Search results

It is a given that multi-GPU scaling is going to be an app by app improvement but at least on the Mac side, their failed trash can endeavor did get app developers to embrace the idea of multi-GPU where it can. This time around there are coherent links between the GPUs and four RX 6800 (albeit...

I'm genuinely surprised that Apple has stuck with Intel and not switched over to AMD Threadripper, Threadripper Pro or even Epyc chips. The improvement on the Mac Pro 2019's IO would be massive and significantly cheaper as the Threadripper Pro and Epyc chips sport a full 128 PCIe lanes which...

There are a couple of things people seem to be forgetting about when it comes to the current state of cutting edge production. First off was that Global Foundries tapped out of the game not too long ago. This is what forced AMD to go to TSMC for the CPU side of their Zen 2 and Zen 3 chips...

Looks like a simulator setup with either a Barco or a Christie Digital setup. Both companies have excellent blending and image warping technologies that will map on to the curved surface of a the simulator's shell. Neat stuff but be prepared for sticker shock. Depending on how bright you want...

There were handful of $60 games back in the 90's, notably those using larger sized carts in the 16 bit days. Since Nintendo stuck with carts, the majority of N64 games were $60 new were as the Sega Saturn and and Sony Playstation 1 leveraged the cheaper manufacturing cost of their mediums to...

The Navi 12 part is an interesting surprise. Probably a play by Apple to reduce motherboard area and further reduce power while keeping performance similar enough to 5600. The real question is if Apple will continue to leverage custom AMD designs (who else has access to Navi 12?) if/when they...

Here is a recent audio card I picked up for work. Goes from various sources to a mixing console and then into the PC system. Neat piece of kit for the professional space. For home, I picked up this that outputs to DSP cluster I've thrown together on the cheap. Technically it should be...

It is Infinity Fabric between the GPUs. It should also be pointed out that Apple's quad configuration isn't optimal either. Imagine the four GPUs in a square and there is an Infinity Fabric link on each edge. Apple's configuration doesn't connect the GPU's catercorner from each other. In...

The bridge thing isn't new, even for Vega 20: Apple uses it in their 2019 revision of the Mac Pro. It's how you get the quad GPU configuration. Both AMD and nVidia are seemingly working on better ways of teaming multiple GPUs together. The advancements in ray tracing make revisiting these...

This has been known about for years but we are just now hitting that barrier without any sort material science break through for the next generation of DRAM. Back when this DRAM barrier was initially discussed, there was a proposed solution that can still work: use SRAM. This is the same...

GDDR standards tend to mimic vanilla DDR standard in terms of cadence. GDDR7 likely implements many changes that has occurred between DDR4 -> DDR5. I do wonder if this requires moving to a package-on-package topology as the GDDR6 currently has a rather short trace length. While I wouldn't say...

AMD seems to have a good product here but looking at the world right now it appears to be the absolute worst time for release.

The game is good but not as great as 2016. Two steps forward and one step back it seems. The game's story is now up front and center compared to 2016's which did indeed have a story but following it was more organic and paying attention to it was far more optional. The story this time around...

Live stream about the music:

Mick Gordon doing a live stream about the game's music:

As pointed out by others, fake. Main reason being is the weird ROP to memory bus width and capacity. 24 GB is possible on a 4096 bit wide bus in various means (two 4 GB stacks + two 8 GB stacks, four 6 GB stacks etc.) However, AMD has aligned the number of ROPs with memory bus width. With...

I just hope that the new socket 1200 doesn't fully isolate power between the CPU and GPU on the chip. Previous F parts still pulled all there power just from the CPU side of things. Power consumption was the same which meant that the CPU side was simply moving more amperage off of the existing...

I strongly suspect that some of those code names are for the announced but not yet shipping version of the Radeon RX 5700 for the Mac Pro's MPX slots. There is a single and dual RX 5700 board and they'll also have a few TB3 controllers too. These probably get their own code name. The big one...

Apple's usage of Vega 20 was two fold: lower power consumption due to HBM (vs. GDDR5) and reduced board space by putting everything onto a single package. Basically a win-win for a company that puts form ahead of function. The big question I have if is multiple "Navi 12" dies can be...

