Vijesti, novosti i glasine o nadolazećim Radeon grafičkim karticama

[The Exiled]

AMD's next compute GPU could have two dies and 128 GB of memory

Aldebaran, the computer part, has appeared a few times in Linux patch notes. Its two-die design was half-implied in February and confirmed last month: "on Aldebaran, only primary die fetches valid power data. Show power/energy values as 0 on secondary die", one note reads. It’s the first AMD Accelerator/GPU to leverage an MCM (multi-chip module) design, a technique that’s been plastered on patents for years but has only just begun to materialize. Using multiple chips/dies connected closely allows for more scalability than traditional monolithic designs, but also reduces the per-compute unit performance of a card, particularly in poorly parallelized workloads. nVidia is expected to challenge Aldebaran with their own MCM designs based on the Hopper architecture next year. Intel, meanwhile, is on the verge of releasing the Ponte Vecchio accelerator based on the Xe HPC architecture and an early MCM implementation. Like Ponte Vecchio, Aldebaran leverages MCM to expand memory capacity, patch notes imply. A note from last week describes Aldebaran as having two dies, four UMCs per die, and eight channels connected to 2 GB of HBM per UMC or 128GB, in total. The patch notes also mention support for a "new HBM2 memory type," presumably HBM2e. If AMD is using SK Hynix’s new HBM2e memory modules with 3.6 Gbps of bandwidth and a 4096-bit bus on each die, then Aldebaran could have 3.6 TB/s of memory bandwidth. AMD CEO Dr Lisa Su has already said that the CDNA 2 architecture, meaning Aldebaran, is slated for release later this year. Presumably, it will release as the MI200 Accelerator.

Izvor: TechSpot

[The Exiled]

Next-gen AMD Radeon RDNA 3 flagship to feature 15,360 stream processors?

AMD's next generation RDNA3 graphics architecture generation could see a near-quadrupling in raw SIMD muscle over the current RDNA2, according to a spectacular rumor. Apparently, the company will deploy as many as 15,360 stream processors (quadruple that of a Radeon RX 6800), and spread across 60 WGPs (Workgroup Processors), and do away with the compute unit. This is possibly because the RDNA 3 compute unit won't be as independent as the ones on the original RDNA or even RDNA 2, which begins to see groups of two CUs share common resources. Another set of rumors suggest that AMD won't play nVidia's game of designing GPUs with wide memory bus widths, and instead build on its Infinity Cache technology, by increasing the on-die cache size and bandwidth, while retaining "affordable" discrete memory bus widths, such as 256-bit. As for the chip itself, it's rumored that the top RDNA 3 part, the so-called "Navi 31," could feature a multi-chip module design (at least two logic dies), each with 30 WGPs. Each of the two is expected to be built on a next-gen silicon fabrication node that's either TSMC N5 (5nm), or a special 6nm node TSMC is designing for AMD. Much like the next-generation "Lovelace" architecture by nVidia, AMD's RDNA 3 could see the light of the day only in 2022.

Izvor: TechPowerUp + Interneti

[Starsylph]

AMD won't play nVidia's game of designing GPUs with wide memory bus widths, and instead build on its Infinity Cache technology, by increasing the on-die cache size and bandwidth, while retaining "affordable" discrete memory bus widths, such as 256-bit.

Hoće li to efektivno značiti jeftinije RDNA3 od prethodnika?

[The Exiled]

Neće, jer nVidia i AMD zadnjih nekoliko godina imaju jebenu potvrdu da su krajnji kupci spremni platiti koliko god treba, neovisno o namjeni i(li) potrebi.

AMD ionako već ima spremnu MCM CDNA 2 MI200 karticu, tako da će RDNA 3 MCM modeli iskoristiti iste chiplet temelje, a cijene će sigurno biti paprene.

Prošla su vremena 200$ do 250$ kartica poput HD 4870 ili HD 5850.

