8 cores, 16 threads, a jump in computing capability that’s a bit higher than between AMD’s previous chips.
Now before dismissing this chip out of hand, it looks like AMD might be learning from their mistakes since they abandoned the Vishera architecture in favor of one closer to what Intel’s doing.
I’m also really hoping they can seriously start competing with Intel again as a result, if not only for the fact it would force Intel to try for bigger leaps in CPU power alongside lower prices for the enthusiast chips (their new 10 core processor is over 1700 bucks for starters).
Exciting times in terms of computing power may be at hand once again if this turns out just as advertised, let’s hope so.
I’m definitely looking forward to an AMD CPU with eight real cores, designed by the father of the almighty Athlon64!
40% better IPC is a massive jump, it’s just about what AMD needs to be competitive with current Intel CPUs.
Now before dismissing this chip out of hand, it looks like AMD might be learning from their mistakes since they abandoned the Vishera architecture in favor of one closer to what Intel’s doing.
Kind of like Intel learned from its mistake with the Pentium 4 architecture in favor of one closer to what AMD was doing at the time.
beer they have zens coming with up to 32 dual thread cores so 64 threads. not sure what the price will be but if you use up all 64 threads and you are not running a server…you have way to much stuff running in the background. shut down the bloat ware. open up task manager and google what all the items are, you will be surprised how many are data miners for google, facebook, etc…just tracking your for advertisers. you can shut that stuff down safely.
AMD has now with Zen not only a potent CPU, but the Polaris architecture introduce UMA, which means you have only an APU(CPU+GPU) and a single HBM/GDDR5X unified memory!
Nvidia WANTS to do this, but really cant, cause Intel would need to implement CUDA cores on the CPU (the Intel GPU is a joke) which will never happen, and NVlink is too slow compared to REAL DMA on HBM (2TB/s) on a single APU!
Like NVMe is way to fast for SATA6, a CPU<>GPU HBM UMA will destroy the 16x GPU slot in the long run.
AMD will win, i tell you, in the long run, unless Intel and Nvidia merge (which will not happen)…
This is very exciting indeed. 8 physical cores is what I initially thought the FX-series was going to be (due to false marketing) so it’s good to see them learning from their mistakes.
Polaris still looks pretty decent though, and pretty affordable as well. Anyone who currently has a mid-range ATI GPU in their machine (and doesn’t feel like spending a large amount of money on new hardware) would probably want to upgrade.
I haven’t heard anything concrete about using GDDR5/HBM on APUs, but it’s certainly interesting. However, there isn’t necessarily that much performance to be gained from uniform memory, it highly depends on your use case and it’s also questionable whether programmers are going to optimize for something like that. Also, GDDR has higher bandwidth but also higher latency, which isn’t good from a CPU perspective. It’s also more expensive, but it does make quite a difference for the GPU.
Another fact of life is that you cannot have one large chip (CPU/CPU combo) with higher peak performance than two large chips (CPU and GPU separate). By “large”, I’m referring to the biggest chips that are still economically feasible. You won’t have an APU beat the best dedicated CPU/GPU system, but the price/performance is bound to be better.
Nvidia WANTS to do this, but really cant, cause Intel would need to implement CUDA cores on the CPU (the Intel GPU is a joke) which will never happen, and NVlink is too slow compared to REAL DMA on HBM (2TB/s) on a single APU!
NVIDIA already has HBM on their Tesla Pascal chips. Putting a CPU on there would be a waste, it couldn’t even handle all that bandwidth. You need a massively parallel processor to take advantage of HBM, but then you’re also massively limited for RAM size because it needs to be so close to the processor.
The point of NVLink is to improve bandwidth to the CPU and potentially hundreds of gigabytes of RAM, which you can’t have with HBM.
By the way, the newer Intel GPUs are actually pretty good.
AMD will win, i tell you, in the long run, unless Intel and Nvidia merge (which will not happen)…
Maybe, if the market decides they’d rather have more economical APUs instead of 400$+ add-in boards.
Oh, this Zen thing looks quite nice. And it just so happens I’m jonesing for an upgrade. It feels like Intel has been capitalizing on its advantage a bit too much. I’ll happily give my money to team red if they can supply the performance.
Now, if only I can find a relatively cheap dual socket AM4 motherboard…
Dual socket mobo with this chip would make it 32 cores. Quite nice indeed. Especially if those cores perform close to what your average i7 is doing right now.
Or you could easily use Linux like I do.
Rendering with CPU on Linux is way faster.
NVidia drivers for Linux are great now, sometimes faster than Windows.
UMA with GDDR5 on an APU is as old as the PS4, which BTW is an AMD 8 Core APU. Its old. Every dev profits from the fact that both CPU and GPU see and can access the DDR5 memory of the PS4.
Yes, HBM is only good for an APU if there is enough RAM, but remember, everything is either stacked or die-sealed. You dont have to have everything on the same DIE.
An APU with a dual DIE with 8 Cores lying beneath a 2560 cores GPU in FinFET 14 would access an UMA of lets say 64/128 GB HBM3@4TB/s. Now imagine the speed of that beast, and the UMA memory efficiency (stored once, accessed by both). Even if you would have to have a bigger APU case, it would still consume less overall and be cooler. No RAM sockets, no PCIe16 for the GPU, etc…
Also that PC case would be really tiny!
UMA is not a new concept either, its only that it was only implemented with HPC (remember 2D torus SGI nUMA 1998?), and UMA IS (as is HBM) the way to go in ANY compute parallelized hardware, and now it comes to the Desktop, finally, but it will not work well over crippled links like NVlink.
The PS4 arch is the best example what can be achieved with that combination for main RAM, even if its “only” GDDR5.
And yes, not only Polaris, but Vega will be a game changer, for sure.
But Intel, good GPUs ? Tell that to any GPU renderer… lol
The same with Intel’s upcoming chips, it might not be a surprise at all since the fab process is getting more difficult (on top of what they are supply chain issues and weak demand for such powerful hardware).
2016 may end up being a year with no new CPU’s for those buying below the enthusiast level (4 core i7 chips and below), so don’t prepare to shop for a new machine yet. Hopefully AMD will take the extra time to make further improvements to the architecture.
EDIT: With a simple Google search, one source states that another reason for the Zen delay is so AMD can place more focus on their new Polaris GPU’s (I’m guessing if Polaris sells well then it can only be a good thing for Zen).
Well between these tech giants its a customer battle, intel might simply drop prices as a answer. (they did that before).
However the strong point here is the lower power usage, because that makes it interesting in rack servers.
As less power means, less heat production, less payment for active cooling of a server park.
That’s a major thing these days in server-parks.
As for consumers, well it would be nice if AMD creates some competition.
I assume that within a few years from now china might also get into play with some home grown chips
As recently they made a super computer with their own developed chips, their goal is to create affordable chips (mips architecture).
So perhaps within 2 ~ 4 years, we get real competition.