Interesting product. The hybrid x86 design is something they announced last year with some very innovative packaging technologies. It won't be bleeding edge in terms of performance but its performance/watt should be. This is a very necessary stepping stone for Intel to remain competitive...

i"m still of the mindset that the DG1 cards going to developers is just a Tiger Lake SoC with the x86 cores fused off. Main reason why it'll perform better than their laptop counter parts is that as a discreet card it has a better cooling solution and power delivery as well as a far higher...

What AMD should be doing is going beyond PCie 4.0 and enabling a full 25 GT/s Infinity Fabric link to certain slots when pairing an AMD CPU and an AMD GPU. Not only there be a nice bandwidth increase but also coherency with a flat memory space between the two device. This is why AMD was hyping...

I can fathom dropping PCIe 4.0 for normal slots but leveraging it for the DMI link would be a nice boost for chipset performance. The chipset has often been over committed in terms of how much bandwidth it can provide vs. the DMI uplink to the CPU but traditionally end users didn't stress it...

30 minutes of music which only feels like two songs due to how DOOM will shift between sections based upon what is going on. NEED MOAR.

Yeah, I mentioned that, hence the term 'like'. It is noteworthy that AMD is actually putting forth the effort to create a new GPU-GPU link where as XMDA was just done over the PCIe bus. This is on top of the transition to PCIe 4.0 on the RX 5700 series. I have a feeling that with the arrival...

Dual GPU isn't exactly dead. Apple is doing it for their latest Mac Pro. Not just dual Vega 20 but also a dual RX 5700 option is in the pipeline. While generally ignored thus far on the PC side, those GPUs are linked via AMD's Infinity Fabric. The two dual Vega 20 cards even come with...

Yes please. I'll take more. And seconds.

I'm half-way expecting this to simply be TigerLake but with the X86 cores disabled. The ability to do this has actually been in Intel SoC's since Haswell as they use it for their VCA cards. Where the performance increase stems from would be higher clocks permitted by a much higher TDP in a...

The main thing is that M.2 drives don't permit many NAND chips on the PCB by default so reaching peak speed there is going to take a generation or two. Flip over to U.2 which has the same 4 lane PCIe 4.0 support, they could saturate that now as additional channels and more NAND per channel are...

Intel also offers this card which is aimed to accelerate graphics workloads, just in an unusual way. The difference between that and DG1 in terms of functionality is that the DG1 has a direct video output. Both cards leverage existing chips and pretty much just disable the x86 side of things...

I also wonder how much of that performance gain is based upon the usage of LPDDR4X due to its higher bandwidth. It may not be as impressive if paired with more vanilla DDR4 at 3200, even if the fair comparison point is Ice Lake with DDR4-3200. I'm really curious how memory/cache latencies fair...

I wonder how many TB3 devices will have the USB4 label retroactively applied.

Are you referring to how the audio is routed on the network? IE how the audio source selected makes it to the destination? If so, there are various routing applications used that can change which transmitter goes to which receiver on a per channel or per bundled channel stream basis. This...

Fair point but being more professional in nature, there tend to be less exaggeration than on the consumer side of things. For analog, balanced is nice because it is indeed more difficult for interference creep in, but not impossible. The more likely scenario is that interference from the end...

The DACs are professional grade with a very high signal to noise ratio. Dante transport is typically 96 kHz, 24 bit so the source is good with some gear supporting all the way up to 384 kHz. The other thing worth noting is that the amount of analog circuitry is kept to a bare minimum so the...

The thing is that SSD manufactures road the SATA3 wagon for while as both NAND performance and per chip capacity increased. The result wasn't an increase in speed (which was limited by SATA3 speeds anyway) or capacities but rather they cut costs by reducing the number of NAND channels in the...

Dante is one of several second generation audio-over-IP technologies (AVB, BluLink, Q-Sys and AES67 being the others) which have replaced several older first generation technologies (Ethertalk, Cobranet). They tend appear in profession gear and for awhile every manufacturer was trying to push...

Have you considered doing this over the network via Dante and one of Audinate's Ethernet-to-XLR adapters?

The thing with EUV is that the maximum die size has decreased vs. current methods. MCM is necessary because they simply can't make a ~800 mm^2 using EUV due to the optics involved.

Intel has gotten their FPGAs working exceptionally well with MCM techniques. That is two large FPGA dies acting as if they were one large monolithic chip. A GPU wouldn't be any different as the internal buses that would normally connect on-die units would just hop across the interconnect...