • NEWS
    • Channels
    • Archive
  • DOWNLOADS
    • New Downloads
    • Categories
    • Archive
  • GAME REVIEWS
  • ARTICLES
    • Rig of the Month
    • Join ROTM
    • PC Buyers Guide
    • Guru3D VGA Charts
    • Editorials
    • Dated content
  • HARDWARE REVIEWS
    • Videocards
    • Processors
    • Audio
    • Motherboards
    • Memory and Flash
    • SSD Storage
    • Chassis
    • Media Players
    • Power Supply
    • Laptop and Mobile
    • Smartphone
    • Networking
    • Keyboard Mouse
    • Cooling
    • Search articles
    • Knowledgebase
    • More Categories
  • FORUMS
  • NEWSLETTER
  • CONTACT

New Reviews
ASUS ROG Rapture GT-AXE11000 WIFI6E router review
Backforce One Plus Gaming Chair review
ASUS GeForce RTX 3080 Noctua OC review
AMD Ryzen 5 5600 review
PowerColor RX 6650 XT Hellhound White review
FSP Hydro PTM Pro (1200W PSU) review
ASUS ROG Radeon RX 6750 XT STRIX review
AMD FidelityFX Super Resolution 2.0 - preview
Sapphire Radeon RX 6650 XT Nitro+ review
Sapphire Radeon RX 6950 XT Sapphire Nitro+ Pure review

New Downloads
Corsair Utility Engine Download (iCUE) Download v4.24.193
Intel HD graphics Driver Download Version: 30.0.101.1994
GeForce 512.95 WHQL driver download
AMD Radeon Software Adrenalin 22.5.2 driver download
AIDA64 Download Version 6.70
FurMark Download v1.30
Display Driver Uninstaller Download version 18.0.5.1
Download Samsung Magician v7.1.1.820
Intel ARC graphics Driver Download Version: 30.0.101.1732
HWiNFO Download v7.24


New Forum Topics
AMD changes its statements, Ryzen 7000 will consume up to 230W Info Zone - gEngines, Ray Tracing, DLSS, DLAA, TSR, FSR, XeSS, DLDSR etc. LG offers a 47,5-inch OLED gaming display with 4K resolution with up to 138 Hz refresh rate. Review: ASUS ROG Rapture GT-AXE11000 WIFI6E router AMD Radeon Software Adrenalin 22.5.1 WHQL driver download and discussion Rumor: Intel Raptor Lake and Sapphire Rapids processors will be available in October Fine Utilise Power of RadeonPRO Software & SweetFX Part 2 MSI AB / RTSS development news thread Free to grab: Bioshock: The Collection at Epic Games Store [3rd-Party Driver] Amernime Zone Radeon Insight 22.5.1 WHQL Driver Pack (Released)




Guru3D.com » Review » GALAX GeForce RTX 2070 Super WTF review » Page 4

GALAX GeForce RTX 2070 Super WTF review - The Turing GPU

by Hilbert Hagedoorn on: 11/08/2019 02:34 PM [ 4] 22 comment(s)

Tweet

The Turing GPU

Looking at the Turing GPU, there is a lot of stuff you can recognize, but starting at Turing there certainly have been fundamental block changes in the architecture; the SM (Streaming Multiprocessor) clusters have separated, and now are core separated isolated blocks. Bear in mind that the base building block for all Turing GPUs will be the TU102, that is the flagship GPU that has been used on the GeForce RTX 2080 Ti. The RTX 2080 uses the TU104, the RTX 2060 TU106 and 2070 SUPER also TU104. In essence, the TU106 is the more simplified revision of the TU102, but shares its architecture. It's fabbed at 12nm FFN.

  

  
 

Turing GPU Specifications

Gamers will immediately look at the shader processors, the Quadro RTX 8000 has 4608 of them enabled and, since everything with bits is in multitudes of eight, while looking at the GPU die photos; a TU104 based GeForce RTX 2070 Super has 40 SMs (streaming multiprocessors) each holding 64 cores = 2560 Shader processors. This GPU is fabbed on an optimized 12nm TSMC FinFET+ node and the full GPU is unlocked. 

