DirectCompute Benchmark for Vista SP2 with DX11 or Win7


Recommended Posts

In case you run into any compatibility issues, I assume. And it's quite obvious as to why the benchmark won't run on Windows XP. It doesn't support DX10, let alone DX11.

Yes thats why i was curious about why dx9 was provided ^_^

Core i7 920

GTX 275

GPU checksum: 1370969.153194

CPU checksum: 1370969.611431

*** PASSED ***

DirectCompute Score: D14171

OpenCL not supported...(will get back to that :p)

CPU Score: M2499

Edited by DDStriker

I get the impression that this benchmark is poorly written at best, not only are cards that are meant to be slower than mine (9800 GT, 9500) achieving far higher results, we also have a faster card (GTX 260) achieving significantly lower results than my card achieved. Something about this benchmarking app doesn't seem right to me :/

I get the impression that this benchmark is poorly written at best, not only are cards that are meant to be slower than mine (9800 GT, 9500) achieving far higher results, we also have a faster card (GTX 260) achieving significantly lower results than my card achieved. Something about this benchmarking app doesn't seem right to me :/

From what I've seen that confusion is becuase some people are still posting results from the .35 version which gives much higher scores.

I can't explain the discrepancy between the scores for NVIDIA but from what I've read, ATI's video cards have less CPU usage which results in a better score. Don't take my word for it though, I could be 100% wrong.

From what I've seen that confusion is becuase some people are still posting results from the .35 version which gives much higher scores.

I've tried the 0.45b version, the thing refuses to benchmark my GPU, but has no problem with my CPU, however that score jumps all over the place.

tsupersonic has the GTX260 he is using 0.45 and his result is a lot lower than mine, but your point is well taken about the 2 folks with the 9500 and 9800

Indeed, the scoring is interesting. Just for comparative measures for the folks who are still posting with v0.35, I got D10670, C59900, M2900, as I mentioned in post #11.

I can't explain the discrepancy between the scores for NVIDIA but from what I've read, ATI's video cards have less CPU usage which results in a better score. Don't take my word for it though, I could be 100% wrong.

I got > 14000 on a single 9800 GT. :)

Can someone please update the first post to link to build 0.45d, these results are so variable it makes direct comparison almost impossible

Done. :)

I got > 14000 on a single 9800 GT. :)

You probably used v0.35. That one gave me a score greater than 150,000. Use v0.45b for more accurate results.

Indeed, the scoring is interesting. Just for comparative measures for the folks who are still posting with v0.35, I got D10670, C59900, M2900, as I mentioned in post #11.

I think even my 0.35 scores where substantially higher than yours, even though you own a faster card :/

Edit: Thanks Anaron, now can everyone who used 0.35 please re-do it, so that I can decide whether the program is dodgy or if it just doesn't like tsupersonic's system for some reason :laugh:

His program does not support the latest ATI drivers properly.

According to his ReadMe:

2) You DON'T need DX11 hardware, DirectCompute is supported on DX10 class devices if only drivers support it (forceware >= 190.62, Catalyst currently doesn't support DirectCompute on DX10/10.1 GPUs)

Now according to the latest ATI drivers (9.12):

This release of ATI Catalyst™ provides full support for DirectCompute 10.1 for the ATI

Radeon™ HD 4800 Series, ATI Radeon™ HD 4700 Series for both single card and ATI

CrossFireX™ supported configurations.

The older version (0.35) seemed to work fine as I used it in the past, but now its almost as if he intentionally blocked using the GPU to do the DirectCompute calculations on ATI cards.

That being said, I cannot get the program to give me a DirectComputer Score, I can get a CPU score, and thats the only score that updates no matter what test API I use. I do not have the ATI OpenCL drivers installed so I do not have a score for that either. But even without OpenCL I should still be able to get a DirectCompute Score, I did so on the past versions with the same exact setup.

The results are very consistent on my machine when I re-run the tests. It is definitely odd that there is such big discrepancies in some systems that have the same video cards. The most likely cause I can think of is people running different driver versions.

His program does not support the latest ATI drivers properly.

According to his ReadMe:

Now according to the latest ATI drivers (9.12):

The older version (0.35) seemed to work fine as I used it in the past, but now its almost as if he intentionally blocked using the GPU to do the DirectCompute calculations on ATI cards.

That being said, I cannot get the program to give me a DirectComputer Score, I can get a CPU score, and thats the only score that updates no matter what test API I use. I do not have the ATI OpenCL drivers installed so I do not have a score for that either. But even without OpenCL I should still be able to get a DirectCompute Score, I did so on the past versions with the same exact setup.

err.... this is something on your end bro. I have cat 9.12 and a 4870. .45 works perfectly for DirectCompute, OpenCL and cpu.

err.... this is something on your end bro. I have cat 9.12 and a 4870. .45 works perfectly for DirectCompute, OpenCL and cpu.