[picajzla0707]

Autor The Exiled

Prošla su vremena 200$ do 250$ kartica poput HD 4870 ili HD 5850.

problem je i u tome što grafe niže vrijednosti daleko slabije profitiraju po pitanju perfomansi s novim generacijama

3080, 3070 imale su popriličan napredak u odnosu na prethodnice 2080 i 2070
jeftinija 3060 je puno manji napredak u perfomansama u odnosu na 2060

a nije nit amd ništa bolji, 6700xt je 20-30 posto jača od prethodne 5700xt, a
jeftina 6600xt će imati puno manji napredak, čak su i broj jezgri smanjili u odnosu na 5600xt, što sa 6700xt nisu radili jer ima isti broj jezgri kao i prethodna 5700xt

dakle namjerno se ide na to da jeftinije grafe srežu i da ispadnu manje isplative za upgrejd

[The Exiled]

Da, nažalost skužili su da im to uredno prolazi, a i Škot se u nekolicini svojih osvrta bavil tim da nVidia i Radeon s godinama sve više dižu cijene uz sve manji napredak u performansama u odnosu na prijašnju generaciju.

Jebiga, dokle god prolazi, nemaju razloga za promjene, a vidimo iz svakodnevnih primjera da su ljudi uistinu spremni platiti svaku cijenu. Dok nekome ne pukne film, obje firme dalje nastavljaju istim putem.

[Mladenxy]

Sami smo si krivi...a oni nisu blesavi da gledaju kako skalperi i trgovci zarađuju ko blesavi. Dignu cijenu max tak da njima ostane manje manevarskog prostora.

[The Exiled]

Pa da, jasno je da npr. jedna RX 6700 XT nema kaj koštati više od 2.500 kn do 3.000 kn, jer je to rang u koji po pitanju performansi spada, ali možemo se slikati kad se prodaju za duplo više bez problema.

[The Exiled]

AMD confirms initial shipment of Instinct MI200 “Aldebaran” accelerator, its first MCM GPU

AMD said this week that it had started initial shipments of its next-generation Instinct compute GPUs based on the CDNA 2 architecture. These aren't the sort of GPU you'd find in the best graphics cards. Instead, the new Instinct MI200 compute accelerator carries 128GB of HBM2e memory and is designed for supercomputers and other high-performance computing applications, such as AI or ML. AMD's Instinct MI200 'Aldebaran' compute GPU will power the world's first exascale-class Frontier supercomputer, due to be delivered in late 2021 (as well as some smaller machines), so it's about time for AMD to start production of the accelerator. AMD has not yet disclosed any architectural peculiarities about CDNA 2, so the only two things we know about the Instinct MI200 for sure is that it features two dies connected with each other using AMD's Infinity Fabric, and it carries 128GB of HBM2e memory that connects to the dies using two 4096-bit interfaces.

Izvor: Tom's Hardware

[Starsylph]

Autor The Exiled

Pa da, jasno je da npr. jedna RX 6700 XT nema kaj koštati više od 2.500 kn do 3.000 kn, jer je to rang u koji po pitanju performansi spada, ali možemo se slikati kad se prodaju za duplo više bez problema.

Kažem ti doći će toliko sredinom sljedeće godine.

[The Exiled]

AMD Radeon PRO W6000X series GPUs bring groundbreaking high-performance AMD RDNA 2 architecture to Mac Pro

Built on groundbreaking AMD RDNA 2 architecture, AMD Infinity Cache and other advanced technologies, the new workstation graphics line-up includes the AMD Radeon PRO W6900X and AMD Radeon PRO W6800X GPUs. Mac Pro users also have the option of choosing the AMD Radeon PRO W6800X Duo graphics card, a dual-GPU configuration that leverages high-speed AMD Infinity Fabric interconnect technology to deliver outstanding levels of compute performance. Up to 64GB of GDDR6 memory with up to 512 GB/s bandwidth provides ultra-fast transfer speeds to power data-intensive professional applications. Up to 256MB (total) of last-level data cache integrated on the GPU die is designed to reduce latency and power consumption.

Izvor: AMD

[The Exiled]

AMD Radeon RX 7000 family may include mix of RDNA 3 and refreshed RDNA 2 GPUs

AMD will use RDNA 3 and RDNA 2 GPUs in its Radeon RX 7000 family. As you may recall, RDNA 3 is tipped to use TSMC's 5nm process node, which should help AMD ramp clock speeds and reap major performance-per-watt benefits.

