The Intel Xe GPU is Intel’s first foray into discrete graphics because the ill-fated Larrabee challenge a few years in the past. But let’s not dwell on the previous, it is a shiny new future for Intel, and probably PC avid gamers too, because it means we’re going to have a real third participant within the graphics market and that may solely be a superb factor. For us, at the least.
Intel’s Xe GPU was successfully introduced final August by way of a teaser tweet promising to “set our graphics free” with a discrete graphics card in 2020. But it wasn’t till December final 12 months, on the firm’s Architecture Day, that we discovered the precise identify Intel had chosen for its new GPU enterprise.
It didn’t actually reveal an entire lot concerning the underlying graphics tech on the occasion, apart from the truth that it’s pronounced X-E, and never Zee as I can’t assist however do. That’s the issue with having each Xe and Xeon in your product portfolio and never following the identical phonetics…
But that doesn’t imply we don’t know a good bit concerning the new entry into the PC graphics card market. Well, we at the least have just a few scurrilous rumours, some real leaks, and some educated guesstimates to whet our appetites for what subsequent 12 months has to supply.
Yes, Intel goes to launch a discrete graphics card for the gaming market subsequent 12 months, and it appears just like the early engineering samples are already on their method out to builders too. For these of us who waited by means of the Larrabee debacle it’s relatively thrilling to have the ability to say that.
Vital Stats
Intel Xe launch date
We’re pondering that. between a tweet from Raja Koduri and a few ‘insider sources.’ we’re a summer time 2020 launch for the brand new discrete Intel Xe graphics playing cards.
Intel Xe specs
It appears doubtless that the preliminary Xe GPUs will largely be primarily based on the constructing blocks put in place by the Ice Lake Gen11 graphics structure. And we’re anticipating 128 EU, 256 EU, and 512 EU components for the entry, mid-range, and high-end markets.
Intel Xe efficiency
Not quite a bit is thought concerning the potential efficiency of the brand new Intel graphics playing cards, however Gen11 GPUs with simply 64 EUs are able to first rate 720p gaming. So with many occasions that determine, the Intel Xe might be genuinely aggressive playing cards. If they clock excessive sufficient…
Intel Xe worth
Spanning entry-level, mid-range, and high-end, you’re a variety of graphics card pricing probably from $200 all the best way as much as $1,000… relying on really how highly effective that 512 EU card seems to be.
What is the Intel Xe launch date?
Our greatest guess for the Intel Xe launch date is June 2020. That’s primarily based on a few issues with Raja Koduri, the favored public face of Intel’s renewed push into the discrete graphics card market, tweeting out an image that’s extensively believed to be some form of launch window tease, in addition to one other supply suggesting a mid-2020 launch.
Intel CEO, Bob Swan, has introduced the primary discrete Xe card, DG1, has completed its power-on testing, and early dev kits have been listed on the EEC database. So issues are undoubtedly transferring, and hopefully that additionally means they’re on observe.
Raja Koduri was as soon as AMD’s GPU engineering maestro, earlier than switching allegiance across the finish of 2017. After taking a depart of absence after the launch of Vega, he introduced his determination to depart AMD “to pursue my passion beyond hardware,” a pursuit which took him straight into the arms of Intel and its {hardware}…
Now he heads up the corporate’s Core and Visual Computing group, and is driving ahead the reason for discrete GPUs. The pun is absolutely supposed, with Raja recently tweeting out an image of his Tesla with a brand new license plate studying “THINKXE”.
@IntelGraphics pic.twitter.com/T2symDHxJ7
— Raja Koduri (@Rajaontheedge) October 4, 2019
The little nugget which has triggered individuals to quote this as a launch date teaser is the expiry date reads June 2020. It might very effectively be a coincidence, and Koduri might simply be displaying off each his inexperienced credentials and love of a customized license plate, or it might be somewhat nod to the way forward for Intel graphics.
Then a subsequent DigiTimes report cited “industry sources” which claimed a mid-2020 Intel Xe launch date, and which appears to tally fairly properly with the Raja tease too.
What are the Intel Xe specs?
Although there was speak of the Intel Xe GPU being constructed from the bottom up, Intel’s Gregory Bryant has defined that its first discrete playing cards can be constructed from the constructing blocks of its current graphics IP. And which means we’re anticipating the Xe tech to essentially be the Gen11 GPU structure of the brand new Ice Lake processors writ massive.
The method the present Intel graphics silicon is designed means it’s comprised of variety of execution models (EUs) arrayed in teams of eight with some shared assets. That’s referred to as a subslice, and there are are eight of those subslices grouped collectively to kind a full slice, which shares L3 cache and the raster backend. This 8×Eight make-up is what offers Ice Lake’s Gen11 GPU its full 64 execution unit depend.