I doubt it, at least not exclusively.

DirectCompute Benchmark 0.35: Generates a DirectCompute Score without a problem.

DirectCompute Benchmark 0.45b: Refuses to generate a DirectCompute Score.

How exactly is that "On my end"?

Edit: Thanks Anaron, now can everyone who used 0.35 please re-do it, so that I can decide whether the program is dodgy or if it just doesn't like tsupersonic's system for some reason :laugh:

It may well be :unsure: but if a benchmark program can't keep scores properly, then it doesn't do its job right. Hopefully the future versions keep the scores consistent.

I doubt it, at least not exclusively.

DirectCompute Benchmark 0.35: Generates a DirectCompute Score without a problem.

DirectCompute Benchmark 0.45b: Refuses to generate a DirectCompute Score.

How exactly is that "On my end"?

The fact that most other people with ATI 4 series cards are able to generate scores suggests a problem with your setup, not the application.

It may well be :unsure: but if a benchmark program can't keep scores properly, then it doesn't do its job right. Hopefully the future versions keep the scores consistent.

Absolutely, suppose a number of things could be causing it to work like that, but regardless of where the fault lies, I believe we can safely say that your score probably isn't overly accurate xD

The fact that most other people with ATI 4 series cards are able to generate scores suggests a problem with your setup, not the application.

While that is true, an older version of the application works fine, while the newer version does not. This means he changed something in his application that caused it to refuse to generate a DirectCompute Score for my setup. It does not mean my setup is faulty, it means his application is not working 100%. Also, Cy-Kill has the same issue as I do, so its not something wrong with my setup and my setup only.