  

GeForce RTX 2070 SuperRTX 2070 FERTX 2060 SuperRTX 2060 FE
GPU   TU104-410-A1 TU106 TU106-410-A1 TU106
Shader cores   2560 2304 2176 1920
Transistor count   13.6 Billion 10.8 Billion 10.8 Billion 10.8 Billion
Base frequency   1605 MHz 1410 MHz 1470 MHz 1365 MHz
Boost frequency   1770 MHz 1710 MHz 1650 MHz 1680 MHz
Memory   8GB GDDR6 8GB GDDR6 8GB GDDR6 6GB GDDR6
Memory frequency   14 Gbps 14 Gbps 14 Gbps 14 Gbps
Memory bus   256-bit 256-bit 256-bit 192-bit
Memory bandwidth   448 GB/s 448 GB/s 448 GB/s 336 GB/s
L2 Cache   4096 KB 4096 KB 4096 KB 4096 KB
RT cores   40 36 34 30
Tensor cores   288 288 240 240
Texture units   184 144 136 120
ROPs   64 64 64 48
TDP   190W 185W 175W 160W
Power connector   8-pin 8-pin 8-pin 8-pin
NVLink   No No No No
Price   $ 499 $ 599 $ 399 $ 349

  * reference specifications

 

Turing architecture

We'll try and be brief here, but Turing is a new and completely overhauled architecture that actually has gotten a new SM (Streaming Multiprocessor) design. As I mentioned, it has a bit of everything, but mostly it hints towards Volta. Two SMs are included per TPC (Texture / Processor Cluster - a group made up of several SMs). Each SM has a total of 64 FP32 Cores and 64 INT32 Cores. Now before you get all confused, yes, that is radically different from Pascal (GeForce series 1000) which had one SM per TPC and 128 FP32 Cores per SM. The Turing SM architecture supports parallel execution of FP32 and INT32 operations, independent thread scheduling similar to the Volta GV100 GPU. That's also described as concurrent execution of FP32 and INT32 operation. Each Turing SM holds eight Turing Tensor Cores. With that out of the way, have a peek at the block diagram below.

  

 

Each Turing SM is partitioned into four processing blocks, each holds 16 FP32 Cores, 16 INT32 Cores, two Tensor Cores, one warp scheduler, and one dispatch unit. Each block includes a new L0 instruction cache and a 64 KB register file. The four processing blocks share a combined 96 KB L1 data cache/shared memory. Traditional graphics workloads partition the 96 KB L1/shared memory as 64 KB of dedicated graphics shader RAM and 32 KB for texture cache and register file spill area. Compute workloads can divide the 96 KB into 32 KB shared memory and 64 KB L1 cache, or 64 KB shared memory and 32 KB L1 cache.

Concurrent Execution of Floating Point (fp32) and Integer Instructions (int32)

Turing’s SM initiates a new unified architecture for shared memory, L1, and texture caching. This unified design allows the L1 cache to leverage resources, increasing its bandwidth by 2x per TPC compared to Pascal, and allows it to be reconfigured to grow larger when shared memory allocations are not using all the shared memory capacity. The Turing L1 can be as large as 64 KB in size, combined with a 32 KB per SM shared memory allocation, or it can reduce to 32 KB, allowing 64 KB of allocation to be used for shared memory. Turing’s L2 cache capacity has also been increased. Combining the L1 data cache with the shared memory reduces latency and provides higher bandwidth than the L1 cache implementation used previously in Pascal GPUs. NVIDIA claims these changes in SM enable Turing to achieve 50% improvement in delivered performance per CUDA core. Turing GPUs add larger and faster L2 caches in addition to the new GDDR6 memory subsystem. The TU102 GPU ships with 6 MB of L2 cache, double the 3 MB of L2 cache that was offered in the prior generation GP102 GPU used in the TITAN XP. TU102 also provides significantly higher L2 cache bandwidth than GP102. Like prior generations of NVIDIA GPUs, each ROP partition in Turing contains eight ROP units and each unit can process a single-color sample. A full TU102 chip would contain 12 ROP partitions for a total of 96 ROPs. The RTX 2080 and 2070 have 64 of these.




25 pages « 3 4 5 6 next »



Related Articles
GALAX GeForce RTX 3060 EX White 12G review
All of your base are belong to GALAX; join us as we evaluate their GeForce RTX 3060 EX White variant, which has the same 12GB of memory and 3584 shading processors but features a factory boost speed o...

GALAX GeForce RTX 3080 SG review
In this review, we benchmark the GeForce RTX 3080 SG (short for Serious Gaming) from none other than GALAX. Armed with a triple fan based blade wing cooling design this card is to seriously out some...

GALAX GeForce RTX 2070 Super HOF AE review
The name is so long that we have to shorten it here and there, but we review the GALAX GeForce RTX 2070 Super HOF (Hall of Fame) 10th Anniversary Edition. The product is big, massive and all-white in...

GALAX GeForce RTX 2070 Super WTF review
We review the new GALAX GeForce RTX 2070 Super in the WTF edition, and yes that is a playful way of wording in many regards. WTF is short for Work The Frames, which realistically you would have never...

© 2022