It's reported that the Navi 32 and Nav 32 will use AMD's Multi-Chip Module (MCM) design, while Navi 33 will go with a more traditional monolithic approach.

Those RDNA 3 GPUs would be joined by refreshed versions of existing RDNA 2 GPUs. Navi 24, Navi 23, and Navi 22 RDNA 2 GPUs will be refreshed on an intermediary 6nm process node instead of the current 7nm process node. This should allow for some slightly tweaked clock speeds compared to current-generation cards, but we shouldn't expect any dramatic leaps in performance as we're bound to get with RDNA 3. Of course, this is all just speculation at this point, but it seems like a highly probable solution for AMD as it looks to maximize yields for its RDNA 3 GPUs in 2022 and leverages the economists of scale for a more mature 6nm process node with an RDNA 2 refresh.

Izvor: HotHardware

[The Exiled]

Killer Instinct: AMD’s Multi-Chip MI200 GPU readies for a major global debut

AMD’s next-generation supercomputer GPU is on its way – and by all appearances, it’s about to make a name for itself. The AMD Radeon Instinct MI200 GPU (a successor to the MI100) will, over the next year, begin to power three massive systems on three continents: the United States’ exascale Frontier system; the European Union’s pre-exascale LUMI system; and Australia’s petascale Setonix system. The new GPU – which began shipping to its first customers this summer – is based on AMD’s CDNA2 architecture and, unlike the PCIe MI100, uses a mezzanine form factor (akin to the SXM versions of Nvidia’s V100 and A100 GPUs). The MI200 will sport two dies – each rumored to have 110 compute units – connected via a passive interposer die, and up to 128GB of HBM2E DRAM. The MI200 is anticipated to deliver at least twice the performance of the single-die, 11.5-teraflops MI100, which had 120 compute units and 32GB of HBM2 memory. The MI200 will certainly be the first multi-chip module (MCM) GPU from AMD and also looks poised to be the first MCM GPU on the general market. AMD’s highest-profile customer for the MI200 might also be the first to launch: just a few weeks ago, it was revealed that the exascale Frontier supercomputer is undergoing installation at Oak Ridge National Laboratory (ORNL), with that installation scheduled to complete sometime this year or early next year.

Izvor: HPCwire

[Disco]

AMD Allegedly Preparing Refreshed 6 nm RDNA 2 Radeon RX 6000S GPU

[The Exiled]

AMD patent details innovative stacked accelerator that could mmpower next-gen RDNA GPUs

In a patent filed a year ago that was published just a few days ago, AMD outlines an additional die stacked on top of the GPU, separate from the existing chiplet structure. AMD calls this extra die an accelerated processing device (APD) and pitches it as a way to enhance machine learning capabilities. It would contain memory and one or more ML accelerators. By stacking the additional die, there is an inherent performance advantage that comes by way of high-speed interconnects to transfer data and commands. AMD notes the memory portion could be configurable as either or both a cache and as directly accessible memory. It also says it could contain matrix multiplication arithmetic logic units (ALUs) configured to perform matrix multiplication operations to boost ML chores.

It's important to note that a patent application is not a foregone conclusion of upcoming technology in actual products. And if if this does come to fruition, it's not clear if it would be bound for high-performance computing (HPC) like its dual-die Instinct MI200 or future gaming products, or both. But we're certainly intrigued at the prospect. There's already been chatter for AMD bringing stacked 3D V-cache to its gaming GPUs, and adding another stacked die to the mix is certainly within the realm of possibility. An APD die could serve some of the same purposes as nVidia's Tensor cores. It might be too late to expect this to be part of RDNA 3, but perhaps RDNA 4 is fair game. That's of course assuming AMD turns its patent into an actual product.