If you take a look at the design of a slice, on the floor at the least, the block diagrams make it look quite a bit like a graphics processing cluster (GPC) from considered one of Nvidia’s current GPUs. For subslice learn SM, and for EU learn CUDA core. Each of Intel’s execution models comprises a pair of arithmetic logic models (ALU), which help each floating level and integer operations, whereas Nvidia CUDA cores comprise one fastened FP and one INT unit. In that sense Intel’s low degree GPU design is nearer to AMD’s Graphics Core Next, or Navi RDNA execution models.
They are completely different beasts – and evaluating Intel execution unit depend to CUDA cores, or RDNA cores, is a little bit of an apples vs. oranges vs. pomegranate comparability – however you possibly can see how Intel might carry a number of slices collectively to create bigger, extra highly effective graphics chips with its current design.
Back in July Intel revealed a developer model of its graphics drivers which featured a bunch of various codenames. It was rapidly taken down, however such is the best way of the interwebs, a person on the Anandtech forums posted all of them. The key entries from the attitude of Intel Xe are those beneath the DG1 and DG2 headers.
- iDG1LPDEV = “Intel(R) UHD Graphics, Gen12 LP DG1” “gfx-driver-ci-master-2624”
- iDG2HP512 = “Intel(R) UHD Graphics, Gen12 HP DG2” “gfx-driver-ci-master-2624”
- iDG2HP256 = “Intel(R) UHD Graphics, Gen12 HP DG2” “gfx-driver-ci-master-2624”
- iDG2HP128 = “Intel(R) UHD Graphics, Gen12 HP DG2” “gfx-driver-ci-master-2624”
From this it appears just like the preliminary discrete Xe GPU (DG1) can be a low-power model with the identical important make-up because the built-in chip contained in the upcoming Tiger Lake processors. That would put it at 96 EUs, as proven in an early CompuBench database listing for the chips.
Intel has promised that its Tiger Lake graphics would double the performance of its present Ice Lake GPUs, and that may point out an entry-level discrete card with 1080p/60fps gaming efficiency. Or at the least that’s what Intel Japan has been saying.
The DG2 listings are extra promising, nonetheless, although whether or not this represents a second-gen model of Xe to launch later than 2020 we’re undecided. I’m hoping they’re simply the playing cards filling out the remainder of the Xe graphics card stack.
The common interpretation of the code is that it’s the high-power (HP) model of Intel’s Gen12 graphics, and the discrete graphics (DG) model. That would imply they reference three completely different Intel Xe graphics playing cards. The remaining quantity on the tip of the codename might be seen as the full EU depend of that specific GPU, and that’s primarily based on commonplace Intel syntax for current codenames.
That would imply we’ll have a 128 EU low-end card, providing twice the silicon as that on supply within the present high Ice Lake built-in GPU {hardware}. That can be an Intel Xe GPU with a pair of full graphics slices in its make-up, after which the mid-range 256 EU half would include 4 slices to create its GPU.
At the highest of the Intel Xe stack would then be a 512 EU chip, sporting eight particular person slices of graphics goodness, and that would make it a really highly effective GPU certainly. So lengthy as Intel can nail excessive clock speeds for its graphics silicon…
Therein lies the rub. If Intel’s discrete Xe graphics playing cards are nonetheless working on the similar 1.1GHz clock speeds as their built-in CPU siblings then they’re going to wrestle to be aggressive with the graphics silicon popping out from both AMD or Nvidia’s GPU skunkworks. The hope is that as a result of the Xe GPU just isn’t having to deal with the thermal restrictions of an added bunch of CPU cores sat cheek by jowl with it, will probably be capable of run faster than the built-in chips.
Now, one of many issues with Intel’s 10nm course of, and a part of the rationale why we nonetheless haven’t seen any desktop 10nm CPUs, is that they wrestle to hit the form of frequencies that Intel’s 14nm processors can. GPUs don’t essentially should clock as excessive as a 5GHz CPU, for instance, so it wouldn’t be past the realms of chance for Xe chips to knock across the 1.7GHz or 1.8GHz marks, or probably even 2GHz.
In phrases of the cache on supply every Ice Lake slice has 3MB of L3 cache and 512KB of shared native reminiscence. Spread that out to a possible eight-slice GPU and also you’re 24MB L3 and 4MB of shared. We don’t but know what video reminiscence answer Intel is probably going to make use of for its discrete playing cards, however Koduri has stated in a redacted interview that it wouldn’t essentially be restricted to GDDR6, however that Intel might additionally match its new GPUs with the costlier excessive bandwidth reminiscence (HBM).