This topic is now closed to further replies.
  • Recently Browsing   0 members

    • No registered users viewing this page.
  • Posts

    • Google adds built-in computer control to Gemini 3.5 flash by Karthik Mudaliar Google has added Computer Use as a built-in tool in Gemini 3.5 Flash, giving developers a single model that can reason about a task and operate graphical interfaces across browsers, mobile devices, and desktop environments. The feature is available through the Gemini API and Google’s Gemini Enterprise Agent Platform, although it remains a preview feature for now. Computer Use enables an AI agent to examine screenshots and return actions such as mouse clicks, scrolling, and keyboard input. A developer’s application must execute those actions, capture the resulting screen, and send it back to Gemini, creating a continuous loop until the task is completed. Google says the integration can be used for activities including repetitive form filling, application testing, research across multiple websites, and longer enterprise workflows. Gemini 3.5 Flash can work with browser, mobile, and desktop environments, whereas Google’s earlier standalone Computer Use model was primarily positioned around browser interaction. The main change is consolidation. Computer control was previously offered through the separate Gemini 2.5 Computer Use preview model. As Neowin reported when that model was introduced, it was designed to interpret a visual interface and generate actions without requiring a website-specific API. Google later brought Computer Use to preview versions of Gemini 3 Pro and Gemini 3 Flash in January 2026. The latest release now incorporates the tool into the stable Gemini 3.5 Flash model rather than requiring developers to select a specialized model solely for interface automation. Gemini 3.5 Flash itself was announced in May as Google’s latest fast model for coding and multi-step agent workflows. It supports a one-million-token input context window and up to 65,000 output tokens, along with adjustable thinking levels that let developers trade additional reasoning for lower latency and cost. Google also added that Gemini 3.5 Flash received targeted adversarial training for computer-use scenarios. The company is also offering safeguards that can require user confirmation before sensitive or irreversible actions and automatically stop a workflow when suspected prompt injection is detected. Its developer documentation describes configurable protections for areas such as financial transactions and changes to sensitive records. Google isn't the first to bring Computer Use to its platform. Anthropic has made computer control available through Claude, while OpenAI has continued improving computer-use performance in its recent models. Microsoft has also applied the concept to business workflows, including a Computer Use capability for the Researcher agent in Microsoft 365 Copilot.
    • After I installed KB5095093, the volume on my ARM laptop won't go above 20%. It's stuck on the hearing protection level, which is pretty much useless if you want to listen to anything. I rolled back.
    • Amazon Prime Day slashes Samsung's newest Galaxy Watch Ultra by 45 percent by Karthik Mudaliar Samsung’s flagship Android smartwatch has received one of its steepest Prime Day cuts. Amazon has dropped the 2025 Samsung Galaxy Watch Ultra in Titanium Blue to $357.24, saving buyers around $292 from its $649.99 list price. That's a 45 percent discount (purchase link below). The 47mm Galaxy Watch Ultra uses a titanium casing and a 1.5-inch Super AMOLED display with a resolution of 480 x 480 and peak brightness of 3,000 nits. It includes LTE connectivity, Bluetooth 5.3, Wi-Fi, NFC, and dual-frequency L1+L5 GPS for more accurate outdoor route tracking. The 2025 model has 64GB of storage, a 590mAh battery, sapphire crystal glass, 10ATM water resistance, IP68 protection, and MIL-STD-810H durability testing. Its health and fitness tools include heart rate monitoring, sleep coaching, Energy Score, Running Coach, body composition analysis, temperature sensing, and ECG support, where available. This model is best suited to Android users who regularly run, hike, cycle, or train outdoors and want cellular access without carrying a phone. The larger battery, rugged construction, bright display, and dedicated Quick Button also make it a stronger option than Samsung’s regular Galaxy Watch models for extended workouts and demanding environments. Grab the Titanium Blue Galaxy Watch Ultra before the Prime Day price resets: Samsung Galaxy Watch Ultra (2025) [Sold and Shipped by Amazon] Good to know This Amazon deal is U.S. specific, and not available in other regions unless specified. We only use first-party seller links (at the time of article publishing); ensure that you purchase from a first-party seller link only. Check out Today's Deals on Amazon | or our recent tech deals. Become a Prime member (for Students or SNAP) via Neowin Get Prime Access - Prime for half price (for qualifying Medicaid, EBT, SNAP) Subscribe to Prime Video, Audible Plus, Music Unlimited or Kindle Unlimited via Neowin As an Amazon Associate, we earn from qualifying purchases.
    • Google begins rolling out its post-Epic Play Store billing model next week by Karthik Mudaliar Google has confirmed that its redesigned Play Store billing and fee structure will take effect on June 30, 2026, in the United States, the United Kingdom, and the European Economic Area. The changes will let eligible developers offer their own payment systems or send users to an external website for purchases, while separating Google’s platform service fee from the cost of using Google Play Billing. The rollout puts concrete dates and detailed rate cards behind the broader Android policy overhaul Google announced in March. That announcement followed a proposed settlement with Epic Games intended to resolve their long-running disputes over app distribution and payments, although the U.S. portion of the agreement still requires court approval. Under the new billing choice program, developers selling digital content or services can display an alternative payment option alongside Google Play Billing. They may also direct users to their own websites to complete a purchase. Developers can use Google’s standard payment-choice screen or design one that complies with the company’s user-interface rules. Choosing another payment processor does not eliminate Google’s cut altogether. The company will continue charging a service fee for transactions associated with apps distributed through Google Play, regardless of whether payment is handled by Google, an alternative provider, or a developer’s website. Google argues that this fee covers the value and infrastructure provided by Android and the Play Store. For developers earning up to $1 million annually, the service fee will generally be 10 percent. That rate also applies to auto-renewing subscriptions. When Google Play Billing is used in the U.S., U.K., or EEA, Google will add a separate 5 percent billing fee, and developers processing payments elsewhere will not pay that additional charge. This means Google’s familiar flat 30 percent commission is disappearing, but developers will not necessarily see a dramatic reduction on every transaction. An in-app purchase from an existing user processed through Google Play Billing can still reach a combined 30 percent. The biggest savings are likely to come from subscriptions, smaller developers covered by the $1 million tier, and companies able to move customers to their own payment infrastructure. Google is also offering lower rates through its Apps Experience and revamped Games Level Up programs. Apps and games that satisfy the company’s requirements can qualify for 15 percent service fees on new-install transactions and 20 percent on existing-install transactions. The criteria include performance and reliability standards, support for additional Android device categories, and selected platform features. Those program rates are scheduled to become available in the initial markets and Australia on September 30. For consumers, the immediate effect will depend on whether developers adopt alternative payments and pass any savings on through lower prices. For developers, however, June 30 begins a more flexible but considerably more complicated Play Store economy in which distribution, billing, install dates, revenue thresholds, and program participation can each affect Google’s final cut. Google is also separately developing a Registered App Stores program designed to simplify the installation of qualifying third-party stores. That initiative is expected to arrive with a major Android release later in 2026 and will launch outside the U.S. first. Google says the rest of the world will receive the changes by September 30, 2027, although billing rates for markets outside the US, UK, and EEA have not yet been announced.
  • Recent Achievements

    • Dedicated
      Scoobystu earned a badge
      Dedicated
    • First Post
      Tom Schmidt earned a badge
      First Post
    • One Month Later
      D0nn13 earned a badge
      One Month Later
    • Rookie
      +ChiefOfNeo went up a rank
      Rookie
    • One Year In
      Tom Schmidt earned a badge
      One Year In
  • Popular Contributors

    1. 1
      +primortal
      464
    2. 2
      +Edouard
      177
    3. 3
      PsYcHoKiLLa
      124
    4. 4
      Michael Scrip
      81
    5. 5
      Xenon
      76
  • Tell a friend

    Love Neowin? Tell a friend!