Izvor: HotHardware

[Disco]

Hajde da malo oživim temu, ugl kratka rasprava o tome koja arhitektura na kraju ima bolji lineup:

[The Exiled]

AMD RDNA 3 Economics: Lovelace may have the Pricing Problem

AMD teases second-gen Upscaling Graphics Tech reveal at GDC, possibly FSR 2.0

Izvor: HotHardware

[Matta]

The First Batch of Radeon RX 7700 XTs are About to Be Tested by AMD: Faster than the RX 6900 XT at Nearly Half the Price

According to Greymon55, it will be the first Navi 3x GPU to be announced, sometime towards the end of the year. Fabbed on TSMC’s 6nm node, Navi 33 will power the Radeon RX 7700 XT with performance exceeding that of the Radeon RX 6900 XT (Navi 21) at roughly half the price.
Overall, the Navi 31 die powering the Radeon RX 7900 XT should be 2.2-2.5x faster than the RX 6900 XT, with the RX 7800 XT (Navi 32) beating the RX 6800 XT by 50-80%.

[stitrek]

Koliko se ocekuje TDP za Navi 31 s obzirom da za Nvidia Rtx 4090 se spominje 600W+

[The Exiled]

AMD Navi 33 Leak: nVidia Lovelace loses the Efficiency War to RDNA 3

[spawn]

Brisi, pogresna tema

[the_ghost]

ako stvarno bude tako, za rasterization performance, to bi bilo jebeno.

[The Exiled]

AMD Navi 31 & MI300 Leak: 3D Stacking the Deck against Lovelace & Hopper

[Mladenxy]

Koliko će ta čuda trošiti da mi je znat, jedna osrednja RTX 4070 navodno ogromnih 300W, za vidjeti je da li će AMD pratiti nebulozni trend.

[The Exiled]

AMD za RDNA 3 ima spreman chiplet dizajn, a sudeći po Zen procesorima unatrag zadnjih 5 godina vidimo da stvar drži vodu, mada nova grafička rješenja budu svakak trošila više od sadašnjih, samo kaj izgleda da nVidia ide u žešće krajnjosti s monolitnim pristupom.

[The Exiled]

AMD's upcoming RX 7900 XT rumored to have 24GB of VRAM, 384-bit memory bus

Graphics cards based on AMD's new RDNA3 architecture are rumored to debut later this year. The flagship Radeon RX 7900 XT will reportedly feature a Multi-Chip-Module (MCM) GPU design with 384 MB of Infinity Cache and a staggering 24 GB of GDDR6 memory. Both AMD and nVidia are expected to launch their next-gen graphics cards later this year. We've already had plenty of leaks involving Nvidia's GPUs, so today, we're focusing on news involving AMD's upcoming RDNA3-based cards. Judging by past releases, AMD is probably going to announce the higher-end models first, which are going to be based on the Navi 31 GPU.

Rumors are that it will feature a 384-bit memory bus, and it will ship with up to 24GB of VRAM. Assuming that the company will use 18Gbps GDDR6 modules again, this would result in a memory bandwidth of 864GB/s. We already knew that the GPU might feature an MCM design, but instead of featuring multiple GCDs (Graphics Compute Dies) as previously thought, AMD will reportedly use six MCDs (Memory Compute Dies) with one big GCD. What this means is that each MCD will feature 64MB of L3 cache, combining for a total of 384MB of what AMD calls Infinity Cache. That is three times more than the monolithic Navi 21 GPU. Splitting the GPU up into multiple dies could also improve yields, resulting in lower production costs. The GCD will be fabricated on one of TSMC's 5nm process nodes, with the MCDs relegated to a 6nm node. AMD's Zen 3 CPUs use a similar design, as the I/O dies use a less-advanced GlobalFoundries process.

Izvor: TechSpot

[MadBoY]

[Consigliere]

Intervju sa podpredsjenikom AMD-a i inženjerom zaduženim za razvoj arhitekture Sam Naffziger-om.
https://www.tomshardware.com/feature...-sam-naffziger

Nadopuna:
Znači dizanje potrošnje je neminovno, pogotovo ako imas konkurenta koji nastoji izvuci max performanse bez obzira na potrošnju, i nemas druge nego ici tim korakom.