And everyone knows how effectively that went down along with your Vega designs, eh Raja?
But Intel issued some clarification on the now eliminated interview suggesting that Koduri had been referencing a possible high-end datacentre half for a possible HBM design. That makes extra sense and means we’re most likely going to see the usual GDDR6 setup for each the entry and fanatic class Xe graphics playing cards.
For Intel’s half it has stated that its 7nm era will kick off with a Xe datacentre GPGPU constructed utilizing the Foveros stacking method. That’s doubtless the place Raja’s HBM goals have gone.
How the person slices can be linked remains to be one thing that’s up for debate. There is the likelihood that Intel may take the chiplet strategy and create every slice as a person piece of silicon. This would scale back the complexity, and subsequently price, whereas additionally probably rising yields… one thing the 10nm tech has struggled with up to now.
But linked the slices collectively, whether or not in chiplet or monolithic kind, goes to be very important within the remaining image of general gaming efficiency.
How will Intel Xe GPUs carry out?
The present 64 EU Ice Lake GPUs are able to some first rate ranges of 720p gaming efficiency. They may even ship a modicum of 1080p enjoyable, albeit at relatively extra staccato body charges. Double, quadruple, and even octuple the {hardware} in these built-in GPUs and try to be considerably extra gaming efficiency down the road.
Where they line up in opposition to the inexperienced and purple competitors will come right down to how briskly the chips can run, how effectively Intel can optimise the {hardware} and the software program when it comes to scaling up the variety of execution models, and the way precisely the completely different slices are linked collectively on the die.
There’s some fascinating back-of-a-napkin maths over at TechSpot suggests the Xe 128 card will sit simply concerning the efficiency of GTX 1650, with the Xe 256 GPU sitting in between the RTX 2060 and 2060 Super. The top-end Xe 512 card in its guesstimate really sits simply above the Nvidia RTX 2080 Ti, which might be fairly an achievement. That’s all predicated on the GPU working at 1.7GHz and TechSpot’s TFLOPS calculations really holding water…
Intel has launched help for variable price shading (VRS) into its Gen11 graphics library, that means that games constructed with that in thoughts ought to carry out higher with the function enabled. In the artificial 3DMark VRS take a look at Intel has promised a 40% uplift in its Ice Lake chips with variable price shading enabled.
The Intel Xe GPUs will even help Adaptive Sync too, so your FreeSync-supporting screens must additionally present easy gaming once they’re plumbed right into a Xe graphics card too.
There was a suggestion lately that an Intel exec. at a Japan occasion had confirmed that Intel was baking ray tracing help into its Xe GPUs. That’s now been clarified by Intel as an error in translation… however hasn’t really detailed whether or not or not it’s true.
Whatever occurs with the ray tracing help in its new Xe graphics playing cards, at the least Intel CPUs now have access to ray tracing. You solely have to take a look at World of Tanks now to see ray traced shadows firing out of your CPU. It’s possibly only a disgrace ray tracing works higher on AMD chips…
Intel can be engaged on Xe’s multi-GPU performance for its first discrete graphics playing cards too, with the potential for them to work with the built-in GPU silicon of its CPUs to spice up gaming efficiency. So far it’s solely evidenced in Linux driver patches, so there’s no assure the multi-GPU work will bear fruit on the gaming facet because it may be restricted to easier compute workloads alone.
How a lot will Intel Xe GPUs price?
Intel said on the Architecture Day final December that it was aiming to offer Xe GPUs throughout a large spectrum of the graphics business. That means going proper from the CPU built-in chips, by means of mainstream, fanatic, and on to datacentre and AI skilled playing cards.
We had thought Intel was going to steer with its discrete GPUs for the gaming market, however the DigiTimes launch date report suggests the corporate goes to be extra aggressive than that.
“Instead of purely targeting the gaming market,” the report says, “Intel is set to combine the new GPUs with its CPUs to create a competitive platform in a bid to pursuit business opportunities from datacenter, AI and machine learning applications and such a move is expected to directly affect Nvidia, which has been pushing its AI GPU platform in the datacenter market, the sources noted.”
In an earlier interview Raja was seemingly misquoted as promising the entry-level Xe GPU would price $200, as a result of not all customers would purchase a $500 – $600 graphics card. But Intel needed to make clear it as a mistranslation, as a substitute saying that was simply an instance of entry-level discrete GPU pricing.
Until we’re quite a bit nearer to launch, and possibly even have an thought of what merchandise Intel is aiming to go up in opposition to, we nonetheless received’t have an actual clue as to how a lot Intel will cost for its first Xe graphics playing cards. But there may be anticipated to be some information at CES 2020, and probably some sort of Architecture Day Redux early in 2020 too.
Source