[The Exiled]

AMD confirms RDNA 3 has "rearchitected compute units" that "enhance ray tracing"

When AMD's David Wang discussed the company's RDNA 3 graphics architecture at their 2022 Financial Analysts Day, Wang stated that RDNA 3 would feature "rearchitected compute units" that would "enhance ray tracing capabilities". These enhanced ray tracing capabilities were not listed on AMD's RDNA 3 slides, so this comment was unnoticed by much of the gaming press. Alongside these rearchitected compute units, RDNA 3 also boasts a "optimised graphics pipeline" that delivers "even faster clock speeds and improved power efficiency". This should enable higher levels of compute performance per compute units, as each unit will be able to complete more clock cycles in any given time. More clock cycles means more work, and more work means more performance. While we can expect a significant boost to AMD's ray tracing performance with RDNA 3, it remains to be seen if AMD will be able to catch up with nVidia.

nVidia aren't exactly sitting on their hands when it comes to ray tracing performance, and you can be sure that nVidia's RTX 40 series will feature its own set of architectural enhancements. Based on AMD's comments, we can expect more ray tracing performance per compute unit and per clock cycle with AMD's RDNA 3 graphics architecture. Beyond that, higher clock speeds should also enhanced Radeon's ray tracing performance (just like its rasterisation performance). Even so, the size of these benefits are unknown, and AMD does not expect AMD to push the gaming industry towards path tracing. With RDNA 3, AMD's still pushing a "hybrid rendering" approach.

Full on patch tracing, where an entire scene and all aspects of a game is ray traced, is not something that AMD is pushing for with RDNA 3. AMD's pushing hybrid rendering, where ray tracing is used alongside traditional rasterised graphics to deliver high performance levels while accessing the visual benefits of ray tracing. AMD is currently investing in techniques that can enable ray tracing in a more performance friendly manner, allowing gamers to get the most performance out of their graphics cards and games. This makes clear that while AMD plans to deliver a ray tracing boost with RDNA 3, they are not promising earth shattering benefits. Hybrid rendering is the future, as we would have to wait for another console generation before the entire gaming industry pushing things to another level. AMD's techniques will benefit current-generation and next-generation GPU hardware, and all of the AMD-powered gaming systems that feature RDNA 2 graphics. There's a reason why few games are fully path traced, and why most path traced games are modded versions of early 3D titles with simple geometry and environments. Simply put, path tracing is too demanding for most modern games at high resolutions and framerates. The future us hybrid, at least for now.

Izvor: OC3D

EDIT:

AMD's answer to GeForce's Tensor Cores may be coming with next-gen RDNA3 architecture

Izvor: TechSpot

[The Exiled]

AMD’s Radeon RX 7950 XT may feature up to 10 chiplets: 8 MCDs, 512MB Infinity Cache, 512-Bit bus, & 16,384 cores

It looks like AMD is going ahead with its mega RDNA 3 GPU featuring several compute and memory complex dies. The Navi 3x (x=4 or 5) SKU will consist of two Graphics Compute Dies (GCDs) and up to eight Memory Complex Dies (MCDs). That’s a total of eight disaggregate chiplets or dies fabbed on TSMC’s 5nm and 6nm process nodes resulting in an absolute behemoth of a graphics card. Each of the two GCDs will be manufactured using the N5 process, packing up to 8,192 stream processors or 16,384 overall. The MCDs, on the hand, will be a byproduct of the N6 node and pack 32MB of Infinity Cache per die via a 64-bit bus. That gives an overall last level cache of 256MB and a bus width of 512-bit. From what we know about Radeon RX 7900 XT (Navi 31), each of the memory dies will be 3D stacked with an additional cache die (3D V-Cache) for a total of 512MB of L3 cache.

The memory buffer is likely to be pegged at 32GB (GDDR6 18Gbps), making it the first consumer graphics card to feature over 30GB of memory if the RTX 4090/4090 Ti doesn’t do it first. Of course, this’ll also push the power draw well over Navi 31’s, somewhere between 450-600W. At the moment, it’s unclear what the codename for this SKU it will supposedly be called Navi 34 or 35. Furthermore, we won’t be seeing this part anytime soon. AMD is likely planning for a mid-to-late 2023 launch, probably around the same time as the RTX 4090 Ti.

AMD Radeon RX 7900 XT to feature 384MB of L3, 192MB 3D stacked Cache, 384-bit bus across 7 MCD chiplets

Izvor: Hardware Times